欢迎访问《太阳能学报》官方网站,今天是 分享到:
ISSN 0254-0096 CN 11-2082/K

太阳能学报 ›› 2023, Vol. 44 ›› Issue (2): 381-390.DOI: 10.19912/j.0254-0096.tynxb.2021-1042

• • 上一篇    下一篇

基于信息熵变权区间组合和边界逼近的短期光伏功率区间预测

杨国清1,2, 李建基1, 王德意1,2, 张凯1, 刘菁1   

  1. 1.西安理工大学电气工程学院,西安 710048;
    2.西安市智慧能源重点实验室(西安理工大学),西安 710048
  • 收稿日期:2021-08-31 出版日期:2023-02-28 发布日期:2023-08-28
  • 通讯作者: 杨国清(1979—),男,博士、副教授,主要从事电力系统新能源消纳、高电压新技术方面的研究。yanggq@xaut.edu.cn
  • 基金资助:
    国家自然科学基金(51507134); 陕西省重点研发计划(2018ZDXM-GY-169); 西安市科技创新平台(201805057ZD8CG41)

SHORT-TERM PHOTOVOLTAIC POWER INTERVAL REDICTION BASED ON INFORMATION ENTROPY VARIABLE WEIGHT INTERVAL COMBINATION AND BOUNDARY APPROXIMATION

Yang Guoqing1,2, Li Jianji1, Wang Deyi1,2, Zhang Kai1, Liu Jing1   

  1. 1. College of Electrical Engineering, Xi'an University of Technology, Xi'an 710048, China;
    2. Xi'an Key Laboratory of Smart Energy, Xi'an University of Technology, Xi'an 710048, China
  • Received:2021-08-31 Online:2023-02-28 Published:2023-08-28

摘要: 针对现有的区间预测在满足高覆盖率的同时区间宽度存在过宽的问题,提出一种基于信息熵变权区间组合和边界逼近的短期光伏功率区间预测方法。首先,对历史天气数据特征进行特征重组,并基于套索交叉的递归特征消除(LassoCV-RFE)算法对重组后的特征进行筛选。然后,采用动态贝叶斯网络模型和基于卷积长短期记忆网络的改进分位数回归模型(CNN-LSTM-QH)分别预测光伏出力的置信区间,根据信息熵进行区间变权组合。最后,结合区间覆盖率和区间宽度指标,构建边界逼近函数和惩罚边界,对两个预测结果加权组合后的区间进行边界逼近。仿真结果表明:相比于一般的单一模型方法,所提方法能在95%、90%和85%的置信水平下分别减小21.86%、16.67%和14.93%的平均区间宽度,同时区间覆盖率也能满足对应的置信度要求。

关键词: 光伏功率, 特征选取, 自适应权重, 组合预测, 边界逼近, 区间预测

Abstract: Aiming at the problem that the interval width is too wide while the existing interval prediction satisfies the high coverage rate, a short-term photovoltaic power interval prediction method was proposed based on the interval combination of information entropy variable weight and boundary approximation. Firstly, the features of historical weather data were reconstructed, and the reconstructed features were screened based on LASSOCV-RFE algorithm. Then, dynamic Bayesian network model and improved quantile regression model based on convolutional long and short-term memory network (CNN-LSTM-QH) were used to predict the confidence interval of photovoltaic output, and the interval variable weight combination was carried out according to the information entropy. Finally, combining with the interval coverage and interval width indexes, the boundary approximation function and penalty boundary were constructed, and the weighted combination of the two prediction results was used to approximate the boundary of the interval. Simulation results show that the proposed method can reduce the average interval widths of 21.86%, 16.67% and 14.93% respectively at 95%, 90% and 85% confidence levels, and the interval coverage also meets the corresponding confidence level requirements.

Key words: photovoltaic power, feature selection, adaptive weight, combined prediction, boundary approximation, interval prediction

中图分类号: