电工技术学报2024，Vol.39Issue(19)：6119-6128,10.DOI:10.19595/j.cnki.1000-6753.tces.231267
考虑综合需求响应的Transformer-图神经网络综合能源系统多元负荷短期预测
Transformer Based Multi Load Short-Term Forecasting of Integrated Energy System Considering Integrated Demand Response
摘要
Abstract
The accurate forecasting of multi load in an integrated energy system is imperative for ensuring the efficient and secure operation of diverse energy sources.Demand response technology not only enhances the equilibrium between energy supply and demand but also induces changes in users'general energy consumption habits,thereby amplifying the complexity and uncertainty of load forecasting.While existing research explores the coupling relationships among different energy sources in integrated energy systems and employs artificial intelligence methods for predictions,there is a noticeable gap in research concerning multi load forecasting that incorporates integrated demand response.To address these issues,this paper presents a Trans-GNN prediction model that incorporates integrated demand response considerations.Through the mathematical modeling of integrated demand response signals and their incorporation as input variables into the deep learning model,the accuracy of load predictions in demand response scenarios is improved. Firstly,adhering to consumer psychology principles,we establish the power demand response center curve.By statistically calculating the fluctuation value of power demand response under various probability conditions,considering the correlation between response uncertainty and electricity price difference,we derive the power demand response signal with due consideration to uncertainty.Employing this signal and the coupling response principle,we ascertain the demand coupling response signal for cold and heat loads,culminating in the derivation of a integrated demand response signal.Then the Trans-GNN model integrates this signal,historical energy consumption data,and meteorological data for prediction.Through Transformer's attention mechanism,the model realizes the probabilistic understanding of the integrated demand response signal,and dynamically extracts and filters the historical energy consumption data.Finally,the graph neural network is used to complete the further analysis of the input data and load forecasting. In this paper,the data set of Tempe campus of Arizona State University in the United States is used as a simulation example to predict the multi load in a week,and the validity of the model is verified.The proposed model yields mean absolute percentage error(MAPE)values of 1.084％,1.186％,and 1.477％for electric load,cooling load,and heat load,respectively.Corresponding root mean square error(RMSE)values are 387.100 kW,703.008 kW,and 24.627 kW,while the weighted mean absolute percentage error(WMAPE)is 1.203％.Through the attention heat map,the way in which the model analyzes the input data is visualized,and the rationality of the model structure is verified.The error distribution of the model is statistically analyzed by the prediction error distribution,and the prediction stability of the model is further analyzed. The following conclusions can be drawn from the simulation analysis:(1)The integrated demand response signal can effectively improve the short-term prediction accuracy of the model in the context of integrated demand response.(2)The reasonable application of attention mechanism can improve model's ability to understand the importance of input information.By extracting historical data at critical time points,the model can grasp the regularity of load forecasting problems and effectively analyze the uncertainty of integrated demand response signals.(3)The application of graph neural network can make the model effectively analyze the coupling relationship between the operation mode of the integrated energy system and the multi load,and improve the prediction performance.关键词
综合能源系统/综合需求响应/耦合响应/图神经网络/Transformer模型/多元负荷短期预测Key words
Integrated energy system/integrated demand response/coupling response/graph neural network/Transformer model/multi load short-term forecasting分类
信息技术与安全科学引用本文复制引用
李云松,张智晟..考虑综合需求响应的Transformer-图神经网络综合能源系统多元负荷短期预测[J].电工技术学报,2024,39(19):6119-6128,10.基金项目
国家自然科学基金资助项目(52077108). （52077108）
