1. Modelling Electricity Transitions with an Agent-Based Model
Jörn C. Richstein

2. Why ABM for my question?

3. Challenges in Transition Modelling
of the Electricity System
Electricity
System
CO2
RES
Cong. Mgtm.
Security of Supply
CO2 Emissions
Prices
Generation
Policy
Fuel Prices
Demand
Technologies
Quite a typical representation of an energy-system: input/output description + influences
However, modelling the system over longer time-periods is a complex task:
Abundance of exogenous uncertainties
Interactions between neighboring electricity systems
Influence of overlapping policy instruments

4. Challenges in Transition Modelling
Path Dependence
Electricity
System
CO2
RES
Cong. Mgtm.
Security of Supply
CO2 Emissions
Prices
Generation
Policy
Fuel Prices
Demand
Technologies
Electricity
System
CO2
RES
Cong. Mgtm.
Security of Supply
CO2 Emissions
Prices
Generation
Policy
Fuel Prices
Demand
Technologies
Differences in:
Technology specific revenue structure
Technology specific learning
Institutional arrangement
Electricity
System
CO2
RES
Cong. Mgtm.
Security of Supply
CO2 Emissions
Prices
Generation
Policy
Fuel Prices
Demand
Technologies
Electricity
System
CO2
RES
Cong. Mgtm.
Security of Supply
CO2 Emissions
Prices
Generation
Policy
Fuel Prices
Demand
Technologies
Electricity
System
CO2
RES
Cong. Mgtm.
Security of Supply
CO2 Emissions
Prices
Generation
Policy
Fuel Prices
Demand
Technologies
Electricity
System
CO2
RES
Cong. Mgtm.
Security of Supply
CO2 Emissions
Prices
Generation
Policy
Fuel Prices
Demand
Technologies
Investments in power systems are very long lived
They create path dependencies, by:
Technology-specific changes of prospective revenues of future investments
Driving technological learning
Create institutional pathways
Example: German decision to invest in PV:
Drove learning-by-doing, which had a global impact on prices
Changed the merit order, by removing the midday peak in summer, reducing peak plant / pumping storage business case
Result:
multiple future equilibria
Historical path dependence
Non-predictability
Possibilities of lock-in
-> so not static efficiencies (or metrics) matter, but also dynamic metrics under uncertainty.
t=0
t=1
t=2

5. Challenges in Transition Modelling
Heterogeneous Agents & Expectations
Electricity
System
CO2
RES
Cong. Mgtm.
Security of Supply
CO2 Emissions
Prices
Generation
Policy
Fuel Prices
Demand
Technologies
Electricity
System
CO2
RES
Cong. Mgtm.
Security of Supply
CO2 Emissions
Prices
Generation
Policy
Fuel Prices
Demand
Technologies
Electricity
System
CO2
RES
Cong. Mgtm.
Security of Supply
CO2 Emissions
Prices
Generation
Policy
Fuel Prices
Demand
Technologies
Opening the black box:
Since liberalisation independent actors
These actors are heterogeneous, and make independent decisions
Decisions are based on expectations:
Price (Fuel & Electricity)
Demand
Technological development
Policy
Competitor behaviour
Traditional approach: equilibrium economics with rational expectations, that are consistent with state of system -> homogeneous beliefs
An energy system in transition will often be out-of equilibrium, the adjustment processes, ie. the investments become important
Investment decision
Investment decision

6. Methodology Agent-based Modelling
Methodology where actors and environment are modelled from the bottom up by algorithmic description:
Naturally allows for heterogeneous agents
Differing levels of detail possible for technologies and actors
Data richness an option (e.g. to include locational specific meteorological data)
Focus on behaviour and emergence of phenomena
Informational asymmetry and multiple equilibria can be represented (L. Tesfatsion, 2006)
Since ABM is a generative methodology, out-of-equilibrium processes can be represented (W. B. Arthur, 2006)
Decisive for modelling slowly adjusting infrastructures in transition

7. EMLab Electricity Market Laboratory
Energy producers are the main agents, but other types exist, which incorporate simpler behaviour:
European government & national governments
Banks, commodity suppliers, etc.
Energy Producers make investments in power plants:
Based on a per-agent merit-order forecast
On a yearly basis
Energy producers bid into two-country electricity system, connected by market coupling and a common CO2 market
Market clearing based on stepwise load-duration curve approximation of demand

8. How?

9. EMLab Implementation Details Implemented in Java, build with Maven
Uses AgentSpring ( as an ABM Framework
Split of Roles (Behaviour) and Agents
Uses a graph database for data storage (Neo4j and Spring Data)
Available as open source on
Single run interfaces:
Web-based
Scriptable via R
Statistical evaluation of Monte Carlo runs with R

10. Agentspring: class structure
Domain
Specification of ‘things’ and their properties and possible relations to other ‘things’.
Role
Coded behavior, to be ‘acted’ by the agents from specific agent classes
Repository
For interaction with the database
Other
Trend: to be able to incorporate various types of trends in data
Util: helper classes
Note: properties and methods are inherited

13. The short term market

14. Short detour: The electricity market – During Peak Load

15. Short detour: The electricity market: During Base Load

16. The medium term

17. EMLab
CO2 market in more detail

18. EMLab - The carbon market

19. The long term

20. EMLab
Investment in generation capacity

21. EMLab
Investment in generation capacity

22. Which parts are typically ABM, which not?

23. Example results

24. EMLab
Example Run – Capacity Development

25. EMLab
Example Run – Capacity Development

26. EMLab
Example Run – CO2 Prices

27. EMLab
Example Run – CO2 Prices

28. EMLab
Example Run – CO2 Prices

33. Thanks for listening

34. Literature
L. Tesfatsion, Agent-based computational economics: A constructive approach to economic theory., in Agent-Based Computational Economics, vol. 2 of Handbook of computational economics, p , North-Holland, 2006.
W. B. Arthur, Chapter 32 Out-of-Equilibrium Economics and Agent-Based Modeling, in Agent-Based Computational Economics (L. Tesfatsion and K. Judd, eds.), vol. 2 of Handbook of Computational Economics, pp , Elsevier, 2006.
J. C. Richstein, E. Chappin, and L. D. de Vries, Impacts of the Introduction of CO2 Price Floors in a Two-Country Electricity Market Model, IAEE European PhD day at the 12th IAEE European conference, 2012.
A. Chmieliauskas, E. J. L. Chappin, C. B. Davis, I. Nikolic, and G. P. J. Dijkema, New methods for analysis of systems-of-systems and policy: The power of systems theory, crowd sourcing and data management, in System of Systems (A. V. Gheorghe, ed.), ch. 5, pp , InTech, March 2012.

35. EMLab
Demand
Approximation of the load duration curve by 20 segments Renewables have a segment specific contribution.

36. EMLab Short-term market Uniform price-volume auction
The carbon price is found via an iteration algorithm. It is adjusted until:
The carbon emissions are close to the cap (+- 5%)
Carbon prize is 0
The price floor is implemented as a complementary tax
If the market price is below the price floor, generators pay the price difference between the market and the price floor

37. Investment decision
Forecasting

38. Forecasting
Running Hour Estimation

39. Forecasting
Cash Flow Estimation

40. Forecasting
DCF Method