Competitive Energy Generation Scheduling in Microgrids

Lian Lu*, Jinlong Tu*,Minghua Chen

Chi-Kin Chau Xiaojun Lin

2

Power Grid Landscape

Transcos

Transmission: enerator to

substation

generation

transmission

distribution

Transmission: from generator to substation, long distance, high voltageDistribution: from substation to customer, shorter distance, lower voltage

generation

substation

3

Power Grid Abstraction and Challenges

Households

Electricity Grid

Electricity

Power Generator

Wind Energy

Electricity

Electr

icity

Grid ReliabilityEconomic loss due to outage: 150 billion USD / year

Renewable IntegrationIntermittent generation makes it difficult to balance supply and demand

Generation Efficiency62% energy loses in conversion

Microgrid opens up new design space here!

Source: U.S. Energy Information Administration, Annual Energy Review 2011, 2012. DoE Intro to SG, 2008

4

Microgrid: A New Paradigm

□ Microgrid is a distributed power system that satisfies heat and electricity demand with two sources of supply:– External electricity and heat supply– Local generation (renewable, combined heat and power(CHP))

Households

Wind Energy

Combined Heat and PowerLocal Heating

Electricity Grid

Natural Gas

Microgrid

Heating

Elec-tricity

Elec-tricity

Heating

Electricity

Gas

Households

Electricity Grid

Electricity

5

Potential Benefits of Microgrids

Today’s Challenge The Microgrid Solution

Reliability Local generation provides heterogeneous power quality and reliability

Renewable Integration

Absorb the renewable uncertainty locally, reducing its negative impact on grid stability

Generation Efficiency

CHP-generation improves generation efficiency by 2-3x

• Worldwide deployment of pilot microgrids in the US, Japan, Europe, and more.

6

Worldwide Deployment of Pilot Microgrids

□ Lawrence Berkeley Lab project of a San Francisco hotel – Cost reduced by 12.4% & carbon footprint by 13.7%

NTT’s microgrid, Japan

Bornholm Microgrid, Denmark

Scheneider microgrid, Berlin, Germany

UCSD microgrid, San Diego, US

7

A Key Problem in Microgrid Operation

□ Real-time balancing supply and demand with minimum cost– Electricity cannot be stored cheaply, yet

Picture source: Lawrence Berkeley National Laboratory, 2007.

Energy Generation

Decision Module

Electricity demand

Heat demand

Grid electricity and natural gas tariffs

Wind/solar generation and CHP generation

8

Prior Approaches: Predict-and-Optimize

□ Assumption: Demand is predictable□ Applications: Unit commitment and economic

dispatching

Time

Electricity Demand

Electricity Supply

Supply Capacity

9

Microgrid Demands Are Difficult to Predict: Prior Approaches Fail

□ Prior assumption:– Demand is predictable

□ Local (net) demands are difficult to predict– Net electricity demand

inherits uncertainty from wind and solar

– Electricity and heat demands express different uncertainty pattern

10

Conventional Wisdom and Our Perspective

□ Model-centric perspective:

Model the future, then optimize accordingly.

□ Optimal but (usually) rely on accurate modeling/prediction

□ Solution-centric perspective:

□ Do not rely on modeling/prediction

□ Characterize the value of modeling/prediction

□ Can explore prediction to further improve performance

11

Our Contributions

Prior Art:Offline

Our Competitive Approach:Online – CHASE

Predict and optimize

No performance guarantee in the

presence of uncertainty

When future input is completely unknown: Competitive Ratio (CR)of CHASEfast-resp. < 3

That is, for arbitrary input,

CR improves with look-ahead

CHASE – Competitive Heuristic Algorithms for Scheduling Energy-generation

12

Problem Formulation

□ Challenges:– Generator characteristics makes decisions across slots coupled: on/off

and output-level decisions depend on future input– Future demand and renewable generation are highly volatile and

unpredictable

Microgrid operational cost in [0,T]

generator operating constraints

supply-demand constraints

capacity constraints

13

“Make It Simple, But Not Simpler.”

