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 ›› 2022, Vol. 43 ›› Issue (8): 58-66.DOI: 10.19912/j.0254-0096.tynxb.2020-1405

• • 上一篇    下一篇

基于DTW-VMD-PSO-BP的光伏发电功率短期预测方法

袁建华1, 谢斌斌1, 何宝林1, 赵子玮1, 刘宇1, 刘邦2   

  1. 1.三峡大学电气与新能源学院,宜昌 443002;
    2.国网安徽省电力有限公司亳州供电公司,亳州 236000
  • 收稿日期:2020-12-27 出版日期:2022-08-28 发布日期:2023-02-28
  • 通讯作者: 袁建华(1978—),男,博士、副教授,主要从事光电子技术应用方面的研究。sdyjh@ctgu.edu.cn
  • 基金资助:
    煤燃烧国家重点实验室开放基金(FSKLCCA1607); 梯级水电站运行与控制湖北省重点实验室基金(2015KJX07); 产学研协同培养研究生实践创新能力机制研究项目(No.SDYJ201604)

SHORT TERM FORECASTING METHOD OF PHOTOVOLTAICOUTPUT BASED ON DTW-VMD-PSO-BP

Yuan Jianhua1, Xie Binbin1, He Baolin1, Zhao Ziwei1, Liu Yu1, Liu Bang2   

  1. 1. College of Electrical and New Energy, China Three Gorges University, Yichang 443002, China;
    2. Bozhou Power Supply Company, State Grid Anhui Electric Power Co., Ltd., Bozhou 236000, China
  • Received:2020-12-27 Online:2022-08-28 Published:2023-02-28

摘要: 针对光伏发电系统短期预测影响因素较多、预测精度较低、稳定度不高等问题,提出一种基于动态时间弯曲(DTW)和变分模态分解(VMD)的粒子群(PSO)优化的BP神经网络光伏发电预测方法。首先使用动态时间弯曲算法对光伏发电功率及影响因素的数据进行测算得到DTW值,再根据DTW值选择对光伏发电功率影响较大的辐射度作为主要影响因素,然后利用变分模态分解将影响因素及光伏发电功率进行分解,降低数据的波动性和非平稳性。运用粒子群优化的BP神经网络对各分量进行预测,然后将预测结果进行叠加,叠加所得结果即为最后预测结果。在Matlab中对该方法和其他神经网络进行算例验证和误差分析,结果表明采用该方法预测结果精度高,稳定性好。

关键词: 光伏发电功率, 粒子群算法, 神经网络, 短期预测, 动态时间弯曲, 变分模态分解

Abstract: Aiming at the problems of many influencing factors, low prediction accuracy and low stability in short-term prediction of photovoltaic power generation system, a BP neural network photovoltaic power generation prediction method based on particle swarm optimization (PSO) optimization of dynamic time warping (DTW) and variational mode decomposition (VMD) is proposed. Firstly, the dynamic time warping algorithm is used to calculate the data of photovoltaic power generation and influencing factors, to obtain DTW value. Then, according to the DTW value, the radiance which has a greater impact on photovoltaic power generation is selected as the main influencing factor. Next, the variational mode decomposition is used to decompose the influencing factors and photovoltaic power to reduce the volatility and non stationarity of the data, and BP neural network optimized by the particle swarm is used to predict each component Finally, the prediction results are superimposed, to obtain the final prediction value This method is verified in MATLAB by comparison with other neural networks and error analysis. The results comparison show that this method has high accuracy and good stability.

Key words: PV power, particle swarm optimization, neural networks, short-term forecast, dynamic time warping, variational mode decomposition

中图分类号: