基于多能需求响应与改进BiLSTM的综合能源系统负荷预测  

Integrated Energy System Load Forecasting Based on Multi-energy Demand Response and Improved BiLSTM

作　　者：张晓佳 王灿 张佳恒 王振 李智威 张赵阳 甘友春 ZHANG Xiaojia;WANG Can;ZHANG Jiaheng;WANG Zhen;LI Zhiwei;ZHANG Zhaoyang;GAN Youchun(School of Electrical Engineering and New Energy,China Three Gorges University,Yichang 443002,Hubei Province,China;State Grid Hubei Economic Research Institute,Wuhan 430077,China;Sichuan Mingxing Electric Power Co.,Ltd.,Suining 629000,Sichuan Province,China)

机构地区：[1]三峡大学电气与新能源学院,湖北省宜昌市443002 [2]国网湖北省电力有限公司经济技术研究院,武汉市430077 [3]四川明星电力股份有限公司,四川省遂宁市629000

出　　处：《电力建设》2025年第4期113-125,共13页Electric Power Construction

基　　金：国家自然科学基金项目(52107108)。

摘　　要：【目的】随着能源消费趋向多样化,多元负荷预测对于综合能源系统(integrated energy system,IES)优化调度与运行规划的重要作用日益凸显。【方法】针对目前综合能源系统负荷预测研究中往往忽略多元负荷间耦合关系的问题,提出一种基于多能需求响应与改进双向长短期记忆(bi-directional long short-term memory,BiLSTM)神经网络的综合能源系统多元负荷联合预测方法。首先,综合用户需求响应行为构建多能需求响应的输入特征变量,并与最大信息系数筛选出的多元负荷预测强相关特征共同构成预测模型的输入特征集;其次,基于混沌映射理论和精英反向学习策略对冠豪猪优化算法进行改进,以优化双向长短期记忆神经网络的模型参数;最后,基于多头自注意力机制自适应调整输入特征权重。【结果】仿真结果表明,所提多元负荷联合预测方法的预测精度相较于单一负荷预测方法有显著提升,与未考虑需求响应的多元负荷预测方法相比,电、热、冷负荷的平均绝对百分比误差分别降低了6.59%、13.04%和24.86%。此外,与其他预测模型相比,所提模型在提高预测精度方面更为有效,能够实现更为精准的多元负荷预测。【结论】同时,将所提负荷预测与综合能源系统调度结合,分析其带来的经济效益。与普通调度相比,引入所提负荷预测方法的系统总运行成本减少了16.49%,能够实现IES综合效益的提升。[Objective]With the trend of energy consumption diversification,multiload forecasting plays an increasingly important role in optimizing the scheduling and operation planning of integrated energy systems(IES).[Methods]To address the problem in which the coupling relationship between multiple loads is often ignored in current integrated energy system load forecasting research,a multiple load joint forecasting method is proposed in this study for integrated energy systems based on the multi-energy demand response and improved bidirectional long short-term memory(BiLSTM).First,by integrating user demand response behavior,the input feature variables of the multi-energy demand response is constructed,and together with multiload forecasting,strong correlation features selected by the maximum information coefficient form the input feature set of the prediction model.Second,the crested porcupine optimizer is improved based on the chaotic mapping theory and elite reverse learning strategy to optimize the model parameters of the BiLSTM neural network.Finally,based on the multihead self-attention mechanism,the input feature weight is adaptively adjusted.The simulation results show that the prediction accuracy of the proposed multiload joint forecasting method is significantly improved compared with the single-load forecasting method.[Results]Compared with the multiload forecasting method without considering the demand response,the mean absolute percentage error of the electricity,heat,and cooling loads was reduced by 6.59%,13.04%,and 24.86%,respectively.In addition,compared with other forecasting models,the model proposed in this study is more effective in improving the prediction accuracy and can achieve more accurate multi-element load forecasting.[Conclusions]The proposed load forecasting method was combined with integrated energy system dispatching to analyze the economic benefits of load forecasting.Compared with ordinary dispatching,the total operating cost of the system using the proposed load forecasting method was

关 键 词：综合能源系统(IES) 双向长短期记忆网络(BiLSTM) 多能需求响应 多头自注意力机制 多元负荷预测 

分 类 号：TM715[电气工程—电力系统及自动化]