The Wavelet Neural Network Method Used For Short-term Electricity Price Forecasting Of The Power Systems

The marketization of power industry is a development trend of the current power industry in the world and a focus of the international electric scientific research and project practice. Electricity price is not only a signal of the power market supply-demand relationship, but also an economic lever of controlling electric marketing, therefore, Electricity price issues are the key problems in the market and how to price the special commodity-electricity is essential for the smooth market operation. And how to use the relative historic data to forecast the future market clearing electricity price is a very meaningful work for every participator in the power market.After discussing the constituents and characteristics of the electricity price, analyzing and comparing the merits and shortcomings of some current forecasting methods. Based on some basic theoretical aspects of the Wavelet Transform and Neural Networks, the advantages in the electricity price forecasting are analyzed, then describing two methods for embedding the discrete wavelet transform into neural network-based short-term electricity price forecasting. The Multi-Resolution Analysis (MRA) approach for tackling feature extraction problems decomposes the electricity price series into one low-frequency and some high-frequency sub-series in the wavelet domain. Using this new representation of the original electricity price signal, two different alternatives are investigated. The first one using different ANN models independently forecasting each sub-series, adding the forecasting results of all sub-series, then the whole forecasted electricity price series are obtained. The second alternative creates a model for short-term electricity price forecasting, whose inputs are based on information from the original electricity price series and from wavelet domain sub-series, too. In order to overcome the shortcoming of BP neural network, this article proposed duplicates forecast method and the threshold value processing method revise forecast result, thus enormously increased the forecast precision and the stability. The outcome of the study clearly indicates that the second model has higher accuracy, and the first one is better in the aspect of the stability. However, bothof two methods reduce the error of peak and valley electricity remarkably.Finally, this paper point out that in order to improve the accuracy, theconstituents and characteristics of the electricity price must be analyzed first based on the given power market, and dealing the choice of the input variables well, designing the model matching the characteristics of the electricity price. The data used in this paper come from California Power Market.