基于遗传算法优化参数的支持向量机短期负荷预测方法

Parameter selection for support vectormachines based on genetic algorithms to short-term power load forecasting

通过研究参数选择和支持向量机预测能力的影响,建立利用遗传算法优化参数的支持向量机负荷预测系统。通过遗传算法对支持向量机(SVM)预测模型的各项参数进行寻优预处理,找到最优的参数取值,然后,代入支持向量机SVM预测模型中,得基于遗传算法的支持向量机(GA-SVM)模型,利用此模型对短期电力负荷进行预测研究。通过实例验证,选择河北某地区2005-03-02至2007-05-22每天各个时点的数据进行分析,并且选择SVM模型与BP(Back propagation)神经网络进行对比。研究结果表明:用GA-SVM算法得到的均方根相对误差仅为2.25%,比用SVM模型和BP神经网络所得的均方根相对误差比分别低0.58%和1.93%。所提出的测试方法克服了传统参数选择方法存在的缺点(如研究者往往凭经验和有限的实验给定一组参数,而不讨论参数制定的合理性),提高了支持向量机的预测精度。

The impact of different parameters of support vector machine （SVM） model on the power load forecasting was studied, and a short-term power load forecasting system was studied with SVM model whose parameters were optimized by genetic algorithms. The parameters for SVM model were pretreated through genetic algorithms to get the optimum parameter values, and these parameter values were used in the SVM model and genetic algorithm-support vector machine （GA-SVM） model was obtained, which was used to make short power load forecasting. Every hour＇s load from 2005-03-02 to 2007-05-22 of an area in Heibei province was taken as the sample data to be analysed. The results show that the root-mean-square relative error of GA-SVM model is only 2.25%, which is less than those of SVM and back propagation （BP） model by 0.58% and 1.93%, respectively. The new model overcomes the shortcomings of parameter selection with traditional methods, for example, researchers often give a set of parameters by virtue of experience or a limited experiments, and don＇t discuss the rationality, and the new model improves the forecasting accuracy.