d3.2 experimentation - restable project...9 levels of aggregation. a reserve capacity forecast is...

56
This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program REstable - Improvement of renewables-based system services through better interaction of European control zones D3.2 – Experimentation Contributors:

Upload: others

Post on 22-Jul-2020

8 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

REstable - Improvement of renewables-based system services through better interaction of European control zones

D3.2 – Experimentation

Contributors:

Page 2: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

This page is intentionally left blank

Page 3: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Versioning and Authors

Version control

Version Date Comments

1.0 29.03.2019 Final Version

Authors

ARMINES: Alexis Bocquet; Simon Camal; Andrea Michiorri

ENERCON: Andreas Linder; Ana Kosareva; Johannes Brombach

ENGIE GREEN: Lina Ruiz; Peng Li

Fraunhofer IEE: Andreas Liebelt; Julia Strahlhoff; Jannic Nagel; Sven Liebehentze; Jonathan Schütt; Stefan Siegl

HESPUL: Marine Joos

HYDRO NEXT: Philippe de Montalembert

INESC TEC: Nuno Fulgêncio; Jorge Filipe; Justino Rodrigues; Ricardo Silva; Carlos Moreira; Ricardo Bessa

Page 4: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Executive Summary

Ancillary Services Forecast: A forecasting platform developed by ARMINES sends automatically production forecasts at plant level and aggregated level, for a total of 13 Wind farms, 5 PV plants and 9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts are issued with a week-ahead horizon to facilitate the scheduling of Ancillary Service tests, and with a day-ahead horizon to inform with enhanced precision on a reliable reserve capacity (see an example of forecast in Figure 1). A database server stores weather predictions from multiple sources and production data from all plants. Results are published on a web-interface with dedicated access for partners and capacity to send forecasts via e-mail.

Figure 1 Example of day-ahead forecast of production and reserve capacity, with highlights on suggested test periods

Field Tests of the European VPP: During the work package 3.2 “Experimentation” the field tests with the European VPP has been performed. The overall goal was to prove that the architecture with all the components (TSO-Emulator, VPP control system, power plants) which has been developed during work package 3.1 “Development” is able to provide ancillary services. Different types of tests have been designed (Emulator-, Pre- and Full-Tests) together with different test signals. Some things has to be prepared for each test to make sure that all systems are working like expected and to prepare all systems. As a result 7 Pre-Tests has been performed and 3 Full-Tests (see Figure 2) which is more than it was planned during work package 3.1. The technical details are described in this document. The evaluations of the tests are shown in the deliverable 3.3 “Evaluation”.

Page 5: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Figure 2 Physical reaction of the REstable VPP during a field test

Laboratory Tests: In the perspective of withdraw extended information on the technical aggregation of the distribution networks with respect to its dynamic response upon large disturbances in the transmission side, an equivalent model for MV converter-dominated distribution networks was tested in the laboratory of INESC TEC. The application considered a hybrid test rig operating in a power-hardware-in-the-loop (PHIL) configuration, enabling the interaction in real-time of a fully-detailed simulated transmission network and the laboratory network acting as a distribution grid. Additionally, a test bed with three controllable power converters were added to the network test case and connected to the virtual transmission network side – loaded into real-time digital simulator (RTDS) unit – accounting for 18, 25 and 30 MW of installed power. The power inverter available at the laboratory (20MW) was also part of the VPP portfolio and used to test a data-driven VPP control function based in reinforcement learning.

Page 6: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Content

Versioning and Authors ........................................................................................................................... 3

Executive Summary ................................................................................................................................. 4

Content .................................................................................................................................................... 6

Figures ................................................................................................................................................. 6

Tables .................................................................................................................................................. 9

1 Ancillary Services Forecasts ........................................................................................................... 10

1.1 Architecture .......................................................................................................................... 10

1.2 External data sources ........................................................................................................... 11

1.3 Forecast models .................................................................................................................... 12

1.4 Data delivery ......................................................................................................................... 16

2 Field Tests of the European VPP .................................................................................................... 19

2.1 Characterization of Field Tests .............................................................................................. 19

2.2 Preparation of Field Tests ...................................................................................................... 23

2.3 Execution of Field Tests ......................................................................................................... 32

3 Field Tests on manual Reserve ...................................................................................................... 45

3.1 Characterization of Field tests ............................................................................................... 45

3.2 Preparation of Field tests ...................................................................................................... 46

3.3 Execution of Field tests.......................................................................................................... 46

4 Laboratory Tests ............................................................................................................................ 50

4.1 ADN equivalent modelling validation .................................................................................... 52

4.2 RL-based VPP control ............................................................................................................ 54

Figures

Figure 1 Example of day-ahead forecast of production and reserve capacity, with highlights on suggested test periods ............................................................................................................................ 4

Page 7: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Figure 2 Physical reaction of the REstable VPP during a field test .......................................................... 5

Figure 3 Architecture of ARMINES’ power prediction platform ......................................................... 10

Figure 4 All wind and PV farms as well as all aggregations considered in the project ...................... 11

Figure 5 REstable Forecasts at Europe aggregation level for all sites. The top graph shows week-ahead forecasts for week between 2019-03-04 and 2019-03-08, the middle graph shows the day-ahead forecasts for day 2019-03-04 and the bottom graph shows the day-ahead forecasts for day 2019-03-07. ........................................................................................................................................... 14

Figure 6 REstable Forecasts at Europe aggregation level for the sites which are available for tests. The top graph shows the week-ahead forecasts for week between 2019-03-04 and 2019-03-08, the middle graph shows the day-ahead forecasts for day 2019-03-04 and the bottom graph shows the day-ahead forecasts for day 2019-03-07. ............................................................................................ 16

Figure 7 Dedicated sharing space for Restable project storing all Restable forecasts generated by Armines’ platform. ................................................................................................................................ 17

Figure 8 Example of email sent daily to Fraunhofer. ........................................................................... 18

Figure 9 Frequency model protocol for the pre-tests ........................................................................... 20

Figure 10 Target Active Power Response of the VPP for the pre-tests ................................................. 21

Figure 11 Frequency model protocol for the full-tests ......................................................................... 21

Figure 12 Target Active Power Response of the VPP for the full-tests ................................................. 22

Figure 13 Preconditions and constraints for a field test ....................................................................... 24

Figure 14 Deviation of calculated possible feeding-in from active power ............................................ 25

Figure 15 Difference between schedules from TSO’s and VPP’s view .................................................. 27

Figure 16 Difference between the TSO’s and VPP’s FCR set point........................................................ 27

Figure 17 Difference between FCR actual value from TSO’s and VPP’s view ....................................... 28

Figure 18 Calculation of the minimum active power ............................................................................ 29

Figure 19 Example of medium- term forecasts ..................................................................................... 30

Figure 20: Example of ARMINES Day-Ahead Forecast .......................................................................... 30

Figure 21 GUI of the field-test-tool ....................................................................................................... 31

Figure 22 Absolute Power Time Series of the first emulator test (Apr, 26th 2018) ............................... 33

Figure 23 FCR Activation Time Series of the first emulator test (Apr, 26th 2018) ................................. 33

Figure 24 Absolute Power Time Series of the emulator test with correction function mx + b (May, 4th 2018) ...................................................................................................................................................... 33

Page 8: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Figure 25 FCR Activation Time Series of the emulator test with correction function mx + b (May, 4th 2018) ...................................................................................................................................................... 33

Figure 26 Physical pre-test with ENERCON wind farms (May, 8th 2018 from 13:30 to 14:00).............. 36

Figure 27 Physical pre-test with ENGIE GREEN wind farms (Jul, 11th 2018 from 16:00 to 16:15) ........ 37

Figure 28 Physical pre-test with ENGIE GREEN wind farms No. 2 (Oct, 11th 2018 from 15:20 to 15:50) ............................................................................................................................................................... 37

