短期负荷预测中支持向量机模型的参数选取和优化方法

讨论了基于支持向量机的电力系统负荷预测模型建模方法。对建立支持向量机模型,通过对模型结构的分析,提出了模型学习参数的选取方法,给出了基于最优方向搜索的高斯核函数参数的优化算法。针对负荷预测模型,对模型的输入变量采用规范化预处理消除了量纲不一致对SVM模型的不利影响,通过对数变换将模型输出的相对误差转化为模型输出的绝对误差,便于模型学习参数的选择。最后采用实际数据对该方法进行了模拟计算,其结果表明该方法可以有效地降低SVM模型的建模误差和测试误差,而且SVM模型比模糊模型和神经网络模型有更好的泛化性能和预测精度。

Parameter Selection and Optimization Method of SVM Model for Short-term Load Forecasting

The modeling method of electrical load forecasting model, which is based on SVM, is mainly discussed in this paper. To build SVM model, a new approach of selecting learning parameters is proposed on analysis of model structure, and an optimizing algorithm of Gauss kernel parameter is presented based on local optimal searching direction. For load forecasting model, input variables is normalized to reduce the influence of different units for SVM model, and logarithm transform is applied to convert relative error of model output into absolute error, in such a way it is convenient to selecting learning parameters. A actual data is employed to simulate computing, the result shows that proposed methods could reduce modeling error and testing error of SVM model, which has better generalized performance and accuracy than fuzzy model and artificial neural network model.