基于ARMA误差修正和自适应粒子群优化的SVM短期负荷预测

SVM short-term load forecasting based on ARMA error calibration and the adaptive particle swarm optimization

利用最小二乘支持向量机(LS-SVM)进行短期负荷预测的精度及其泛化性能很大程度上取决于其参数选择。对于支持向量机中的核参数σ和惩罚系数C采用基于适应度函数惯性权重自适应调整的粒子群优化算法进行选择。在对LS-SVM回归模型参数优化的基础上,建立自回归滑动平均(ARMA)误差预测模型来修正负荷预测结果从而提高预测精度。选择某地区夏季96点负荷数据作为训练样本和测试样本进行分析,并且选择SVM模型进行对比。实验结果表明,同标准的SVM回归模型相比,APSO-ARMA-SVM负荷预测模型能明显改善预测精度,能够推广到电价预测等其他预测领域。

To a large extent,the forecasting accuracy and generalization performance of short-term load forecasting by using least squares support vector machines（LS-SVM）depend on selection of its parameters.Adaptive particle swarm optimization（APSO）based on fitness function inertia weight is put forward to optimize the kernel parameter σ and regularization parameter C of LS-SVM.Based on the optimized regession model,ARMA error forecasting model is also presented.The error forecasted by ARMA is used to update the forecasted load so as to improve the forecasting accuracy.The load data from some area in summer are analyzed as training and forecasting samples and they are compared with the SVM model.Case analysis results show that the proposed forecasting model（APSO-ARMA-SVM）has better forecasting accuracy compared with the method of standard SVM,and it can be applied to other forecasting areas such as the electricity price forecasting.