Figure 29 Physical pre-test with ENGIE GREEN wind farm (March, 9th 2019 from 10:30 to 11:00) .... 38

Figure 30 Physical pre-test with HESPUL/EDISUN POWER pv plant (March, 11th 2019 from 14:22 to 14:27)..................................................................................................................................................... 38

Figure 31 Physical pre-test with ENGIE GREEN, HESPUL/VALOREM and ENERCON wind farms and HESPUL/EDISUN POWER pv plant (Mar 18th 2019 from 14:00 to 14:30)............................................. 39

Figure 32 Physical pre-test with HESPUL/BORALEX pv plant (Mar 19th 2019 from 14:30 to 15:00) ... 39

Figure 33 Physical full-test with ENERCON and ENGIE GREEN wind farms (Dec, 12th 2018 from 15:00 to 16:15) ................................................................................................................................................ 40

Figure 34 Physical full-test share of ENERCON wind farms (Dec, 12th 2018 from 15:00 to 16:15)...... 41

Figure 35 Physical full-test share of ENGIE GREEN wind farms (Dec, 12th 2018 from 15:00 to 16:15) 41

Figure 36 Physical full-test with ENERCON and ENGIE GREEN wind farms (Jan, 9th 2019 from 13:00 to 14:15)..................................................................................................................................................... 42

Figure 37 Physical full-test share of ENERCON wind farms (Jan, 9th 2019 from 13:00 to 14:15) ........ 42

Figure 38 Physical full-test share of ENGIE GREEN wind farms (Jan, 9th 2019 from 13:00 to 14:15) .. 42

Figure 39 Physical full-test with ENGIE GREEN wind farms and HESPUL/EDISUN POWER pv plant (Mar 12th 2019 from 14:00 to 15:15) ............................................................................................................ 43

Figure 40 Overview of the testing setup, including assets and corresponding interaction per case. .. 50

Figure 41 INESC TEC’s Smart-Grids and Electrical Vehicles Laboratory setup. ..................................... 51

Figure 42 Fully-detailed transmission network used in the laboratory tests. ....................................... 52

Figure 43 Schematic representation of the methodology adopted for the equivalent ADN model parametrization, implemented in at the SGEVL. ................................................................................... 53

Figure 44 Transmission network used for RL-based VPP control, and corresponding VPP portfolio under control. ........................................................................................................................................ 54

Figure 45 Schematic representation of the methodology adopted for the RL-based VPP control, implemented in at the SGEVL. ............................................................................................................... 55

Page 9: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Tables

Table 1: Power plants integrated in the VPP control system ................................................................ 22

Table 2 Maximum and Effective Energy Loss ........................................................................................ 32

Table 3 Summary of performed field tests ............................................................................................ 34

Table 4 Comparison with the planning in D3.1 ..................................................................................... 43

Page 10: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

1 Ancillary Services Forecasts

1.1 Architecture

The ARMINES’ power prediction platform has been installed on a dedicated Linux server, which distribution is Ubuntu 16.04. The platform, whose structure is shown in Figure 3, is a N-tier architecture, including:

A database server (MySQL 5.7) on which all time series data are stored for all sites

considered of the REstable project (observed power production data, Numerical Weather

Predictions as well as Wind and PV power forecasts). All logs generated by the modules of

the platform are also stored on a dedicated database.

An application server (Apache Tomcat 7) on which two web services are running. Those two

web services allow to store and retrieve all the data included in the databases (measured

data, forecasts data, execution logs, …). It also permits to get some statistics over the

contained data.

A data-importer module developed in Java allows to retrieve the NWP forecasts for specific

locations corresponding to all different Wind and PV farms of the project.

A generic Wind and PV forecasting module developed in R called by a Java program making

the link between the application side and the calculation side.

Figure 3 Architecture of ARMINES’ power prediction platform

Page 11: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

The platform is running continuously for all sites of the project and delivering daily and weekly forecasts for each of those sites as well as for each aggregation levels, which means:

13 wind farms (8 from Engie, 3 from Enercon and 2 from Hydronext)

5 PV farms (2 from Edisun, 2 from CNR and 1 from Boralex)

9 level of aggregations

Figure 4 below illustrates all levels of aggregations.

Figure 4 All wind and PV farms as well as all aggregations considered in the project

1.2 External data sources

Several sources of Numerical Weather Predictions are used to produce all the required forecasts. For an offline evaluation, we used weather forecasts from ECMWF. Then for the online forecasts, we are retrieving the following data, every morning for all geographical points linked to all wind and pv farms of the Restable project:

Hourly ARPEGE1 weather forecasts for the next 48 hours with a spatial resolution of 0.1.

These forecasts are retrieved from an external server at MeteoFrance through HTTP in GRIB2

format and are used to provide Day-Ahead power production forecasts.

Hourly GFS2 weather forecasts for the next 48 hours with a spatial resolution of 0.25. These

forecasts are retrieved from an external server at NOAA through HTTPS in DODS format

(based on NetCDF) and are used to provide a backup solution of Day-Ahead power

production forecasts.

1 ARPEGE Numerical Weather Predictions are described and downloaded from following HTTPS URL:

https://donneespubliques.meteofrance.fr/?fond=produit&id_produit=130&id_rubrique=51 2 Hourly GFS Numerical Weather Predictions are described and downloaded from following HTTPS URL:

https://nomads.ncep.noaa.gov:9090/dods/gfs_0p25_1hr

Page 12: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

3-hourly GFS3 weather forecasts for the next 10 days with a spatial resolution of 0.25. These

forecasts are retrieved from an external server at NOAA through HTTPS in DODS format

(based on NetCDF) and are used to provide a Week-Ahead power production forecasts.

For all these NWP sources, the retrieved parameters are:

U and V components of wind at 100 meters,

the temperature at 2 meters,

the down-ward radiation flux at surface,

the total cloud cover.

1.3 Forecast models

1.3.1 Model

The core of the module is based on a Quantile Regression Forest algorithm developed in R. For more information the reader is invited to refer to Deliverable D3.1 “Development”. This core module is called by a Java program which aims to recover the input data (historical measurements and NWP forecasts) from the database, format them before calling the core of the module and stores the predictions generated by the generic module in the database. It also handles logs, errors and exceptions to make the program robust.

1.3.2 Input

As input to the model, we use several kinds of information:

Static data linked to the pv and wind farms (name of each site, geographical coordinates and

installed capacity). This data is provided in an Ascii file.

Memory files are being built upstream with measured production and NWP forecasts

historical data, so that they can be used in the Day-Ahead and Week-Ahead forecasting

processes. This data is provided in an RDS file (R data file).

Last available NWP forecasts for a specific site for the following 48 hours to provide Day-

Ahead forecasts and for the following 10 days to provide Week-Ahead forecasts. This data is

provided in CSV files.

1.3.3 Output

The forecasting module outputs probabilistic power production forecasts for all sites and aggregation levels. In addition to a mean power production forecast, it outputs 10%, 50% and 90%-quantile forecasts. It also outputs low quantile forecasts (0.1% to 1% every 0.1%-quantile and 1% to 9% every

3 3-Hourly GFS Numerical Weather Predictions are described and downloaded from following HTTPS URL:

https://nomads.ncep.noaa.gov:9090/dods/gfs_0p25

Page 13: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

1%-quantile). These low quantile forecasts give the required information for ancillary services forecasts. Therefore, every day in the morning the module outputs ancillary services forecasts for the next day and every Wednesday the module outputs ancillary services forecasts for the next week.

In addition to the existing aggregation levels, different aggregation levels were created to give outputs for those sites which were available for REstable tests (which was not the case of all sites). All output forecasts are delivered in CSV format.