Problem with multiple generators

Problem with multiple fast-responding generators

Problem with single fast-responding generator

Solve the simplified problem

Problem with single fast-responding generator

Find an optimal offline solution

CHASE the offline optimal in an online fashion

Characterize the Competitiveness

14

Problem with Single Fast-Responding Generator

CHP generator

External generation:on-demand but

expensive

Local CHP generation: cheap but has start-

up cost

Electricity Grid

Separate Heating

Net electricity demand

Heat demand

15

Dense Demand: Local GenerationSparse Demand: External Generation

CHP generator

External generation:on-demand but

expensive

Local CHP generation: cheap but has start-

up cost

Electricity Grid

Separate Heating

Net electricity demand

Heat demand

Net electricity demand

Heat demand

Sparse demandDense demand

16

OFAyOffline optimal solution

CHASEsyCHASE ‘s solution without look-head

CHASElk( )sy

CHASE’s solution with look-head

Solutions and IntuitionsArrivals of demand

(t) type-start type-1 type-2 type-1 type-end

-

0

𝑇 1𝑐 𝑇 2

𝑐 𝑇 3𝑐 𝑇 4

𝑐𝑇 5𝑐𝑇 0

𝑐~𝑇3

𝑐~𝑇2

𝑐~𝑇1

𝑐

𝜔 𝜔 𝜔𝜔

Net electricity demand

Heat demand

Function captures “sparseness/denseness” of

the demand

17

Extending CHASE to Multiple Generators

□ Theorem: Layered partition induces zero optimality loss.

Gen. #1

Gen. #2

Electricity Grid

Separate Heating

External generation

18

CHASE Is the Best Deterministic Algorithm

□ Theorem: For the problem with multiple fast-responding generators, CHASE achieves the best competitive ratio (CR) of all deterministic online algorithms:

CR(CHASE) ,

where captures maximum price discrepancy between local generation and external generation.

□ Price of uncertainty is at most 3

19

Worst Demand vs. Real-world Demand

𝑇 0𝑐 𝑇 1

𝑐 𝑇 2𝑐

OFAy

SCHASEy

0

(t)

OFAy

SCHASEy

a(t)

h(t)

Worst Case Demand Real-world Demand

20

Look-Ahead Improves Performance

□ Theorem: For the problem with multiple fast-responding generators and look-ahead window size ,

CR(CHASE) ,

where captures the benefit of look-ahead.

𝑡𝑡+𝜔

look-ahead window

21

Numerical Evaluation

□ Electricity/heat demand traces from a San Francisco college

□ Power output from a nearby wind station

□ Compare OFFLINE, CHASE, and an alternative RHC

(a) Summer week

(b) Winter week

22

Benefit of Combined Heat and Power (CHP)

□ CHP leads to additional 10% cost saving

□ CHASE performs close to the offline optimal

□ RHC performs badly and may give zero cost saving

(a) With CHP

(b) Without CHP

23

Benefit of Looking Ahead

□ CHASE saves 20% cost even without look-ahead□ CHASE is robust to prediction error

24

Conclusions

□ We propose CHASE: a paradigm-shift solution for energy generation scheduling in microgrids– Does not rely on demand prediction– Achieves the best competitive ratio without look-ahead– Performance improves with look-ahead– Leads to 20% cost reduction in case studies

Households

Wind Energy

Combined Heat and PowerLocal Heating

Electricity Grid

Natural Gas

Microgrid

Heating

Elec-tricity

Elec-tricity

Heating

Electricity

Gas

Minghua Chen (minghua@ie.cuhk.edu.hk)http://www.ie.cuhk.edu.hk/~mhchen

Thank you!

CHASE – Competitive Heuristic Algorithms for Scheduling Energy-generation

CHASE – Track the offline optimal in an online fashion