Power System Short-term Load Forecasting Based on Adaptive Weighted Least Squares Support Vector Machine

The presence of outliers in load data can corrupt a model's performance, giving undesirable results. In this study, a novel adaptive weighted least squares support vector machine (AWLS-SVM) regression method is proposed for modeling the short-term load forecasting of a power system. An improved normal distribution weighted scheme is used to determine weights for the training samples adaptively. To improve the prediction accuracy and model generalization, the particle swarm optimization genetic algorithm is utilized to obtain optimal model parameters. Furthermore, assuming the seasonal power system load of a northern city, the AWLS-SVM method is applied to test the forecasting performance of the model. The computational results demonstrate that in terms of the prediction accuracy and model generalization, the proposed AWLS-SVM approach is better than conventional LS-SVM and weighted LS-SVM models.