基于极限学习机方法的短期负荷预测  

Extreme Learning Machine Method for Short-Term Load Forecasting

作　　者：成天乐[1] 周胜瑜[1] 李斯[2] 赵慧材[1] 黄佩[3] 蒋凌[1] 

机构地区：[1]长沙理工大学智能电网运行与控制湖南省重点实验室,湖南长沙410004 [2]中机国际工程设计研究院有限责任公司,湖南长沙410021 [3]邵阳学院,湖南邵阳422000

出　　处：《电力科学与工程》2013年第4期24-29,共6页Electric Power Science and Engineering

基　　金：国家自然科学基金资助项目(51277016);湖南省高校创新平台开放基金项目(12K001);湖南省研究生科研创新项目立项(CX2011B359)

摘　　要：将极限学习机(ELM)方法引入电力系统短期负荷预测领域。该方法预测能力强,具有计算时间短、计算准确性高、全局搜索等显著特点。在运用ELM算法建立短期负荷预测模型过程中,采用归一化处理输入数据,使用主成分分析法选取计算样本,并由交互验证法确定最优主成分因子数和ELM隐含层节点数。实际算例表明,在于短期负荷预测的预测精度和运算时间方面,ELM方法较传统神经网络方法具有其独特的优势。The paper presents a new algorithm for short term load forecasting based on the extreme learning ma-chine （ELM） method. The advantages of this method include short calculation time, high forecasting accuracy, global optima property. The extreme learning machine is employed to calibrate model for short term load forecas- ting. In the model, the principal component analysis is used to select calculation sample, and then the optimal number of PCs and the optimal number of hidden-layer nodes of ELM network are determined by the cross valida-tion. The practical examples for short term load forecasting of power system shows that the extreme learning ma- chine method outperforms the traditional networks on both the forecasting accuracy and the computing speed.

关 键 词：短期负荷预测 极限学习机 主成分分析法 交互验证法 

分 类 号：TM714[电气工程—电力系统及自动化]