Research On Hybrid Forecasting Model Of Short-term Electricity Demand

In recent years, with the continuous reform of electric power industry system in China, the competition mechanism in power market has gradually formed, in the competitive electricity market, how to forecast the short-term electricity demand accurately is becoming more and more important. More accurate demand forecasting is conducive to the power suppliers to adjust the power supply strategy, thus to make the maximum profit, at the same time it is also useful for the ordinary users to make power purchase plan which can get the minimum cost. However, the electricity demand is affected by weather, season, economic, policy and several other factors, making the accurate forecasting of electricity demand work becomes very difficult and complex.Therefore, based on the research status of the short-term electricity demand forecasting and the main factors affecting the demand, this paper proposes a hybrid forecasting model by cluster analysis and least square support vector machine.This paper first gives an in-depth analysis to the Australian NSW and Queensland electricity demand data, then the experimental data set is classified into different clusters by the K-means or SOM network, and each of them contains similar objects. When the clustering process has been completed, the LSSVM regression algorithm for each cluster is constructed. For the LSSVM model, there seems no ideal way to determine the parameters γ and δ2. Thus, based on the MATLAB parallel compute toolbox, we propose a parallel with 2-step grid search algorithm. It is a local optimal search algorithm which can get the global optimization result in a shorter period of time than the traditional grid search algorithm with the cost of prediction accuracy is limited. Finally, the empirical testing shows that the results of the mean absolute percentage error obtained with the SOM-LSSVM methods can achieve better prediction accuracy compared with the K-means-LSSVM and a single LSSVM models. At the same time the forecasting accuracy of our method is better than ARIMA, ARIMA-BP and WT-LSSVM model. The hybrid model proposed in this paper is a comparative ideal forecasting method of the short-term electricity demand.