Research And Application On A Hybrid Method For Short-term Load Forecasting Based On Ensemble Empirical Mode Decomposition

Electricity demand forecasting is not only the basic work of the electric power system, but also the focus and hotspot in data mining area. Reasonable demand forecasting can effectively reduce the costs of power enterprise as well as improve the economic and social benefits of the whole community under conditions of ensuring social normal production and people’s daily life. Therefore, it is a very meaningful and important thing to carry out electricity demand forecasting. Moreover, with the rapid development of power industry and the marketization of power enterprises, the requirements of forecasting methods and forecasting accuracy are continuous to enhance and improve. In recent years, providing effective, reasonable and accurate electricity demand forecasting methods has become more and more critical.As a single linear or nonlinear method may not capture all information of a real complex electricity systems, this paper proposes a novel hybrid method based on ensemble empirical mode decomposition, time series, grey theory, phase space reconstruction, least square support vector machine and particle swarm optimization. Firstly, the adaptive, fully data-driven ensemble empirical mode decomposition method is utilized to eliminate the noise signals from original data to reduce adverse influence on the forecasting accuracy. Secondly, Four typical methods including seasonal aggressiveness integrated moving average model SARIMA, chaotic time series CHAOS, grey model GM(1,1)and least square support vector machine LSSVM are respectively adopted to forecasting the following electricity demand data utilized the original data preprocessed by ensemble empirical mode decomposition. Lastly, The final forecasting result of the hybrid method is obtained by combining the forecasting results of four typical methods with corresponding weight factor optimized by particle swarm optimization.Last but not least, a case study is carried out with the actual electricity demand data collected from New South Wales of Australia. Meanwhile, three evolution criteria are employed to evaluate the forecasting performance of various forecasting methods mentioned in this paper. Through research and analysis on the finally prediction results, the proposed hybrid method can significantly enhance forecasting performance and improve the forecasting accuracy comparing with the other four forecasting methods used in this paper.