一种自适应选择样本的用电负荷预测方法  

An electric load forecasting method based on adaptive selection of samples

作　　者：方芳 田世明[2] 卜凡鹏[2] 苏运 Fang Fang Tian Shiming Bu Fanpeng Su Yun(State Grid Beijing Changping Electric Power Supply Company, Beijing 102200, China China Electric Power Research Institute, Beijing 100192, China Shanghai Electric Power Company, Shanghai 200437, China)

机构地区：[1]国网北京市电力公司昌平供电公司,北京102200 [2]中国电力科学研究院,北京100192 [3]国网上海市电力公司,上海200437

出　　处：《电子技术应用》2017年第11期18-21,26,共5页Application of Electronic Technique

基　　金：国家863计划(2015AA050203);国家电网公司科技项目(52094016000A)

摘　　要：针对传统短期负荷预测中预测模型缺乏自适应性、预测影响因素复杂难于筛选的问题,提出一种结合自适应技术的岭回归预测模型。通过引入岭回归技术,能在预测中多方面考虑各种复杂因素而不会受到因素间多重共线性的影响;引入虚拟预测日,同时设置不同权重对相似历史样本进行自适应筛选并训练,能够对每一个预测日减小预测误差。算例分析表明,应用结合自适应技术的岭回归预测方法后,实际预测误差得到显著降低。Aiming at the problem that the traditional forecasting model of short-term load forecasting is lack of adaptability and the factors affecting the prediction are complex and difficult to filter, a ridge regression model combined with adaptive technology is proposed. Through ridge regression technique is introduced, many can predict the consideration of various complex factors and will not be affected by factors of multieollinearity. Introducing virtual prediction and setting different weights for adaptive training on screening and similar historical samples can reduce the prediction error. The example analysis shows that the prediction error is sig- nificantly reduced by using the ridge regression prediction method combined with adaptive technique.

关 键 词：岭回归 自适应 虚拟预测日 短期负荷预测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]