Modelling the Electricity Market European Summer School München, July 3rd, 2009 Dipl. Wi.-Ing. Serafin von Roon 11 Research Center for Energy Economy Am Blütenanger 71 80995 München SRoon@ffe.de
Agenda Overview over Electricity Markets Different Models of Pricing Approaches of Modelling Electricity Prices Results and Comparison of two different Approaches 2
Overview over Electricity Markets before Liberalisation and Unbundling Industry Electric Power Company Trade & Commerce Customer Private Households 3
Overview over Electricity Markets since Liberalisation and Unbundling Power Generation Load-Frequency Control OTC Trade Exchange (EEX) Power Network Sales & Distribution Industry Trade & Commerce Customer Household 4 EEX: Derivatives; Day-ahead; Intra-day
Pricing: European Energy Exchange (EEX) without a price (no alternative), demand must be covered successful unsuccessful 50 Supply Demand Price [a.u.] 40 30 20 10 theory:{ profit Market Clearing Price theory: marginal costs of different power plants theory: costs of best alternative to cover the demand, e.g. individual contracts, self-generating 5 0 0 10,001 20,002 30,003 Load [a.u.] FfE 502.13_00207
Pricing: Load Frequency Control Power successful unsuccessful 40 demand 30 Price [ /MW] 20 10 marginal costs + calculated profit technical need 0 0 1,000 2,000 3,000 4,000 5,000 6,000 Cumulated Reserve Power [MW] FfE 502.13_00203 6
Approaches of Modelling Electricity Prices Approach Fundamental by Neural Networks Statistical Agent-based Essential Factors costs of generation of electricity technical conditions consumer load profile historic data of correlations of input and output modelling correlation function with least error historic data of the energy market individual optimisation agents improve their strategies based on experience FfE 502.13_00294 7
Fundamental Approach - Methodology based on the load profile that has to be covered by conventional power plants the merit order is a result of the power plant data base For each hour there we get a certain price according to the load needed to be provided 8
Fundamental Approach - Results Mon Tue Wed Thu Fri Sat Sun 200 150 EEX Fundamental Approach 250 % 200 % EEX Fundamental Approach Price in per MWh 100 50 150 % 100 % 50 % 0 FfE 502.13_00235 0 % FfE 502.13_00233 0 100 200 300 15-Jan-07 16-Jan-07 17-Jan-07 18-Jan-07 19-Jan-07 20-Jan-07 21-Jan-07 Day of the Year 2007 2007 EEX Fundamental Approach average price in per MWh 37.98 54.76 FfE 502.13_00234 9 The two graphs show the characteristic of a whole year and an exemplary week (normalised with the week mean price). The left graph shows that the fundamental approach mostly overestimates the real price for electricity, this is also proved by the data. The other graph illustrates that the characteristic on weekdays is achieved pretty well while the weekend price fluctuations can hardly be recognised.
Fundamental Approach - Weaknesses weaknesses of this approach: In this calcualtion the model is too static, changing costs for fuel or CO 2 -certificates are not regarded extreme weather conditions are not taken into account as the graph shows about 800 hours - which equals almost 10 % of the year s prices - cannot be reached by the model, especially high prices do not occur 10
Design of Synthetic Neural Networks background is the composition of the human brain (real neural network) neurons are stimulated by input signals; depending on processing functions in neurons output signals will result by repeating correlations of input and output signals (training) the connections to the neurons will increase (learning) synthetic neural networks consist of layers with synthetic neurons by entering input parameters and defining associated output parameters the network can be trained backpropagation algorithm (change of the connection weightings) Inputs Source: Westdeutscher Rundfunk, Köln, 2009 Input Layer Hidden Layer Output Layer Output 11 Connection Weightings
Implementation of a Synthetic Neural Network for Modelling Electricity Prices training network with influencing factors on the electricity price in 2007 input parameters in resolution of one hour consumer load wind load coal price gas price oil price price of CO 2 emission certificates temperature atmospheric fallout 12 input parameters with a resolution lower than one hour are approximated (e. g. coal with quarterly prices) modelling electricity prices for 2007 and 2008 by entering associated input parameters to the 2007-trained network
Results of Modelling Electricity Prices for 2007 and 2008 300 250 Simulated Electricity Price in 2007 Real Electricity Price in 2007 300 250 Simulated Electrcity Price in 2008 Real Electricity Price in 2008 200 200 Price in per MWh 150 100 50 0 Price in per MWh 150 100 50 0-50 -100 Hour of the Year 2007 FfE 502.13_00229 0 2000 4000 6000 8000 comparison over a time period of 3 days: -50-100 0 2000 4000 6000 8000 Hour of the Year 2008 FfE 502.13_00230 13 Energy Price in per MWh Time of Day 12 AM 6 AM 12 PM 6 PM 12 AM 6 AM 12 PM 6 PM 12 AM 6 AM 12 PM 6 PM 12 AM 240 Simulated Price 200 Real Price 160 120 80 40 FfE 502.13_00232 0 23-Jun-08 24-Jun-08 25-Jun-08 26-Jun-08 Date some days match bad Energy Price in per MWh Time of Day 12 AM 6 AM 12 PM 6 PM 12 AM 6 AM 12 PM 6 PM 12 AM 6 AM 12 PM 6 PM 12 AM 120 Simulated Price 100 Real Price 80 60 40 20 0 FfE 502.13_00231 26-Jul-08 27-Jul-08 28-Jul-08 29-Jul-08 Date some days match well
Statistical Approach 350 300 Historic EEX-Prices Day-Ahead-Price in Day-Ahead-Price in 250 200 150 100 50 0 FfE 502.13_00296 01-Jan-02 01-Jan-03 01-Jan-04 01-Jan-05 01-Jan-06 01-Jan-07 31-Dec-07 Date 800 Modelled Price Paths 600 EEX-Data 400 200 0-200 Thedatabaseforthestatistical approach are the years 2002-2007. The changes of the daily mean prices are evaluated and are the basis of the model. The model does not reproduce the course of 2008 s prices at all. Instead the price reach extreme values and are not realistic. Therefore this model is not very useful. -400 14-600 FfE 502.13_00295 20-Oct-07 1-Jan-08 14-Mar-08 26-May-08 7-Aug-08 19-Oct-08 31-Dec-08 Date
Agent-based computational Economics (ACE) supply side agent 1 agent 2 apply their strategies market 1 market 2 demand side time variable fixed demand or consumer agents agent.. market.. next time step agents offer products on different markets agents learn and adopt their strategies pricing according to supply and demand profit payout 15
Agent-based computational Economics (ACE) application to electricity markets: agents can be: utilities, renewable energy producers, governments/regulators, consumers, traders, grid operators, etc. according to the research question and abstraction level the agents for the models are chosen the markets can be: spot market, balancing market, CO 2 market advantages: realistic behaviour of individual agents can be simulated the effects of a wide range of influences on the pricing can be tested disadvantages: high complexity of the model, difficult to comprehend and implement parameters can be set in order to achieve desired results 16
Summary and Conclusion There a different markets for electricity with individual pricing In Germany the most important market is the Spot market at the European Energy Exchange. These prices are reference prices for other markets There a different approaches to model electricity markets The presented approaches have difficulties to reproduce the historic extreme prices Predicting future electricity prices not only mean prices but also the temporally price structure - is a challenge 17
Discussion Thank you for the attention! 18