SOM特征提取与ELM在电力负荷预测中的应用  

Application of SOM Feature Extraction and ELM in Power Load Forecasting

作　　者：马立新[1] 尹晶晶[1] 郑晓栋 

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093

出　　处：《电力科学与工程》2015年第5期1-5,共5页Electric Power Science and Engineering

基　　金：国家自然科学基金资助项目(61205076);上海市张江国家自主创新重点项目(201310-PI-B2-008)

摘　　要：随着电力负荷内涵复杂度和非线性增加,单纯追求电力负荷预测精度将变得困难。研究根据负荷样本分析其趋势、抽取特征来解决预测精度问题,即提出一种基于自组织特征映射网络(SOM)进行特征提取并与极限学习机(ELM)相结合的短期电力负荷预测方法。通过SOM特征提取找出与预测日同类型的历史数据作为训练样本;然后采用ELM进行预测,该方法预测过程简捷,能得到唯一的最优解。实验以某市的电力负荷数据进行仿真和比较。结果表明,基于SOM特征提取的ELM方法不仅精简了训练样本数量,且使训练更具有针对性,提高了预测精度和泛化性能,具有一定的理论意义和较好的应用前景。With the increase of power load connotation complexity and nonlinearity, the pure pursuit of power load forecasting accuracy becomes more difficult. This paper aimed to improve prediction accuracy in line with the anal- ysis of load sample trend and feature extraction and proposed a short-term load forecasting method based on the combination of self-organizing feature mapping network （SOM） feature extraction and extreme learning machine （ ELM）. First, the same type data as that on the forecasting day were selected as the training sample by using the feature extraction of SOM algorithm. Then, ELM was used for prediction, since the forecasting process of ELM is simple and can generate a unique optimal solution. The power load data of one city were used for simulating and comparing. The experimental results showed that ELM method based on SOM feature extraction downsized the num- ber of training samples, made the training more targeted, and improved forecasting accuracy and generalization per- formance. This method has certain theoretical significance and good application prospects.

关 键 词：自组织特征映射 特征提取 极限学习机 短期负荷预测 

分 类 号：TM715[电气工程—电力系统及自动化]