中国电力 ›› 2020, Vol. 53 ›› Issue (5): 100-111.DOI: 10.11930/j.issn.1004-9649.201809133

• 新能源 • 上一篇    下一篇

宁波地区基于统计升尺度的新能源区域功率预测算法

王威1, 王波1, 张俊2, 陆春良2, 贺旭1   

  1. 1. 国网浙江省电力有限公司宁波供电公司,浙江 宁波 315000;
    2. 国网浙江省电力有限公司,浙江 杭州 310007
  • 收稿日期:2018-09-30 修回日期:2019-09-28 发布日期:2020-05-05
  • 作者简介:王威(1979-),男,硕士,高级工程师,从事电力技术、市场研究,E-mail:nbdyjww@163.com;王波(1985-),男,硕士,高级工程师,从事电力系统分析研究,E-mail:247622018@qq.com;张俊(1981-),男,博士,高级工程师,从事电力系统分析研究,E-mail:zhang-jun@zj.sgcc.com.cn;陆春良(1967-),男,高级工程师,从事电力技术经济、电力负荷预测研究,E-mail:13858031590@139.com;贺旭(1990-),男,硕士,工程师,从事电力系统分析研究,E-mail:810492523@qq.com
  • 基金资助:
    国家电网有限公司科技项目(基于微气象大数据的新能源发电预测及优化调度研究与应用,5211NB160007)

New Energy Regional Power Prediction Algorithm Based on Statistical Upscaling in Ningbo Region

WANG Wei1, WANG Bo1, ZHANG Jun2, LU Chunliang2, HE Xu1   

  1. 1. State Grid Ningbo Power Supply Company, Ningbo 315000, China;
    2. State Grid Zhejiang Electric Power Co., Ltd., Hangzhou 310007, China
  • Received:2018-09-30 Revised:2019-09-28 Published:2020-05-05
  • Supported by:
    This work is supported by the Science and Technology Project of SGCC (Research and Application of Renewable Energy Generation Forecast and Optimal Scheduling Based on Micro Meteorological Big Data,No.5211NB160007)

摘要: 为实现电网调度对大规模风电场和光伏站的监控和调度,需要进行新能源区域功率预测。提出了一种基于子区域划分和统计升尺度的区域功率预测算法,通过浙江区域9个风电场和16个光伏站2016年4—9月的历史数据,对比了6种不同的组合方案,发现利用互信息理论,基于最大相关–最小冗余原则累加选取4个代表站点或者直接选取9个代表站点,采用布谷鸟搜索算法训练得到各个代表站点的权重,升尺度得到区域功率预测误差较低,月均方根误差(RMSE)分别为8.51%和7.64%。说明在夏季风盛行时,在浙江区域采用互信息为指标,基于最大相关–最小冗余的原则选取代表站点后,再采用布谷鸟搜索算法得到子区域功率预测值,累加各子区域功率预测结果为最终的区域功率预测结果最优。

关键词: 新能源, 区域功率预测, 互信息理论, 最大相关-最小冗余, 布谷鸟搜索算法

Abstract: In order to monitor and dispatch the large-scale wind farms and photovoltaic stations for power grid, it is needed to predict the regional power of new energy. A regional power prediction algorithm is proposed based on sub-region partition and statistical upscaling. According to the historical data of 9 wind farms and 16 photovoltaic stations in Zhejiang province from April to September 2016, six different combination schemes are compared. The mutual information theory (MI) and the minimal redundancy maximal relevance principle (mRMR) are used to accumulatively (mRMR-A) select four representative stations or directly (mRMR-D) select nine representative stations. The weight of each representative site is trained by cuckoo search algorithm, and the regional power is determined by scaling up. It is found that the MI-mRMR-A-CS-4 and MI-mRMR-D-CS-9 algorithms have low prediction errors with their monthly root mean square error being 8.51% and 7.64% respectively. It is concluded that when the summer monsoon is prevalent in Zhejiang region, the representative stations are selected based on mRMR principle with MI used as the index, and then the cuckoo search algorithm is used to obtain the power prediction values of sub-regions, and the cumulative power prediction results of all sub-regions are the best for the final regional power prediction results.

Key words: new energy, regional power prediction, mutual information theory, minimal redundancy maximal relevance, cuckoo search algorithm