As shown on Figure 5, the module also produces PNG files and HTML files showing graphical results for ancillary services forecasts and reserve flexibility available for tests. In the different figures below, we can see some green periods telling when we suggest the best time for proceeding to REstable tests. The criteria for suggested tests periods is that the reserve forecast should be continuously higher than a capacity threshold fixed to a parametrized value (fixed to 5 MW in REstable project). The reserve forecast is computed as the minimal value of the 1%-quantile of aggregated production over a parametrized interval (fixed to 1 hour in REstable project) to which is subtracted the largest 1%-quantile of all plants in the aggregation. For more information, the reader is invited to refer to Deliverable D3.3 “Evaluation”.

Page 14: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Figure 5 REstable Forecasts at Europe aggregation level for all sites. The top graph shows week-ahead forecasts for week between 2019-03-04 and 2019-03-08, the middle graph shows the day-ahead forecasts for day 2019-03-04 and the bottom graph shows the day-ahead forecasts for day 2019-03-07.

Page 15: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Page 16: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Figure 6 REstable Forecasts at Europe aggregation level for the sites which are available for tests. The top graph shows the week-ahead forecasts for week between 2019-03-04 and 2019-03-08, the middle graph shows the day-ahead forecasts for day 2019-03-04 and the bottom graph shows the day-ahead forecasts for day 2019-03-07.

Figure 5 shows reserve forecasts at Europe level for all sites. The graph on the top shows week-ahead forecasts while the middle and bottom ones show day-ahead forecasts for different days of the forecasted week-ahead week. On this figure, we can see that day-ahead forecasts confirm that tests are possible for the next day.

Similarly, Figure 6 shows week-ahead and day-ahead reserve forecasts at Europe level but only for the sites which are available for tests. This figure highlights that even though the week-ahead forecasts might tell that tests would not be possible after the first day of the week, day-ahead forecasts tell that tests are possible during the 4th day of the week. Differences can occur, especially for the end of week-ahead forecasts when the forecasts were computed 7 days before.

1.4 Data delivery

All partners have access to their own forecasts through a dedicated sharing space accessible from a web based interface.

Figure 7 below is a capture of this interface showing the directories organisation.

Page 17: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Figure 7 Dedicated sharing space for Restable project storing all Restable forecasts generated by Armines’ platform.

The partners are alerted when new forecasts are available with an email containing links to the web-based interface as well as PNG and HTML graphical results. Figure 8 below is a capture of an example email sent daily to Fraunhofer.

Page 18: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Figure 8 Example of email sent daily to Fraunhofer.

Finally, all forecasts computed for Hydronext wind farms (2 wind farms) are also being sent daily to their FTP server in CSV format.

Page 19: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

2 Field Tests of the European VPP

2.1 Characterization of Field Tests

2.1.1 Different Types of Tests

There are different types of field tests which are performed during REstable project.

Emulator4 Tests

With the Wind Farm Emulators of the Fraunhofer IEE a physical field test is simulated to check all tools and interfaces as well as the whole communication chain and the generation of reports, but only within the infrastructure of Fraunhofer IEE.

Pre-Tests

The pre-tests are performed to check the interfaces and the communication to the power plants of the different partners as well as the physical reaction. Pre-tests are used to determine whether the power plants in general and over a longer period of time react with a certain accuracy to the set-points of the VPP control system. The pre-tests last 30 minutes and have the lowest meaningful curtailment (considering e.g. control constraints, regulator constraints) to minimise the loss of energy.

Full-Tests

The full-tests are performed with as many successfully pre-tested power plants as possible. An international pool is important for the full tests, as this is an important goal of the project. They last 75 minutes and have a higher curtailment.

2.1.2 Different Types of Evaluations

There are different types of active power reserve (APR) depending mainly on the requirements for activation speed and accuracy. A distinction is made between Frequency Containment Reserve (FCR), automatic Frequency Restoration Reserve (aFRR) and manual Frequency Restoration Reserve (mFRR).

The field tests are developed to test and evaluate the reaction of the power plants to a frequency rise or drop according to FCR, which has the highest requirements in terms of speed. The frequency model protocol is designed in a way to be able to also evaluate the fulfilment of aFRR requirements

4 In this context, an emulator is a software program that imitates the function of a real system and can be seen

as a black box that has the same interfaces, in- and outputs as the real system.

Page 20: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

from the same physical reaction to obtain as many evaluations as possible from a single test. The aFRR requirements are for example more stringent in terms of accuracy.

In order to evaluate the mFRR KPIs from the time series adequately, a longer time interval is actually required but the evaluation of the tests. But the evaluation of the tests may indicate a trend in how well the mFRR requirements are met.

Unfortunately, the tests and evaluations for voltage support could not be carried out in the project.

2.1.3 Test Signal

The frequency model protocol used for the pre-tests is provided by the frequency emulator and causes 6 minutes of each positive and negative constant APR activation and an about 3.5 minutes lasting realistic frequency drop. Figure 9 shows the 30 minutes lasting frequency model protocol for the pre-tests, the target response of the VPP is shown in Figure 10.

Figure 9 Frequency model protocol for the pre-tests

Page 21: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Figure 10 Target Active Power Response of the VPP for the pre-tests

The frequency model protocol for the full-tests lasts 75 minutes and is adapted to the new test protocol for the prequalification process for FCR and aFRR [PQ S.25]. Therefore the provision and activation periods have a duration of each 15 minutes (see Figure 11, Figure 12). In order to keep again the loss energy as low as possible, there will be only one positive and one negative activation. The real historical frequency signal was not used here, since a well-founded evaluation of continuously changing activation heights is comparatively difficult and was not clearly specified by the TSOs.

Figure 11 Frequency model protocol for the full-tests

Page 22: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Figure 12 Target Active Power Response of the VPP for the full-tests

2.1.4 Summary of Power Plants in the VPP Control System

The power plants which are integrated into the VPP control system are shown in Table 1. A “√” in the column “Pre-Test” means that a successful pre-test has been performed and “x” means that a pre-test was performed but it was not successful.

Table 1: Power plants integrated in the VPP control system

Partner in Charge Name Manu-facturer

Type of Power Plant

State Nominal Active Power

[MW]

Pre-Test

Full-Test

ENERCON Bischberg ENERCON Wind DE 2.4 √ √

Faas ENERCON Wind DE 52.8 √ √

Kassieck ENERCON Wind DE 54.5

Schinne ENERCON Wind DE 55.8 √ √

Sum: 165.5

ENGIE GREEN Beau Regard SENVION Wind FR 14.4

Champ Vert SENVION Wind FR 10.3

Champ Vert Sommereux SENVION Wind FR 12.3

Couturelle-Biaches SENVION Wind FR 10.3 √ √

Couturelle-Flaucourt SENVION Wind FR 8.2 √ √

Crête Tarlare SENVION Wind FR 8.2 √

Epivent SENVION Wind FR 12.4 √ √

Haut de la Vausse SENVION Wind FR 16.4

Page 23: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Haut de Bâne SENVION Wind FR 12.3

Joncels III ENERCON Wind FR 11.8 x

L’Epine SENVION Wind FR 12.3

La Haute Borne SENVION Wind FR 8.2

La Monjoie SENVION Wind FR 10.3

La Saurrupt SENVION Wind FR 10.3

La Solerie SENVION Wind FR 12.3

Le Boutonnier SENVION Wind FR 16.4

Les Près Hauts SENVION Wind FR 12.3

Mont de Ponche SENVION Wind FR 8.2

Moulin de Sehen SENVION Wind FR 12.3

Prévoterie-Perrière SENVION Wind FR 12.3

Prévoterie-Rhèges SENVION Wind FR 12.3

Prévoterie-Savinien SENVION Wind FR 12.3

Prévoterie-Vaudon SENVION Wind FR 12.3

Saint Saumont SENVION Wind FR 10.3

Sole du Moulin Vieux SENVION Wind FR 14.4

Vieux Moulin SENVION Wind FR 12.3

Sum: 305.4

HESPUL / EDISUN POWER

Sainte Maxime Solaire Solar FR 1.0 √

EPF-Chatuzange Solar FR 0.7 √ √

Sum: 1.7

HESPUL / BORALEX Cigallettes Solar FR 10.0 √

Sum: 10.0

HESPUL / VALOREM La Planche SENVION Wind FR 10.3 x

Sum: 10.3

Sum of the REstable VPP: 492.9

2.2 Preparation of Field Tests

2.2.1 Preconditions, Constraints and Processes

In order to be able to perform a successful field test, a few aspects have to be considered before (see Figure 13).

