考虑影响因素的短期负荷预测核函数ELM方法  

Kernel function ELM method for short-term load forecasting considering influencing factors

作　　者：张宁[1] 刘天键[1] ZHANG Ning;LIU Tianjian(Department of Physics and Electronic Information Engineering,Minjiang University,Fuzhou 350108,China)

机构地区：[1]闽江学院物理学与电子信息工程系

出　　处：《武汉大学学报（工学版）》2018年第8期703-707,714,共6页Engineering Journal of Wuhan University

基　　金：福建省中青年教师教育科研项目(编号:JAT170437,JAT160390)

摘　　要：结合结构风险最小化原则,使用核函数映射代替基本极限学习机（ELM）模型中的隐层节点特征映射,在考虑温度、相对湿度、日期类型、历史负荷等影响因素情况下提出了基于核函数极限学习机模型的短期负荷预测新方法.该方法具有较强的泛化能力,并能避免基本ELM模型可能产生的过学习现象.对实际负荷数据进行预测分析,其研究结果表明核函数极限学习机模型的预测精度要优于基本ELM模型、最小二乘支持向量机模型以及BP神经网络模型.同时也验证了核函数极限学习机方法用于短期负荷预测中的可行性和有效性.According to the principle of structural risk minimization,using kernel function mapping instead of the hidden layer feature mapping in basic extreme learning machine（ELM）model,considering influencing factors of temperature,relative humidity,date type and historical load,a new method for short-term load forecasting based on kernel function extreme learning machine model is proposed.The method proposed has strong generalization ability and can avoid the over learning phenomenon of basic ELM model.Through the prediction analysis of actual load data,the research results show that the prediction accuracy of kernel function extreme learning machine model is better than the basic ELM model,least squares support vector machine model and back propagation（BP）neural network model.Meanwhile,the feasibility and effectiveness of kernel function extreme learning machine method for short-term load forecasting are verified.

关 键 词：结构风险最小化原则 极限学习机 负荷影响因素 核函数极限学习机 最小二乘支持向量机模型 

分 类 号：TM715[电气工程—电力系统及自动化]