Forecasting and decision making in electricity markets with focus on wind energy

Forecasting and decision making in electricity markets with focus on wind energy

PhD project by Tryggvi Jonsson

Short introduction video with Tryggvi Jonsson

This thesis deals with analysis, forecasting and decision making in iberalised electricity markets. Particular focus is on wind power, its interaction with the market and the daily decision making of wind power generators. Among recently emerged renewable energy generation technologies, wind power has become the global leader in terms of installed capacity and advancement. This makes wind power an ideal candidate to analyse the impact of growing renewable energy generation capacity on the electricity markets. Furthermore, its present status of a significant supplier of electricity makes derivation of practically applicable tools for decision making highly relevant.

The main characteristics of wind power differ fundamentally from those of conventional thermal power. Its effective generation capacity varies over time and is directly dependent on the weather. This dependency makes future production uncertain and difficult to contract even on a day-to-day basis. Consequently decisions about market bids for next-day delivery are based on production forecasts which are bound to come with some uncertainty. Naturally markets that experience large scale integration of wind power are affected by these different characteristics. The thesis presents analyses of how this impact is realised in markets significantly penetrated by wind power. Due to its representation by forecasts in the supply curve, such predictions are used to describe their non-linear influence on the market prices.

Methods adequately accounting for this effect in models for day-ahead forecasting of the prices are also presented in the thesis. Prompted by the volatile behaviour of electricity markets, considerable focus has been on time-varying and robust parameter estimates. The models derived are all based on well know methods from the statistical literature.

The stochastic production of wind turbines prompts the need for alternative methods for optimally bidding wind power to day-ahead markets. Such bidding strategies are formulated in this thesis, which utilise the information provided by the market models. Bids that maximise expected revenues are found and the possibility of risk averse behaviour is discussed.

The main differences between the renewable energy sources like those previously mentioned and conventional power plants are owed to the characteristics of the fuel. Directly relying on nature for fuel supply causes the effective generation capacity of these plants to vary over time in a manner best characterised as stochastic. Thus contracting exact amount for future delivery can be problematic. With the fuel available however, these plants can generate electricity at virtually no cost and are not subject to the same constraints as the conventional ones.

These differences naturally call for different approaches to various aspects of their operation in general. Among those things is their participation in the market which is what eventually ensures the investor's payback. This thesis addresses some aspect of the interaction between wind power and electricity markets. The effect of its presence on the market is analysed and tools for assisting its participation in the market are presented. The latter involves analysis and development of tools for optimally bidding wind power to the market and also derivation of models that can provide the forecasts to be used in the decision making. More precisely the objectives of the thesis are as follows:

  • To analyse and improve understanding of how electricity prices are influenced by large scale wind power integration.
  • To analyse proposed bidding strategies for wind power producers, develop them further and understand which forecasts are required for adopting such strategies in practise.
  • To utilise the interaction between wind power production and the market along with other known impact factors to construct models that are capable of issuing accurate forecasts of the day-ahead and real-time market prices.

The focus is thus on challenges that directly relate to the daily operation of a wind power producer. No attention is however paid to the long-term management challenges such producers face, e.g. pricing of financially settled contracts for risk management and pricing of long-term delivery contracts.

Due to the stochastic generation of wind turbines, their operators must rely on forecasts for trading on a day-ahead basis. The production forecasts therefore represent the wind power in the supply curve and not the actual production. This motivated that, in contrast to previous analysis of the relationship between wind power production and day-ahead prices, the one presented in this thesis considers production forecasts instead of actual production. In addition, by accounting the non-linear aspects of the analysed relationship through non-parametric models, the impact of wind power on the prices is shown to be more substantial than previously demonstrated. The figure ... demonstrates how this effect actually appears.

These effects become an additional source of variation for the prices which already are known to exhibit features such as non-stationarity, multiple seasonal cycles, mean-reverting spikes, positive skewness and high kurtosis. A common methodology for modelling the expected prices has therefore been the well known ARIMA model, fitted in terms of the logarithmically transformed prices. The model presented in this thesis is however estimated on the original scale of the prices. Correspondingly, forecasts are issued on the same scale. Recursive and robust parameter estimation are instead used to accommodate the varying dynamics of the prices. Similar transformations and parametric assumptions are also the literature's prevailing approach to density models for the prices. Conversely, a semi-parametric density forecast method is presented in this thesis which also deals with untransformed data.

Alternatives to contracting the expected wind power production in the day-ahead market have received increased intention in the literature in recent years. The bidding strategy presented here builds on previous work in the field and aims at maximising expected hourly revenues. Modifications are made to account for practical constraints and on the contrary to many previous studies, who base their results on constant expectations for the market, trading is simulated based on actual forecasts for the cost of imbalances. The derivation of the model used to generate these forecasts is presented in the thesis as well. There, efforts are made to accommodate the regime switching behaviour of the imbalance costs in the modelling process.

The models presented are all developed with practical applicability in mind. Results are therefore in all cases derived by mimicking the real-life circumstances. The models used for analysis and forecasting are all statistical ones, built on established and well known models from the statistical literature. The characteristics of the subject has prompted focus on adaptive and robust parameter estimation along with other models that are able to capture varying dynamics. Most of the work presented in this thesis is done using data from the Western Danish grid area which comprises Jutland, Funen and the islands west of Storebælt.

<html>
  <head>
    <title>
      IMM - Energy
    </title>
  </head>
  <body>
    ...
<!--
Under construction,
check back later.
-->
    ...
  </body>
</html>