For each partner it has been agreed a maximum amount of energy loss for the project, which is caused by the curtailment of the power plants for providing positive APR capacity or for activating negative APR. Before each field test the remaining energy loss for the tests has to be considered. If the remaining loss energy for a partner it’s power plants would be excluded from the field test to avoid exceeding the agreed loss energy.

Page 24: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Apart from that, specific technical constraints of the involved power plants have to be clarified before the test. This could be a maximum curtailment or a lower control limit5 of a power plant, which leads to a minimum set point, for example.

For the planning of a specific test, some preconditions have to be considered. First the proper offset correction function has to be implemented (see chapter ‎2.2.2). Second the minimum necessary APR, considering the lower control limit, has to be calculated (see chapter ‎2.2.3). To select a time slice for the field test with sufficient expected active power, the wind forecast has to be consulted (see chapter ‎2.2.4).

Figure 13 Preconditions and constraints for a field test

Finally, a defined procedure allows the selection of a timeslot for a pre-test with involvement of the concerned partners. The following exemplary procedure is agreed with ENGIE GREEN:

1. IEE selects dates (Tue., Wed. and/or Thu.) and MW before Thu. 24:00 of previous week [e-mail]

2. IEE receives confirmation of selected dates before Fri. 16:00 of previous week [email]

3. IEE selects test time window (10:00-13:00 or 13:00-17:00) before 12:00 of each day before [e-mail]

4. IEE selects exact test time period before two hours before [phone]

-> Cancellation of one or more tests is possible in and between each step [e-mail]

2.2.2 Calculation of an Off-Set for the Set Point

The activation of positive APR is calculated by

( ) ( ) ( ( ) ( ))

5 Some wind farms have a lower control limit, i.e. they cannot be curtailed below a specific active power which

can be, for example, 10 % of nominal active power.

Page 25: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

( ): activated positive APR (activation)

( ): active power

( ): available active power

( ): ordered/contracted positive APR capacity (provision)

Following this, the calculated activated positive APR gets lower if the available active power is overestimated. P(t) and PAAP(t) are read out by the control system from the power plants.

In order to compensate the deviations of the AAP from the real active power and to avoid an underfulfillment6 during the activation of positive APR an offset correction function was implemented into the control system of the VPP. By doing this, a specific higher value is sent to the wind farms during a positive activation. This can compensate a possible systematic overestimation of the available active power. Currently for example sometimes electrical losses inside the windfarms are not considered for the calculation of the AAP which leads partly to this overestimation.

In order to calculate this offset a statistical analysis of the difference between the available active power and the active power is generated in times of no curtailment. The following exemplary analysis shows an average overestimation of the available active power of 0.9 MW during the evaluated time period of a specific wind farm.

Figure 14 Deviation of calculated possible feeding-in from active power

6 Underfulfillment means, that the activated active power reserve is lower than the called active power reserve.

Page 26: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

This evaluation has to be done for each wind farm or cluster of wind farms as every wind farm has a different precision for the calculation of the available active power.

In addition to the described absolute offset, it can be useful to implement a relative offset for safety reasons as well. This may compensate fluctuation and avoid underfulfillment in the limit of tolerated overfulfillment by sending for example a 5 % higher set point for each positive and negative APR.

Combining the absolute and relative offset, a correction function und is implemented in the control system of the VPP, whereas

⁄ ⁄

⁄ ⁄ .

By using this correction function, the results of the evaluation are expected to be better, which means that the actual APR value is closer to the set point.

At the moment, the compensation of positive provision and activation is not part of the real APR mechanisms and evaluations, which is why it could not be applied in such a way in real operation. The following example will explain this more detailed.

For a FCR test the VPP shall provide +/- 10 MW. The control system sends a specific higher set point (e.g. 10.2 MW) disaggregated to the power plants, which try to reduce their active power in summary by 10.2 MW. The schedule for the power plants is calculated by the difference between the AAP and the ordered positive active power.

( ) ( ) ( )

Following this, the calculated real time schedule for the pool is for example

- by the VPP incl. offset correction: ( )

- by the TSO: ( )

The difference between the VPP’s and the TSO’s calculation of the schedule is visualized in Figure 15.

Page 27: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Figure 15 Difference between schedules from TSO’s and VPP’s view

As already mentioned before, the APR set-point sent by the VPP for activating positive (negative) FCR is higher (lower) than the actual ordered APR (see Figure 16). The power plants try to follow the higher (lower) set-point and thus the APR actual value is closer to the TSO’s set point and avoids underfulfillment.

Figure 16 Difference between the TSO’s and VPP’s FCR set point

However, the APR actual value is a calculated time series as well which is calculated by the difference between the schedule and the active power.

Page 28: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

( ) ( ) ( )

This leads to the fact, that the TSO would possibly see a different (e.g. lower) APR actual value, because he has already calculated a different schedule than the VPP. This means, the TSO would see a time series which is shifted down (see Figure 17).

Figure 17 Difference between FCR actual value from TSO’s and VPP’s view

For the field tests in the project and the evaluations this just described issue can be disregarded, since this offset correction procedure has not been agreed with the TSOs yet and is just a technical feature of the VPP control system that leads to better control results to show what quality would be reachable if the AAP would be more accurate. The necessity of this feature will be discussed further in WP 3.3 in consideration of new APR requirements.

2.2.3 Calculation of Minimum Active Power

For a successful execution of a field-test the wind farms have to provide a specific minimum of active power so that the curtailment does not cause an operating point below the lower technical regulator limit.

Figure 18 illustrates the calculation of the minimum necessary active power for a field test. First the technical boundaries of e.g. 10 % of the nominal power have to be considered. Second the offset correction factor for the available active power has to be added. At least the maximum possible activation of APR has to be adapted for each positive and negative APR, including the 5 % safety. In the illustrated example with a technical boundary of 1.7 MW and an offset of 0.9 MW the minimum

Page 29: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

active power for an activation of +/- 3 MW would be 8.9 MW and for an activation of +/- 1 MW would be 4.7 MW. In that case a regulation of Δ 3 MW is adequate for a good regulator / control response.

Figure 18 Calculation of the minimum active power

2.2.4 Wind Forecast for Field Test Planning

Based on the calculation of the minimum active power, a timeslot for the field test has to be chosen when the wind speed is high enough to guarantee the minimum active power. For this a daily forecast of the wind speed at the concerned wind farms is used. Regarding the Capacity Factor (CF [%]), the expected active power can be calculated as the percentage of the nominal power.

LOCAL TIME V80

(m/s) CF (%)

P(V>5) (%)

P(G>10) (%)

P(G>15) (%)

T (°C)

P(RAIN) (%)

P(FZRA) (%)

P(SNOW) (%)

P(TS) (%)

ACC_PCPN FROST_I ADVICE

Wed, 10 Oct 2018

Page 30: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

06:00 8 55 60 80 40 8 - - - - - - 0

07:00 8 58 40 80 20 7 - - - - - - 0

08:00 8 46 30 60 10 7 - - - - - - 0

09:00 8 51 35 65 10 9 - - - - - - 0

10:00 8 43 55 80 20 12 - - - - - - 0

11:00 6 20 60 85 35 16 - - - - - - 0

12:00 5 12 55 85 25 18 - - - - - - 0

13:00 5 9 45 80 15 21 - - - - - - 0

14:00 5 10 20 60 10 22 - - - - - - 0

Figure 19 Example of medium- term forecasts

Figure 20: Example of ARMINES Day-Ahead Forecast

For the arrangement of a pre-test with more involved partners or of a more complex full test, FRAUNHOFER has developed a so called “Field Test Doodle” (see Figure 21).

