Scholarlink

With the widespread adoption of distributed photovoltaic (PV) generation equipment on the user side, higher demands have been placed on the forecasting and regulation technology of distributed PV output. Considering the strong spatiotemporal correlations of distributed PV clusters, this paper proposes a short-term power forecasting method for distributed PV clusters based on a combination optimization strategy and Temporal Graph Neural Network (TGNN). This method effectively leverages spatiotemporal feature information to improve power forecasting accuracy. Firstly, the temporal and spatial correlations inherent in PV output data are analyzed. An adjacency matrix encompassing all sites is constructed based on the correlation coefficients, which characterizes the spatiotemporal correlation properties among distributed PV sites. Then, by integrating distributed PV meteorological data and historical PV output data, a power forecasting model based on TGNN is constructed. On this basis, three hyperparameter optimization algorithms—random search (RS), Tree-structured Parzen Estimator (TPE), and Covariance Matrix Adaptation Evolution Strategy (CMA-ES)—are combined for stepwise optimization of the power forecasting model. This approach effectively exploits the spatiotemporal correlation characteristics among the sites. The case studies indicate that the proposed forecasting method can significantly enhance the accuracy of distributed PV cluster power forecasting.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Short-Term Power Forecasting for Distributed Photovoltaic Clusters Based on a Combination Optimization Strategy and TGNN

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Jian Peng,
  • Jiajia Huang,
  • Shengjun Luo,
  • Xiaoqiang Huang,
  • Zhaojie Zeng,
  • Fengchu Liu,
  • Yu Lai,
  • Hanwen Wu,
  • Cheng Chen,
  • Zhe Lin

摘要

With the widespread adoption of distributed photovoltaic (PV) generation equipment on the user side, higher demands have been placed on the forecasting and regulation technology of distributed PV output. Considering the strong spatiotemporal correlations of distributed PV clusters, this paper proposes a short-term power forecasting method for distributed PV clusters based on a combination optimization strategy and Temporal Graph Neural Network (TGNN). This method effectively leverages spatiotemporal feature information to improve power forecasting accuracy. Firstly, the temporal and spatial correlations inherent in PV output data are analyzed. An adjacency matrix encompassing all sites is constructed based on the correlation coefficients, which characterizes the spatiotemporal correlation properties among distributed PV sites. Then, by integrating distributed PV meteorological data and historical PV output data, a power forecasting model based on TGNN is constructed. On this basis, three hyperparameter optimization algorithms—random search (RS), Tree-structured Parzen Estimator (TPE), and Covariance Matrix Adaptation Evolution Strategy (CMA-ES)—are combined for stepwise optimization of the power forecasting model. This approach effectively exploits the spatiotemporal correlation characteristics among the sites. The case studies indicate that the proposed forecasting method can significantly enhance the accuracy of distributed PV cluster power forecasting.