最小二乘小波支持向量机在电力负荷预测中的应用

Application of Least Square Wavelet Support Vector Machines to Electric Load Forecasting

针对中期电力负荷预测问题,提出了一种基于多维允许小波核的最小二乘小波支持向量机(least squares wavelet support vector machines,LS-WSVM)方法,并且给出了一种可有效求解LS-WSVM的Cholesky分解算法.该方法结合小波技术和最小二乘支持向量机,其中小波核函数具有近似正交以及适用于局部信号分析的特性.将LS-WSVM应用于电力负荷预测的两个实例中,结果表明,与LS-SVM、标准SVM、多层前向神经网络等方法相比,LS-WSVM均能给出相当好的预测性能,所提出的用于中期电力负荷预测的LS-WSVM方法显示了其有效性和应用潜能.

In allusion to the mid-term electric load forecasting,a form of least square wavelet support vector machines （LS-WSVM） using multi-dimensional admissible wavelet kernel is pro- posed. Simultaneously,an efficient implementation algorithm via Cholesky factorisation for LS- WSVM is also presented,which combines the wavelet techniques with LS-SVM,and the wavelet kernel is characterized by its local analysis and approximate orthogonality. LS-WSVM is then applied to the mid-term electric load forecasting of two regions. Experiment results show that the method based on LS-WSVM gives considerably good performance in comparison to LS-SVM, SVM,neural networks and etc. The proposed LS-WSVM method for the electric load forecasting shows its effectiveness and applicability.