Research On Analysis And Forecasting Of Short-term Electricity Price In Electric Power Market

Electricity price is the key to electric power market in a competitive environment. Accurately forecasting electricity price is not only helpful to constructing an optimal bidding strategy and maximizing profits for generation companies, but also conducive to reducing the costs for consumers. Therefore, the analysis and forecasting of electricity price is one of the most important issues to be solved in electric power market. So far, different forecasting methods have been proposed for achieving forecasting accuracy. Among these proposed methods are time series methods, artificial neural network algorithms, and so on. Affected by many factors, electricity price movements exhibit a larger volatility, making the accurate price forecasting very difficult. Therefore, the forecasting methods and accuracy should be further improved.This thesis focuses on the short-term day-ahead electricity price analysis and forecasting:Firstly, the history of restructuring the electric power industry is briefly introduced, and the significance of electricity price forecasting is emphasized, the advantages and disadvantages of the proposed forecasting methods are also discussed. Secondly, based on the data from a truly electric power market, factors that affect the electricity price and characteristics that are distinct the electricity price from those of other commodities are analyzed and verified in detail. Thirdly, electricity price forecasting models are established based on the BP neural network algorithm and the least squares support vector machine (LS-SVM) method, respectively. Comparison of these two forecasting models is conducted. Results show that the forecasting accuracy of the LS-SVM model is higher than that of BP neural network model. Finally, a short-term electricity price forecasting method based on time series decomposition is proposed, combining the LS-SVM algorithm.In this newly proposed method, original time series of electricity prices are firstly decomposed into weekday series and weekend series. And then, these two series are further divided into three components:trend component, periodic component and random component, using moving average method and discrete Fourier transform. To obtain the forecasted values of these two time series of electricity prices, these three components are forecasted using moving average method, extrapolation method and LS-SVM method, respectively. Finally, the performance of the proposed method is verified based on the data from a truly electric power market. It is proven that the proposed method is effective and practical. Compared with the traditional BP neural network method and the LS-SVM method, the proposed method in this thesis has higher forecasting accuracy, its mean absolute percentage errors are within 7%.