With the help of the field test doodle all partners can vote for selected weeks, days and timeslots to find a date with highest availability of the partners and best wind situation.

Page 31: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Figure 21 GUI of the field-test-tool

2.2.5 Summary of Energy Loss

The involved partners and third parties have each confirmed a specific amount of maximum energy loss which may be caused during all tests of REstable project. The maximum and effectively caused energy loss during the field tests is shown in Table 2.

The energy loss is calculated by

∫ ( ) ( )

Or with consideration of an overestimation of the available active power (AAP) which corresponds to the “b” in the offset correction function

∫ ( ) ( )

Page 32: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Table 2 Maximum and Effective Energy Loss

Partners / Third Parties Maximum Energy Loss Effective Energy Loss

ENERCON 40 MWh 1st Pre: 2 MWh 1st Full: 7.724 MWh 2nd Full: 12.187 MWh = 21.911 MWh

ENGIE GREEN 100 MWh 1st Pre: 0.003 MWh 2nd Pre: 0.6 MWh 3rd Pre: 1.322 MWh 1st Full: 0.7393 MWh 2nd Full: 3.228 MWh 3rd Full: 10.2973 MWh

5th Pre: 3.949 MWh = 20.1383 MWh

HESPUL / BORALEX 8 MWh 7th Pre: 0.3863 MWh

= 0.3863 MWh

HESPUL / EDISUN POWER 1,5 MWh 4th Pre: 0.005 MWh 3rd Full: 0.129 MWh 5th Pre: 0,0321 MWh = 0.1661 MWh

HESPUL / VALOREM 10 MWh none

2.3 Execution of Field Tests

2.3.1 Emulator Tests

Before the first physical field test some emulator tests were executed in order to check the tools and interfaces as well as the whole communication chain and the generation of automated reports.

For the emulator tests the wind farm emulators were connected to the VPP to simulate real wind farms and the reaction to set points of the control system. For activating the APR, the same frequency model protocol was used as for the physical field tests.

Figure 22 shows the reaction of the wind farm emulators to the APR activation. The result of the emulator test with implementation of a correction function mx + b is shown in Figure 24. It is

Page 33: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

demonstrated that the deviation from the set point during the activation of positive APR is smaller than before (see Figure 23 and Figure 25).

Figure 22 Absolute Power Time Series of the first emulator test (Apr, 26th

2018)

Figure 23 FCR Activation Time Series of the first emulator test (Apr, 26th

2018)

Figure 24 Absolute Power Time Series of the emulator test with correction function mx + b (May, 4th

2018)

Figure 25 FCR Activation Time Series of the emulator test with correction function mx + b (May, 4th

2018)

Page 34: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

2.3.2 Physical Tests

The degree of success of a physical field test depends on different aspects which have to be considered and which get more complex the more different wind farms participate. Due to that, many planned field tests had to be canceled for different reasons. The following table shows a summary of planned and executed field tests.

Table 3 Summary of performed field tests

Partner Wind farms Test Type

Date Status Comments

ENERCON Faas, Bischberg Pre Test

08.05.2018 13:30-14:00

Executed No reaction of Bischberg, delay in reaction time

ENGIE GREEN COB, COF Pre Test

10.07.2018 16:00-16:15

Executed technical lower control limits reached

ENGIE GREEN COB, COF Pre Test

10.10.2018

15:20-15:50

Executed Partly remaining at same absolute active power level

ENERCON, ENGIE GREEN

Faas, Bischberg, Schinne, COB, COF

Full Test

20.11.2018 Cancelled TSO Management

ENERCON, ENGIE GREEN

Faas, Bischberg, Schinne, COB, COF

Full Test

28.11.2018 Cancelled Wind situation in Germany too bad

ENERCON, ENGIE GREEN

Faas, Bischberg, Schinne, COB, COF

Full Test

29.11.2018 Cancelled

ENERCON, ENGIE GREEN

Faas, Bischberg, Schinne, COB, COF

Full Test

06.12.2018

15:00-16:15

Executed Overestimation of AAP, French wind farms practically did not fulfil the negative activation

ENERCON, ENGIE GREEN

Faas, Bischberg, COB, COF

Full Test

09.01.2019

13:00-14:15

Executed Better reaction time, relatively high fluctuation during

Page 35: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

positive provision

ENGIE GREEN, HESPUL / EDISUN POWER

EPV, CTE, EPF Pre Test

07.03.2019

10:30-11:00

Executed Only EPV reacted

HESPUL / EDISUN POWER

EPF Chatuzange Pre Test

11.03.2019

14:22-14:27

Executed EPF reacted this time

ENGIE GREEN, HESPUL / EDISUN POWER

COB, COF, EPV, EPF Chatuzange

Full Test

12.03.2019

14:00-15:15

Executed Very good overall quality

ENERCON, ENGIE GREEN, HESPUL / EDISUN POWER, HESPUL / VALOREM

SMS, CTE, Joncels III, La Planche

Pre Test

18.03.2019

14:00-14:30

Executed Strange behaviour of some PP

HESPUL / BORALEX

Cigallettes Pre Test

19.03.2019

14:30-15:00

Executed Very good quality at the end

ENERCON, ENGIE GREEN, HESPUL / BORALEX, HESPUL / EDISUN POWER, HESPUL / VALOREM

Faas, Bischberg, Schinne, COB, COF, EPV, CTE, EPF Chatuzange, SMS, Cigallettes

Full-Test

19.03.2019 – 21.03.2019

Cancelled An extended high pressure area over Europe caused a calm wind and prevented the last full test of the project.

Page 36: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

2.3.2.1 Pre-Test ENERCON (1st)

In the pre-test with ENERCON two wind farms participated with a total nominal power of 55.2 MW – wind farm Faas with 52.8 MW and Bischberg with 2.4 MW. The result of the reaction to the simulated frequency change is shown in Figure 26.

It became obvious that there was no reaction of one wind farm (Bischberg). Later research figured out that the wind farm has not got the needed technical permission to be curtailed. Bischberg had a share of only 4.2 % of nominal power, so this did not affect the result very seriously.

Another effect was that latencies in the algorithms and the set-point communication between the control system of the VPP and the wind farms occurred which result in physical reaction delays of up to 30 seconds. The source of the delays could not be identified in this project. For setpoint-based communication products the current reaction seems sufficient, nevertheless could be improved for operational environments and for faster products. For FCR in real operation the measurement and realization would be performed locally anyway and this delay would not occur.

After the test the curtailment lasted on and had to be reset manually. This was a problem in the control system which could only occur in certain rare situations. The problem was fixed immediately and did not occur in the following tests.

The energy loss for this pre-test was 2 MWh.

Figure 26 Physical pre-test with ENERCON wind farms (May, 8th

2018 from 13:30 to 14:00)

2.3.2.2 Pre-Test ENGIE GREEN (2nd)

In the ENGIE GREEN pre-test the wind farms Couturelle-Biaches (COB 10.25 MW) and Couturelle-Flaucourt (COF 8.2 MW) participated with a total nominal power of 18.5 MW.

Because of a bad wind situation during the test, the curtailment caused that the technical lower control limits of each wind farm were reached. For that reason, the real time schedule could not get a forced value below these boundaries (1.24 MW for COB and 0.82 MW for COF). The test had to be cancelled after half of the time.

Page 37: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

The energy loss for this cancelled pre-test was 0.003 MWh.

Figure 27 Physical pre-test with ENGIE GREEN wind farms (Jul, 11th

2018 from 16:00 to 16:15)

2.3.2.3 Pre-Test ENGIE GREEN (3rd)

The third pre-test with ENGIE GREEN was also performed with wind farms COB and COF.

