电力科技与环保2025，Vol.41Issue(2)：323-332,10.DOI:10.19944/j.eptep.1674-8069.2025.02.015
基于空间网络和时间序列的发电量预测优化研究
Optimization study of power generation forecasting based on spatial network and time series
摘要
Abstract
[Objective]In the power system,effective management and scheduling rely heavily on accurate forecasting of future load demand,and accurate forecasting of power generation on the supply side of the power plant,determined by the power load on the demand side of society,can provide important support.[Methods]In this paper,the daily power generation data of 17 power plants in Jiangxi Province from October 2020 to September 2021 and the spatial latitude and longitude coordinates of the power plants are used to construct L-GAT and G-LSTM prediction models by combining graph attention network(GAT)and long short-term memory network(LSTM)in different orders for integrating spatialnetwork features and time series features for power system power generation prediction.The L-GAT and G-LSTM prediction models,which combine the graph attention network(GAT)and long short-term memory network(LSTM)in different orders,are used to synthesize spatial network features and time series features for power system generation prediction.[Results]The results show that the L-GAT model has the best prediction performance,with the average absolute error,root mean square error and average percentage error reduced by 17.47%,6.17%and 12.79%,respectively,compared with the LSTM model.The enhancement of 106.48%,66.52%and 1205.68%compared with the GAT model.The average absolute error,root mean square error and average percentage error of the L-GAT model arelower than those of the traditional model,the autoregressive model,and the integrated sliding average model.integralsliding average model,the average absolute error,root mean square error and average percentage error are reduced by 63.28%,59.18%and 592.30%,respectively.The temporal feature is the core of prediction,and the introduction of spatial feature can further optimize the accuracy.While G-LSTM results in the performance degradation due to the spatial error transfer,which verifies the importance of the processing order.[Conclusion]This study has certain reference value in the application of forecasting methods.It has certain practical significance for improving the digitalmanagement level of power system.关键词
电力系统发电量预测/图注意力网络/长短期记忆网络/空间网络特征/时间序列特征Key words
power generation prediction/graph attention network/long short-term memory network/spatial network features/time series features分类
能源与动力引用本文复制引用
胡睿,刘典,张文波,陈植元..基于空间网络和时间序列的发电量预测优化研究[J].电力科技与环保,2025,41(2):323-332,10.基金项目
国家自然科学基金项目(72471180) （72471180）
