Time Series Aggregation Module (TSAM)

Hannes Kachel

Technische Universität Berlin | Department of Energy and Resource Management | 05.12.2019

Dezentrales Logo

optional Page 2

1

2

3

4

Idea TSAM – without storages

Idea TSAM – with storages

Application and first results

Motivation 4

6

14

17

Page

Outline

Dezentrales Logo

optional

1

2

3

4

Idea TSAM – without storages

Idea TSAM – with storages

Application and first results

Motivation 4

6

14

17

Page

Outline

Page 3

Dezentrales Logo

optional

Why do we use TSAM?

Page 4

Bildquelle: Iks-gmbh (2017).

Dezentrales Logo

optional

1

2

3

4

Idea TSAM – without storages

Idea TSAM – with storages

Application and first results

Motivation 4

6

14

17

Page

Outline

Page 5

Dezentrales Logo

optional

Sequence of time series reduction

pre-processing of

time series

aggregation

algorithm

add extreme periods

back-scaling

Page 6

Quelle: Kotzur et al. (2018).

Dezentrales Logo

optional

1. Pre-processing of time series

normalisation

-> ∈ (0, 1)

choose length and number of typical periods

Page 7

Quelle: Kotzur et al. (2018).

Dezentrales Logo

optional

Example pre-processing:

typical day for one application

• one application -> a = 1

• one year with 365 days

-> i = 365 candiates for

typDays

• one day with 24 hours

-> g = 24

Page 8

Quelle: Kotzur et al. (2018).

Dezentrales Logo

optional

2. Clustering days of two applications

• # cluster a priori

• clustering similar days

• -> k typDays

Page 9

Quelle: Kotzur et al. (2018).

Dezentrales Logo

optional

3. There are 4 methods for extreme periods

Page 10

Quelle: Kotzur et al. (2018).

Dezentrales Logo

optional

4. Back scaling of aggregated time series

• goal: no systematic errors

average aggregated time series = average

full time series

Page 11

Quelle: Kotzur et al. (2018).

Dezentrales Logo

optional

1

2

3

4

Idea TSAM – without storages

Idea TSAM – with storages

Application and first results

Motivation 4

6

14

17

Page

Outline

Page 12

Dezentrales Logo

optional

There are two time layers:

intra and inter-periods

1. intra-periods

– As before

2. inter-perioden

– Storage level between periods

3. Superposition

– sum of sub storage levels equals storage level

Page 13

Quelle: Kotzur et al. (2018).

Dezentrales Logo

optional

Storage level is deconstructed

• i: days

• g: hours

• blue: intra

• red: inter

superposition:

𝑥𝑖𝑛𝑡𝑟𝑎 + 𝑥𝑖𝑛𝑡𝑒𝑟 = 𝑠𝑡𝑎𝑡𝑒 𝑥

Page 14

Quelle: Kotzur et al. (2018).

Dezentrales Logo

optional

Calculation of a storage level delta

for each typical period

• i: days

• g: hours

• red: intra

• blue: inter

• k: typical

Day

Superposition:

𝑥𝑖𝑛𝑡𝑟𝑎 + 𝑥𝑖𝑛𝑡𝑒𝑟 = 𝑠𝑡𝑎𝑡𝑒 𝑥

Page 15

Quelle: Kotzur et al. (2018).

Dezentrales Logo

optional

1

2

3

4

Idea TSAM – without storages

Idea TSAM – with storages

Application and first results

Motivation 4

6

14

17

Page

Outline

Page 16

Dezentrales Logo

optional

Application in a dummy dispatch model

1. Easy application

• great documentation

• many examples

2. Special features

• Pre aggregation with chosen data

• Disaggregation

• RMSE, RMSE duration, MAE

3. Chances

• Cluster Indices only for hierarchical clustering

• Idea and calculation of storage level deltas

Page 17

Dezentrales Logo

optional Page 18

Results with dummy model

Variation of

• number typical periods

• length of typical periods (not shown)

• 1 year horizon

-35

-30

-25

-20

-15

-10

-5

0

5

10

0 10 20 30 40 50

diffe

rence

in s

olu

tion

tim

e

number of typical periods

Absolute difference of solution time to no TSA

-0,02

-0,015

-0,01

-0,005

0

0,005

0,01

0,015

0,02

0,025

0 10 20 30 40 50

rela

tive e

rror

number of typical periods

Relative error of objective to no TSA

Quelle: Eigene Darstellung.

Dezentrales Logo

optional

Thank you

very much

for your

attention

Dezentrales Logo

optional

Literaturverzeichnis

Seite 20

Kotzur, L., Markewitz, P., Robinius, M., Stolten, D. 2017: Impact of different time series aggregation methods on optimal

energy system design. In: Renewable Energy Vol. 117, S. 474 – 487.

Kotzur, L., Markewitz, P., Robinius, M., Stolten, D. 2017: Time series aggregation for energy system design: Modeling

seasonal storage. In: Applied Energy Vol. 213, S. 123 – 135.

Dezentrales Logo

optional

Bildquellenverzeichnis

Seite 21

Iks-gmbh.com. 2017: Anmeldung zum IKS-Thementag "Komplexität beherrschen". https://www.iks-

gmbh.com/unternehmen/news/einladung-zum-iks-thementag-komplexitaet-beherrschen-agiles-arbeiten-und-

microservices-architekturen/online-anmeldung-zum-iks-thementag-komplexitaet-beherrschen-am-28-03-217/, 30.

12.2019.