One could see that the APR mode was activated too late. In the VPP it was activated manually 45 seconds before the start of the test protocol, but physically it was activated 30 seconds after the start of the test protocol.

It is assumed that a set point of +/- 1 MW is quite low for high fluctuations of the wind farms which leads to high fluctuations around the set point. It is proposed to activate at least +/- 3 MW to get a good regulator response.

Moreover, it is striking, that from minute 12 to 15 and from minute 18 to 21 the active power is quite constant during a rising AAP which is an untypical behaviour.

The energy loss for this pre-test was 0.6 MWh.

Figure 28 Physical pre-test with ENGIE GREEN wind farms No. 2 (Oct, 11th 2018 from 15:20 to 15:50)

Page 38: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

2.3.2.4 Pre-Test ENGIE GREEN and HESPUL/EDISUN POWER (4th)

The fourth pre-test was performed with wind farms EPV and CTE from ENGIE GREEN and the pv plant EPF from HESPUL/EDISUN POWER. For this test EPF and CTE didn’t react properly. Therefore, Figure 29 only shows EPV. The average AAP during the test was 10.259 MW with a provision of APR capacity of +/- 2.5 MW.

The energy loss for this pre-test was 1.322 MWh.

Figure 29 Physical pre-test with ENGIE GREEN wind farm (March, 9th 2019 from 10:30 to 11:00)

2.3.2.5 Pre-Test HESPUL/EDISUN POWER (5th)

The fifth pre-test was performed with EPF from HESPUL/EDISUN POWER. Figure 30 shows the absolute power time series. The average AAP during the test was 0.186 MW with a provision of APR capacity of +/- 75 kW.

The energy loss for this pre-test was 0.005 MWh.

In addition it should be mentioned that the set-points after the pre-test could not be reseted correctly, therefore EPF was curtailed till the next day where it was noticed and taken back.

Figure 30 Physical pre-test with HESPUL/EDISUN POWER pv plant (March, 11th 2019 from 14:22 to 14:27)

Page 39: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

2.3.2.6 Pre-Test ENGIE GREEN, HESPUL/EDISUN POWER, HESPUL/VALOREM and ENERCON (6th)

In the sixth pre-test three wind farms (CTE, Joncels III, La PLanche) and one PV plant (SMS) participated with a total nominal power of 31.4 MW. The average AAP during the test was 11.961 MW with a provision of APR capacity of +/- 1.5 MW.

The total loss energy for the pre-test was 4.051 MWh.

Figure 31 Physical pre-test with ENGIE GREEN, HESPUL/VALOREM and ENERCON wind farms and HESPUL/EDISUN POWER pv plant (Mar 18th 2019 from 14:00 to 14:30)

2.3.2.7 Pre-Test HESPUL/BORALEX (7th)

In the seventh pre-test the PV plant Cigallettes participated with a nominal power of 8.16 MW. The average AAP during the test was 6.764 MW. The provision of APR capacity was +/- 1 MW.

The total loss energy for the pre-test was 0.4733 MWh.

Figure 32 Physical pre-test with HESPUL/BORALEX pv plant (Mar 19th 2019 from 14:30 to 15:00)

2.3.2.8 Full-Test ENERCON and ENGIE GREEN (1st)

In the first full test three ENERCON wind farms (Faas, Schinne, Bischberg) and two ENGIE GREEN wind farms (COB, COF) participated with a total nominal power of 129 MW. The average AAP during the

Page 40: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

test was 23 MW altogether. The provision of APR capacity was +/- 6 MW and the activation of APR as well +/- 6 MW. In summary the first international field test in REstable was successful, but some aspects occurred:

In the beginning, the French wind farms needed about 3 minutes to get into the curtailment.

Major overestimates of the available active power occurred for the German wind farms. The compensation algorithm did not have enough buffers, which is why frequent underfulfillments occurred during the positive activation.

When changing the set-point value, the German wind farms sometimes experienced delays of 30 to 50 seconds (in comparison to the input time of the TSO set-point signal in the control system) until it was physically realized. The delay observed depends on the parametrization of the wind farms. As the wind farms are currently optimized for maximum yield and grid code compliance the reaction may be optimized by changing the default parametrization. The French wind farms had delays of 70 to 100 seconds in some cases. For a more detailed analysis please refer to D3.3.

COF in particular showed extremely sluggish behaviour and in some cases remained at the same absolute level for many minutes, although it should have changed.

Both French wind farms practically did not fulfil the negative activation at all, which is why there was a constant underfulfillment for the negative activation.

The loss energy for the full- test was 7.75 MWh on ENERCON’s and 0.75 MWh on ENGIE GREEN’s side.

Figure 33 Physical full-test with ENERCON and ENGIE GREEN wind farms (Dec, 12th 2018 from 15:00 to 16:15)

In order to examine the behaviour of the wind farms of a specific partner, Figure 34 and Figure 35 show the results of either only ENERCON wind farms or only ENGIE GREEN wind farms.

Page 41: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Figure 34 Physical full-test share of ENERCON wind farms (Dec, 12th 2018 from 15:00 to 16:15)

Figure 35 Physical full-test share of ENGIE GREEN wind farms (Dec, 12th 2018 from 15:00 to 16:15)

2.3.2.9 Full-Test ENERCON and ENGIE GREEN (2nd)

In the second full test two ENERCON wind farms (Faas and Bischberg) and two ENGIE GREEN wind farms (COB and COF) participated with a total nominal power of 73.7 MW. The average AAP during the test was 27.2 MW altogether. The provision of APR capacity was +/- 10 MW and the activation of APR as well +/- 10 MW.

In summary the second international field test in REstable was successful. The overall quality was similar compared to a real conventional pool. There was very quick physical responses (under 30 seconds) and relatively exact adherence to the set-points. During the full positive activation major deviations between the active power and the AAP were observed. It is assumed that by improvements of the AAP-signal the deviations may be lowered.

The loss energy for the full- test was 12.2 MWh on ENERCON’s and 3.2 MWh on ENGIE GREEN’s side.

Page 42: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Figure 36 Physical full-test with ENERCON and ENGIE GREEN wind farms (Jan, 9th 2019 from 13:00 to 14:15)

In order to examine the behaviour of the wind farms of a specific partner, Figure 37 and Figure 38 show the results of either only ENERCON wind farms or only ENGIE GREEN wind farms.

Figure 37 Physical full-test share of ENERCON wind farms (Jan, 9th 2019 from 13:00 to 14:15)

Figure 38 Physical full-test share of ENGIE GREEN wind farms (Jan, 9th 2019 from 13:00 to 14:15)

2.3.2.10 Full Test ENGIE GREEN and HESPUL/EDISUN POWER (3rd)

In the third full test three ENGIE GREEN wind farms (COB, COF and EPV) and one PV plant from HESPUL/EDISUN POWER (EPF Chatuzange) participated with a total nominal power of 31.55 MW. The average AAP during the test was 31.177 MW altogether. The provision of APR capacity was +/- 6 MW and the activation of APR as well +/- 6 MW.

Page 43: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

In summary the third international field test in REstable was a great success. The overall quality was better than in the other Full Tests. There was very quick physical responses (under 30 seconds) and low fluctuation in the resulting active power time series.

The loss energy for the full-test was 10.297 MWh on ENGIE GREEN’s and 0.129 MWh on HESPUL/EDISUN POWER’s side.

Figure 39 Physical full-test with ENGIE GREEN wind farms and HESPUL/EDISUN POWER pv plant (Mar 12th 2019 from 14:00 to 15:15)

2.3.3 Comparison with the Field Test Planning in D3.1

Overall 5 Up-Tests and 5 Down-Tests with a total test-duration of 150 minutes were planned in work package 3.1 “Development”. It was possible to perform more than this planned number of tests. In total 7 Pre-Tests with at best each 2 Up-Tests and 2 Down-Tests and 3 Full-Tests with each 2 Up-Tests and 2 Down-Tests was performed. As a result overall 12 Up-Tests, 12 Down-Tests and 380 minutes could be realized during the tests.

