Price forecast in the competitive electricity market by support vec - tor

Price forecast is a task challenging and very important in competitive electricity market context. Both market players and regulators concern very much about the price evolution, on one hand, the prediction of the market price is a crucial information for the production arrangement and bidding strategies. On the other hand, the regulators need to analyze the market behavior and monitor the market evolution. Price forecast tools provide them a useful tool to foresee the future market and examine the potential regulations. The price forecast tools are categorized in two fields. One is the detailed market simulation [1]-[4], which needs lots of market information, the other price forecast technology refers to those mathematic approaches without market modeling. The most popular approach is the time series algorithm, which has been widely used in many forecast fields. Garcia introduced GARCH forecast model to capture the price volatility for next day electricity price prediction [5]. Contreras applied ARIMA models to predict the next day electricity price[6], while Conejo developed a hybrid ARIMA approach with wavelet transform for ill-behaved price series[7]. In [8] dynamic regression and transfer function models are proposed by Nogales. Artificial Intelligence (AI) that has been also widely used in load forecast is popular for price forecast. Szkuta proposed a three-layered artificial neural network (ANN) with back propagation for price prediction in [9]; support vector machine (SVM) as a new AI method has shown its excellent performance for classification, regression and pattern recognition, and has been successfully applied in load forecast [10], some preliminary attempt has been implemented in price forecast [11] as well. Price forecast can be classified in three types according to the time frame, Short term price forecast; Medium term price forecast; Long run price forecast. Apparently, the short-term price forecast is the most elusive, due to the incomplete information or uncertain strategic bidding. With the time frame enlarged, the price is more and more reasonable with reference to the corresponding context, namely it is easier to find the mapping relation between the context and the price. Hence the major difficulty for the long run price forecast is the uncertainty of the future context. The market price will be hugely influenced by the subjective bidding strategies and the