中长期负荷预测的异常数据辨识与缺失数据处理  

Abnormal Data Identification and Missing Data Filling in Medium-and Long-Term Load Forecasting

作　　者：毛李帆[1] 姚建刚[1] 金永顺 李文杰 关石磊 陈芳 

机构地区：[1]湖南大学电气与信息工程学院,湖南省长沙市410082 [2]湖南湖大华龙电气与信息技术有限公司,湖南省长沙市410082

出　　处：《电网技术》2010年第7期148-153,共6页Power System Technology

摘　　要：负荷历史数据是进行中长期负荷预测的基础。历史数据异常及缺失将严重影响负荷预测模型的精度及有效性。针对传统异常数据辨识方法和缺失数据填补方法的不足,提出了基于T2椭圆图的异常数据识别和基于最小二乘支持向量机(least square support vector machine,LSSVM)的缺失数据填补方法。采用偏最小二乘法(partial least square,PLS)提取历史数据主成份,计算各历史样本对主成份的累积贡献率(accumulative contribution rate,ACR),并绘制T2椭圆,从而识别出历史样本贡献率过大的异常数据。用最小二乘支持向量机拟合历史数据变化趋势,从而实现缺失数据的填补。算例结果表明:T2椭圆图能有效识别历史数据中的异常样本;最小二乘支持向量机具有良好的数据填补特性,具有较强的实用价值。Historical load data is the basis of medium- and long-term load forecasting, thus the abnormal historical data and historical data missing seriously affect the accuracy and effectiveness of load forecasting. To remedy the insufficiency of traditional methods for abnormal data identification and data filling, a method for missing data filling, which is based on both T2 ellipse map to identify abnormal data and least square support vector machine （LSSVM）, is proposed. The principal component of historical data is extracted by partial least square （PLS） to compute the accumulative contribution rate （ACR） of historical data to principal component and draw T2 ellipse, thereby the abnormal historical data that possesses too high contribution rate can be identified; the variation trend of historical data is fitted by LSSVM, thus the missing data can be filled. Results of calculation example show that the T2 ellipse map can effectively identify the abnormal samples in historical data and LSSVM possesses good data filling performance, therefore the proposed method is practicable.

关 键 词：数据异常 数据缺失 累积贡献率 T^2椭圆 最小二乘支持向量机 负荷预测 

分 类 号：TM715[电气工程—电力系统及自动化]