Table 4 Comparison with the planning in D3.1

Tests planned in D3.1 Performed

Number of Up-Tests 5 6 during Full-Tests

6 during Pre-Tests

Number of Down-Tests 5 6 during Full-Tests

6 during Pre Tests

Test-Duration 150 minutes 225 minutes Full-Tests

155 minutes Pre-Tests

Page 44: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Page 45: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

3 Field Tests on manual Reserve

3.1 Characterization of Field tests

The goal of this Field Test is to provide manual reserve to the Grid in real commercial conditions. Therefore a dedicated ephemeral VPP was set up to fully fulfill REMA rules (“Les règles relatives à la Programmation, au Mécanisme d'Ajustement et au dispositif de Responsable d'équilibre”), and then was offered on RTE’s “Mécanisme d’ajustement”

The power plants used for this field test differ from the one of the other field tests, as the prior approval from its current Balance Responsible could not be obtained by Restable. An alternative VPP could be constituted with 2 wind sites operated by ENGIE, and which balancing responsible CNR granted such prior approval.

As of February 1st 2019, the EDA RESTAT 1 was created and filled within RTE systems. It is constituted of:

- Wind park Barly : 5 x Vestas V100/2000 - 10 MW

- Wind park Scaer le Merdy : 4 x Senvion MM92/2050 - 8,2 MW

Page 46: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

3.2 Preparation of Field tests

3.2.1 Implicit vs explicit offers

The REMA rules allow 2 type of flexibility offers: explicit vs implicit which have their own pro and cons

- Implicit offers consist in providing to the grid a production schedule and technical constraints

associated with it. The Grid is then allowed to request any variation of said schedule as long

as staying within the technical constraints. The assessment of an activation is performed by

comparing the actual production vs the production schedule amended with the grid orders.

- Explicit offer consist in providing a flexibility schedule and technical constraints associated

with it. No production schedule is supplied. The assessment of an activation is performed

comparing the actual production vs the production level observed before the activation

amended with the grid order.

Pros Cons

Implicit The full flexibility of the plant can be offered. It corresponds to the production forecast

The reliability highly depends on the quality of the forecast, and the possibility to update it regularly.

Explicit No forecast is requested

The assessment of an activation is not based on a production forecast

The production level before the activation is a relevant reference point only for short activations

The offered flexibility is limited to the lowest production level of a given period

Based on the above, the test was conducted on implicit offer scheme.

3.3 Execution of Field tests

RESTAT 1 was offered on RTE adjustment mechanism platform during the week n° 9, eg from 25/02/2019 to 03/03/2019 on a 24h/7 basis.

Page 47: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

3.3.1 Schedule and technical constraints

The schedule of production was communicated to RTE on D-1 basis. It is based on the forecast delivered to Hydronext by Armines on D-1 at 09:30. This forecast cover the current day and D+1, allowing therefore a revision of the schedule in intraday.

It has been agreed a maximum amount of energy loss of 10 MWh for this manual reserve test. Therefore the bidding had to be performed accordingly. The technical constraints associated with such schedule were:

Technical constraint Value for manual Reserve Field Test

Max allowed load curve The production forecast

Min allowed load curve null

Max allowed Energy The energy of the production forecast

Min allowed Energy The energy of the production forecast minus 10 MWh

Preparation time 30 minutes

Other Not relevant

3.3.2 Commercial bids

The commercial bids consist of a series of prices in €/MWh for up and down flexibility. As per REMA rules, the 24hours are divided in 6 price intervals : 0h to 6h ; 6h to 11h ; 11 to 14h; 14h to 17h; 27h to 20h ; 20h to 24h

Only the downward bids should be considered (upwards bids where always maintained at a high level of 500€/MWh). For sake of clarity, the downward bid is the price that the operator (Hydronext) is ready to pay to the Grid in order to receive it from the Grid without actually having to produce it. It therefore corresponds to the saving generated by the non-production of this power, and would theoretically correspond to the production marginal costs.

In case of renewable, such marginal costs is 0 €/MWh. This is why renewable plants rank after thermal plants in the merit order. The bidding strategy for the test resulted in a combination of offering prices that are:

Sufficiently attractive (ie high) in order to maximize the chances of an activation by RTE

In line with real business practice.

Page 48: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

The bidding strategy on 28/02/2019 went as as follow:

Initial bid on 27/02 at 10:00 (D-1): no price level indication - the bids are kept conservatively low/unattractive

Period of 28/02 Price Downward (€/MWh)

0h à 6h -100

6h à 11h -100

11h à 14h -100

14h à 17h -100

17h à 20h -100

20h à 24h -100

Revised bid on 27/02 at 15:00: Result of spot auction published at 13:00 on D-1. Bid prices are adjusted to Spot levels. The periods 14h to 17h and 17h to 20h are set voluntarily attractive (high wind).

Period of 28/02 Price Downward (€/MWh) Avg Spot (for information) (€/MWh)

0h à 6h -20 32.7

6h à 11h -8 46.8

11h à 14h -13 38

14h à 17h 34 35.3

17h à 20h 38 45.3

20h à 24h -9 44.1

Page 49: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Revised bid on 28/02 at 15:55 : still no activation is observed, therefore Hydronext decides to improve the bid on 17h to 20h period to make it more attractive

Period of 28/02 Price Downward (€/MWh) Avg Spot (for information) (€/MWh)

0h à 6h -20 32.7

6h à 11h -8 46.8

11h à 14h -13 38

14h à 17h 34 35.3

17h à 20h 43 45.3

20h à 24h -9 44.1

3.3.3 Execution and chain of command

Activation of the offers were communicated by RTE to Hydronext command center through the traditional media, eg via RTE’s TAO machine to machine interface.

The order was then communicated from Hydronext to Engie command center via telephone, from operator to operator. Given that the expected response time to execute such activation was set to 30 minutes, no machine to machine command chain was deemed necessary between Hydronext and Engie.

The curtailment of the 2 plants was then executed from Engie command center.

Page 50: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

4 Laboratory Tests

The laboratorial experiments developed by INESC TEC described in this report were generally divided under two main sections, both serving as validation procedures for the developments achieved in the previous tasks, regarding:

1. Active distribution networks (ADN) equivalent modelling;

2. Reinforcement-Learning (RL) based virtual power plant (VPP) control.

A hybrid laboratorial configuration was used to implement both cases (generically depicted in Figure 40) that, although developed independently, took advantage of the interaction between the real-time simulation platform – provided by the real-time digital simulator (RTDS) unit – and the physical assets available at the Smart-Grids and Electrical Vehicles Laboratory (SGEVL) running altogether in a power-hardware-in-the-loop (PHIL) configuration. Either cases considered the laboratory network to act as a distribution network, operating at medium voltage (MV), established at the 30kV level, and accounting with 20MW of installed power. This network interacts dynamically, in real-time, with a fully-detailed generic transmission network loaded into the RTDS unit. In either cases, events and devices in the laboratory and at the simulated transmission network were controlled using the Lab Device Manager, also developed at INESC TEC, which has the ability to fully control the operation of the network of the SGEVL.

Figure 40 Overview of the testing setup, including assets and corresponding interaction per case.

