发电技术 ›› 2019, Vol. 40 ›› Issue (6): 521-526.DOI: 10.12096/j.2096-4528.pgt.19010

• 能源互联网 • 上一篇    下一篇

基于KPCA-SVM模型的电力负荷最大值短期预测方法

张永伟1(),潘巧波2   

  1. 1 安徽邮电职业技术学院通信工程系, 安徽省 合肥市 230000
    2 华电电力科学研究院有限公司, 浙江省 杭州市 310030
  • 收稿日期:2019-01-16 出版日期:2019-12-30 发布日期:2019-12-31
  • 作者简介:张永伟(1991),男,硕士,讲师,研究方向为电力系统自动化、电力负荷预测等, zyw13844244150@163.com|潘巧波(1993),男,工学硕士,工程师,研究方向为微电网控制、电力电子与新能源发电
  • 基金资助:
    国家科技支撑计划项目(2015BAA06B02)

Short-term Prediction Method of Maximum Power Load Based on KPCA-SVM Model

Yongwei ZHANG1(),Qiaobo PAN2   

  1. 1 Telecommunication Engineering Department, Anhui Post and Telecommunication College, Hefei 230000, Anhui Province, China
    2 Huadian Electric Power Research Institute Co., Ltd., Hangzhou 310030, Zhejiang Province, China
  • Received:2019-01-16 Published:2019-12-30 Online:2019-12-31
  • Supported by:
    National Science and Technology Support Program of China(2015BAA06B02)

摘要:

电力负荷最大值预测是电网企业调度工作的重要组成部分,其预测结果的准确度将对电能的配送、有效利用率、供电服务的质量以及电力系统的发展产生重要影响。以安徽某市81天的电力负荷最大值数据为基础,选取影响当天电力负荷最大值的10个因素,并采用核主成分分析(kernel principal component analysis,KPCA)算法将10维的影响因素降为5维,其累计贡献率可达93.70%。以降维后的5维数据为输入,以径向基函数为核函数,并采用交叉验证选择支持向量机(support vector machine,SVM)回归的最佳参数,随机选取54组数据训练SVM预测模型,最后进行27组数据的拟合预测,拟合预测的均方误差为0.004 1,相关系数为0.963 1。研究结果表明,应用KPCA结合的SVM预测模型对电力负荷最大值具有很好的预测能力。

关键词: 电力系统, 负荷, 核主成分分析(KPCA), 支持向量机(SVM), 预测模型

Abstract:

The maximum power load is an important part of power enterprise dispatching work. The accuracy of its prediction results will have an important impact on the distribution of power, the effective utilization rate of power, the quality of power supply service and the development of power system. Based on the 81-day maximum power load data of a city in Anhui province, 10 factors influencing the maximum power load of that day were selected, and the influencing factors reduced from 10 dimensions to 5 dimensions by kernel principal component analysis(KPCA). The cumulative contribution rate can reach 93.70%. The best parameters of SVM regression were selected by cross validation. 54 groups of data were randomly selected to train SVM prediction model. Finally, 27 groups of data were fitted and predicted. The mean square error of fitting prediction was 0.004 1, and the correlation coefficient was 0.963 1. The results show that the KPCA combining the SVM prediction model for maximum power load has good prediction ability.

Key words: power system, load, kernel principal component analysis (KPCA), support vector machine(SVM), prediction model