基于STGCN-Transformer的短期电力净负荷预测

打开文本图片集

中图分类号：TB9;TM714 文献标志码：A 文章编号：1674-5124（2025）06-0160-10

Short-term electricity net load forecasting based on STGCN-Transformer

MENG Wei¹，YU Bin1²，BAI Long¹，XU Jie’，GU Jinhao¹，GUO Feng³

（1.Automationcollege，Nanjing Universityof Information Science & Technology，Nanjing210o44，China;2.Wuxi

College ofAutomation，Wuxi 214105，China; 3.Taizhou Power Supply， Company of State Grid Zhejiang Electric Power Co.， Taizhou 318000， China）

Abstract： The development of smart grids recognizes the importance of short-term electricity net load forecasting for integrated energy systems （IES）.Net load forecasting represents the difference between the electricity consumption load and the installed renewable energy sources，and is the basis for energy management and optimal dispatching，which is poorly predicted by traditional statistical models due to the high volatility of IES. In this paper， we propose an integrated energy system short-term load forecasting model based on Spatial Temporal Graph Convolutional Networks （STGCN） and Transformer. First， STGCN is used as an input embedding layer to encode multivariate input sequences， which fills the gap in Transformer where relevant information is not fully considered. Then， the time-dependence of the sequence data is captured using theself-attentive mechanism in Transformer.Finally，the feedforward neural network is used to output the predicted load values. In this paper， a regional electricity dataset in Zhejiang Province is used as an example， and the comparison with other four prediction models shows that the model in this paper has high prediction accuracy and stability.

Keywords： spatio-temporal graph convolutional network; Transformer; multi-headed atention mechanism; short-term net load forecasting

0 引言

随着可再生能源发电容量的增加和智能电网的快速发展，电网供电需求的不确定性增加。（剩余13533字）