SGEVL Electric Panel

15kW Power Amplifier(Triphase PM15)(LV 230/400V)

3ph + N

RTDS unit(OPAL RT 5600)

Transmission network offline

loading

Physical assets Virtual environment

Lab Device

Manager

TCP/IPRL-based VPP

Control Unit

TCP/IP

Real-time communication

link (RT)

Power/VoltageScaling

15kW : 20MW

230V : 30kV

RL-based VPP Control

ADN Dynamic Eq. Modelling

TCP/IP

Page 51: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

The laboratory test rig consists in a scalable and configurable low voltage (230/400V) network comprising a bidirectional power inverter suitable to emulate solar photovoltaic (PV) or storage units (10kW), and able to comply with fault ride through (FRT) capabilities, a resistive load (20kW) and an induction motor (3kW). A power amplifier (15kW) (Triphase PM15) is used to perform the power interface between the simulated transmission network and SGEVL network. It also provides the proper power scaling to the laboratory network, so it can act (virtually) as a real MV ADN, while interacting with the transmission network, at the RTDS unit. A comprehensive scheme of the laboratory electrical configuration is depicted in Figure 41.

Figure 41 INESC TEC’s Smart-Grids and Electrical Vehicles Laboratory setup.

For either cases, a 22-buses generic transmission network (see Figure 42) operating at extra-high voltage (230kV and 500kV) was used and fully-detailed in the simulation platform of MATLAB/Simulink®. It comprises six conventional generation units (synchronous thermal) connected at either 13,8kV or 21,6kV, summing a total of around 4,5GVA of installed power, as well as around 3GW/1,5Gvar of total load power. All the components (synchronous machines and their controllers, loads, lines and power transformers) were modelled recurring to well-established state-of-the-art models from the MATLAB/Simulink® libraries, using standard parametrizations.

15kW Power Amplifier(Triphase PM15)

3ph

3ph + N

R

LV (230/400V)

Network

N

R3ph

PMSG

ResistiveLoad

ResistiveLoad

InductionMotor

⁄Power

Inverter(w/ FRT)

+_ 4 Quadrant

DC Power Source Interface withvirtual environment

Page 52: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Figure 42 Fully-detailed transmission network used in the laboratory tests.

The two cases under study and reported in this document are explained in detail in the following subsections.

4.1 ADN equivalent modelling validation

The results obtained in D2.2 have provided the required level of confidence to extend validation procedures at the laboratorial level, to assess whether the proposed ADN dynamic equivalent model has the ability to properly represent a near-real-world application and also to improve the solution’s robustness.

The laboratory’s closed network is aimed to act as a representation of a given ADN, fully dominated by power converter-interfaced generation. The idea is to compare its aggregated response to large voltage disturbances occurring at the transmission network level, to the equivalent model’s simulated response, in similar conditions. In this case, the focus was solely on the evaluation of the ADN’s aggregated response of reactive power injection upstream, during faulty operation, as part of the FRT service. A comprehensive schematic of the methodology adopted is presented in Figure 43.

Page 53: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

Figure 43 Schematic representation of the methodology adopted for the equivalent ADN model parametrization, implemented in at the SGEVL.

As previously depicted, the equivalent model aims to represent the aggregated behavior of ADNs for large voltage disturbance at the transmission side, meaning several three-phase short-circuits with variable magnitudes and locations were exploited, to cover a wide diversity of disturbances types. To ensure similarity between the laboratory responses and the equivalent model’s responses, the voltage “seen” by the laboratory network during the faults ( ) is applied to a three-phase

controllable voltage source, in the simulation side. The scaled reactive power responses from the SGEVL network ( ( )) are then compared with the equivalent model’s reactive power responses ( ( )). The equivalent model is subsequently reparametrized iteratively in an offline procedure

recurring to an evolutionary particle swarm optimization (EPSO) algorithm, to fit the laboratory’s behavior. In the studies developed in this task, the “grey-box” dynamic model structure was adapted to the existent assets used in the laboratory. It is, as depicted in Figure 43, composed by two main components, each connected in parallel through an equivalent impedance: an equivalent power converter (FRT capability compliant), an equivalent composite load (comprising static and dynamic loads types).

SGEVL Electric Panel

Power Amplifier(Triphase PM15)

Real-time communication

link3ph + N

RTDS unit(OPAL RT 5600)

Transmission

Network

Simulation ModelTransmission

network offline loading

Physical assets Virtual environment

( ) ( )≈

Controlled

Voltage Source

( )

MEq. Load

Eq. Power

Converter with FRT

Equivalent Model for Active Distribution Network

Model fitting using EPSO

Page 54: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

4.2 RL-based VPP control

The trained Reinforcement Learning (RL) agent, previously tested on a REST API that simulated the operation of the VPP, was now used for calculating and sending power setpoints for the setup prepared at the INESC TEC laboratory in order to simulate real-time, near real-world operation scenarios. A series of tests were therefore designed where the RL agent could be put to test in order to ascertain the feasibility of its usage for real-world applications.

In order assess the application, three controllable power converters were added to the previously presented network test case, and connected to the virtual transmission network side (loaded into the RTDS unit), accounting for 18, 25 and 30 MW of installed power (Figure 44). The power inverter available at the laboratory (20MW) was also part of the VPP portfolio, being also considered when applying the method. It is important to notice that given the agnostic nature of the training method, the algorithm could be used for optimizing the output of the DER with any desired installed capacity.

Figure 44 Transmission network used for RL-based VPP control, and corresponding VPP portfolio under control.

A comprehensive representation of this section’s methodology and testing setup is represented in Figure 45.

As inputs, the agent requires not only the 4 seconds AGC signals, but also the day ahead (DA) and frequency restoration reserve (FRR) bids and the flexibility (or state) of each DER comprising the VPP. For simplicity of the tests, and because no further conclusions could be extracted from simulating random DER states and different DA bids for this purpose, fixed values for each of these inputs were

C VPP controllable power converters

C1 C3 – Triphase

C2 C4

20MW

18MW

30MW

25MW

Page 55: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

defined a priori, with the VPP maximum output threshold being always greater than the DA + FRR bids. The FRR bid was then defined as ±10% of an 80MW DA bid, and the DER maximum output equaled to each DER’s installed capacity.

Figure 45 Schematic representation of the methodology adopted for the RL-based VPP control, implemented in at the SGEVL.

The overall management of the experiment is performed via TCP-IP communication by the Lab Device Manager, which is responsible to communicate the power set-points generated by the VPP control unit to the physically represented DER (power inverter) and the virtual DER units represented in the virtual environment running in the RTDS (see Figure 44). It also collects the relevant data measured from the physical power inverter and from the virtual units represented in the virtual environment for further analysis. Using this scheme, the Lab Device Manager communicates the resulting participation factor for each DER unit at the laboratory setup, which is kept unchanged during the 4 second intervals, while recording the relevant measurements with a 200ms sample time for later analysis and comparison with the desired output, i.e. DA bid power reference plus the effect of the AGC signal.

Meanwhile, the electrical connection between the physically represented DER unit represented by the power inverter and the virtual network running in the RTDS (represented in Figure 44) is performed using a PHIL setup, where a duplex fiber optic real-time communication link is used to close the power loop between the RTDS and the power amplifier.

Since the simulated one hundred 1-hour scenarios used for previous in silico tests could not be replicated given the direct correspondence of time in a physical laboratory setup, a total of 6 hours of test episodes, with a time horizon of 5 minutes each were randomly selected from an automatic

SGEVL Electric Panel

15kW Power Amplifier(Triphase PM15)

(LV 230/400V)

3ph + N

RTDS unit(OPAL RT 5600)

Transmission network offline loading

Physical assets

Virtual environment

Lab Device

Manager

Real-time communication

link (RT)

RL-based VPP

Control Unit

TCP/IP

⁄Power

InverterDay-ahead market bids

4 seconds AGC setpoints

Real-time data acquisition

TCP/IP

TCP/IP

Page 56: D3.2 Experimentation - Restable Project...9 levels of aggregation. A reserve capacity forecast is also provided, based on the low quantiles of aggregated production forecasts. Forecasts

This project has been funded by the ADEME « Investissement d’Avenir » (FR), the BMWi (DE) and the Foundation for Science and Technology (PT), within the ERAnNet Smart Grid Plus program

generation control (AGC) signal historical dataset. This equaled to 75 episodes summing to an overall total of 5400 time steps tested in laboratory.