A Combined Model Based On EMD And Seasonal Exponential Adjustment Application In Power Load Forecasting

Due to the rapid development of power industry and modernization of the electricity grid management, people were becoming more and more widespread to attention to the power load forecasting. For the power system operations, the electric power load forecasting was a very important task. It was very important to ensure the normal operation of the power industry and the effective development of the whole national economy. Therefore, improving the accuracy of electric power load forecasting for the healthy operation of the power system is very important. At the same time, to solve and improve the problem is also one of the important and difficult tasks to scholars in current.This paper present a new combined model for electric power load forecasting. Electric power load data is a time series, and the electricity market has a strong volatility by the influence of many uncertain factors. Therefore, the electric power load data has high noise. If using the noisy data to predict directly, the prediction results will produce large errors. Therefore, this paper used empirical mode decomposition method to eliminate the noise in the raw data. This method can retain the low-frequency signal and remove the high-frequency signal of the original data. As electric power load data has obvious seasonal, the method of seasonal exponential adjustment to eliminate the seasonal effect of the de-noised data; then using Elman neural network and improved gray neural network to forecast the preprocessed load data. Finally, using particle swarm optimization algorithm to optimize the model weights, we can get the new combined forecasting model which proposed in this paper. In empirical study, the electric power load data were collected on a half-hourly which were got from New South Wales in Australia for short-term forecasting. The results show that the particle swarm optimization of grey neural network prediction accuracy was significantly higher than the prediction accuracy of gray neural network. At the same time, the combination of proposed model is better than individual model, and the de-noising model’s prediction effectiveness is better than the un-de-noising model, thus demonstrated the feasibility of combination model proposed in the electric power load forecasting.