Short-term load forecasting based on weighted least squares support vector machine within the Bayesian evidence framework

• Published in: Dianli Xitong Baohu yu Kongzhi Vol. 39; no. 7; pp. 44 - 49
• Main Authors: Wang, Lin-Chuan, Bai, Bo, Yu, Feng-Zhen, Yuan, Ming-Zhe
• Format: Journal Article
• Language: Chinese
• Published: 01.04.2011
• Subjects: Bayesian analysis; Forecasting; Least squares method; Mathematical analysis; Mathematical models; Nuclear power generation; Support vector machines; Vectors (mathematics)
• ISSN: 1674-3415

A short-term load forecasting model and algorithm based on the weighted least squares support vector machine within the Bayesian evidence framework is proposed. On the basis of pre-processing of historical data, the author analyzes the important factors of affecting the load change, and then selects the best input data as the input vector of LS-SVM training model. The optimal parameters of models can be found through three-layer Bayesian evidence inference: The weight vector w and bias value b of LS-SVM can be determined in the first layer, and the hyper-parameter of the model can be inferred in the second layer, the hyper-parameter of the nuclear function finally can be determined in the third layer. To improve the robustness of the model, WLS-SVM regression model with good generalization performance is established by giving a different weight coefficient to each sample error, which further improves the prediction accuracy of the model. Applying the proposed method to short-term load of Heilongjiang power sy