Short-term Load Forecasting Based On Wavelet Transform And Least Squares Support Vector Machine

Power system as the basic facilities of national life plays a very crucial role in economic and social development of the country. High accuracy electric load forecast data obtained through short-term power load forecasting technology provide an effective data for the dispatch of the power system, installation of generators and network planning. It can also ensure stable operation of the entire grid system.But the electric load data are vulnerable to changes of the weather, holidays, social events and other factors, resulting in noise data. If the data contain noise direct modeling to predict that will affect the accuracy of the results. The noise must be effectively filtered out to improve forecasting accuracy. For this, the paper combined with wavelet transform, particle swarm optimization and least squares support vector machine (LSSVM) method and propose a hybrid prediction method, which named WPLSSVM to forecast short-term power load data. The main idea is to decompose the original sequence by using the wavelet transform. Then predict the decomposition characteristics of enriched sequence. Finally the predicted value is got by wavelet reconstruction.Firstly, the wavelet decomposition is used to decompose the original sequence to remove high frequency noise in the signal sequence. Signal components with the same characteristics will also be divided into a group and the local characteristic information of the signal can be ensured. Then the signal components with the same characteristics model using particle swarm optimization LSSVM. The parameter selections of LSSVM have great impact on the performance of learning machine. The LSSVM optimization parameters can be obtained by the Particle Swarm Optimization algorithm which has global search capability through automatic optimization of model parameters to ensure the validity of predictive model parameters. Finally, the predictive value of electric load data can be obtained by wavelet reconstruction.In this paper, the proposed method WPLSSVM has been applied in New South Wales (Australia) electricity market for short-term load forecasting. It can be seen from the simulation examples that WPLSSVM method predicted with high accuracy and can capture the historical data model with strong generalization ability. At the same time, the predict results of WPLSSVM were comparative analysis of multiple sets of experimental data with the prediction values of a single model and two combination predictions model. The results show that WPLSSVM prediction error is less than a single model and the combination of the two models.