Research On Combination Model For Short-term Load Forecasting Based On Emd And Gp Regression Theory

Power short-term load forecasting is an important part of dispatch and operation for power system, which accuracy will directly affect the security, economical efficiency and quality of power systems. With the development of the artificial intelligence, many new methods and technologies of short-term load forecasting are emerged in large numbers, which raise a high requirement for short-term load forecasting. Therefore, how to improve the forecasting precision is the emphasis on the study of short-term load forecasting.At first, the constituents, characteristics and influenced factors of the short-term load was analyzed in this paper. It indicated that the short-term load is a non-liner and a non-smooth process, the changes of load data are influenced by many factors. The relevance is most evident between average temperature and load data, so the paper takes the average temperature into account. Then summarized the methods recently, and firstly used Gaussian regression processes in the short-term load forecasting. As to the nonlinear relationship between load and the influencing factors, historical load data and meteorological factor were taking into account, and the maxima load data of last five day before predicting day and the predicting daily average temperature were used as the input data, then the short-term load forecasting model was built based on Gaussian regression processes. The proposed model has such advantages as few parameters, easy to optimize parameter, and fast convergence.Short-term load of power systems can be considered as linear combination of sub-series characterized by different frequencies, so the Empirical Mode Decomposition (EMD) was used for pre-procession of short-term load forecasting data. From the composition of load data, based on EMD the load series is decomposed into different lots of stable series and a surplus component, then take Gaussian regression processes theory establish model, finally reconstructed the forecasted signals of the components and abstain the ultimate forecasting result. The simulation shows the accuracy of the proposed combined model is considerable, and it is feasible to introduce Gaussian regression processes to the electric power load forecasting field. The proposed method has high accuracy while the change trend of load can be well reflected.