Short-term Load Forecasting Research Based On EMD And LSSVM

Short-term Load Forecasting has important research significance and practical application value for the security and economy of regional power system operation. In this paper, a hybrid short-term load forecasting model is established based on EMD(Empirical Mode Decomposition) and LSSVM(Least square support vector machine).This paper introduces the necessity of data processing in detail and how to select the effective sample and the normalization of the sample. This paper introduces the necessity of data processing, the selection of effective samples and the normalization of the sample in detail, and analysis the characteristics of power load in Hefei. These results lay a solid foundation for the selection of the input features of the following samples. Power load data is vulnerable to climate change, major holidays and so on, resulting in a lot of noise signals in the data signal. If the noise signal is not eliminated, the accuracy of prediction will be reduced, Therefore, this paper introduces the EMD algorithm in order to eliminate the noise signal and improve the prediction accuracy. EMD algorithm can convert the input signal into several intrinsic mode function, improving the smoothness of data. This paper builds a LSSVM short-term load forecasting model based on particle swarm optimization. Selecting the appropriate parameters is very important to performance prediction of model. Using particle swarm parameters can solve the blindness of traditional experience and cross validation in parameter selection, greatly shorten the search time of training, finally, meet the expectations of the short-term forecast. Finally, by EMD decomposition algorithm for the numerical combination model of load forecasting. This paper builds a EMD-PSO-LSSVM hybrid forecasting model, and applies this model in short-term load forecasting of Hefei. Through the simulation, this paper found that EMD-PSO-LSSVM has good performance on the prediction accuracy and processing ability in historical data, the model also has excellent generalization ability. The hybrid model, single model and combined model of the two algorithms were compared; results show that the prediction error is significantly lower than the single EMD-PSO-LSSVM model and the two algorithm combination model.