JIANG Lihui, ZHOU Lidong, REN Jianguo, KE Peng, TIAN Linjing, WANG Fei
Accurate forecasting of day-ahead electricity price is of great significance to the decision-making optimization for electricity market participants. At present, most day-ahead electricity price forecasting methods do not distinguish daily fluctuations of electricity prices and adopt unified models. When the forecasted day's fluctuation pattern is different from historical data, the forecasting accuracy cannot be guaranteed. Based on different fluctuation patterns, classification modeling that uses similar historical data is an effective way to solve this problem. As a result, it is necessary to establish a classification recognition model for different patterns of historical data and a prediction model for future fluctuations. For this reason, this paper proposes a day-ahead weighted combination forecasting method for daily fluctuations of electricity price that serves for classification forecasting. Firstly, the K-means algorithm is used to cluster the daily electricity price series. On the basis of clustering analysis results, the feature vector reflecting the difference of daily fluctuation patterns is extracted. The support vector machine for classification （SVC） method is used to establish a recognition model for daily fluctuation patterns of electricity price data. Secondly, multiple conventional methods are applied to establish day-ahead electricity price forecasting models, and the forecasting results are input into the daily fluctuation pattern recognition model to obtain corresponding pattern recognition results. On the basis of the different pattern forecasting precisions of historical results from multiple models, a credibility based combination mechanism is designed to realize weighted combination forecasting considering the prediction accuracy, and the final forecasting results of the daily fluctuation model is obtained. Simulation analysis using the electricity price data of the PJM electricity market in the United States shows that the proposed forecasting method for future electricity price fluctuations can obtain relatively accurate forecasting results|and compared with unified forecasting the accuracy using this method for classification forecasting of electricity price is improved.
