摘要
负荷预测是综合能源系统(integrated energy system,IES)高效运行的前提,面对综合能源系统多元负荷强耦合相关性、强随机性的特点,单一模型在运行负荷特征提取方面存在不足。为充分利用负荷间的相关性、降低负荷数据的非平稳性、弥补单一模型的不足,提出一种基于TCN-TPABiLSTM组合模型和多任务学习框架的IES多元负荷超短期协同预测方法。首先对负荷间耦合相关性、负荷时间相关性和负荷影响因素进行分析以构建模型输入,再通过变分模态分解将负荷数据分解为一定数量的模态以降低非平稳性,最后以TCN-TPA-BiLSTM组合模型作为多任务学习框架的共享层进行预测。通过实际数据进行验证和对比,结果表明该方法能够充分发挥模型各部分优势,相较于其他模型也获得了更优的结果。
Load forecasting is a prerequisite for the efficient operation of integrated energy system(IES),and in the face of the strong coupling correlation and stochasticity of multiple loads in IES,a single model is insufficient in the feature extraction of operating loads.In order to fully utilize the correlation between loads,reduce the non-stationarity of load data,and make up for the deficiencies of a single model,a collaborative ultra-short-term forecasting method based on the TCN-TPA-BiLSTM combined model and multi-task learning framework is proposed for multivariate loads in IES.Firstly,the coupling correlation among loads,load temporal correlation and load influencing factors are analyzed to construct the model inputs,and then the load data are decomposed into a certain number of modes by variational mode decomposition(VMD)to reduce the non-stationarity,and finally,the combined TCN-TPA-BiLSTM model is used as the shared layer of the multi-task learning framework for prediction.Validation and comparison with real data show that this method can fully utilize the advantages of each part of the model,and superior results are obtained compared with other models.
作者
朱丽
侯靖轩
李子睿
ZHU Li;HOU Jingxuan;LI Zirui(School of Architecture,Tianjin University,Nankai District,Tianjin 300072,China;National Industry-Education Platform for Energy Storage,Tianjin University,Jinnan District,Tianjin 300354,China;APEC Sustainable Energy Center,Nankai District,Tianjin 300072,China)
出处
《全球能源互联网》
北大核心
2025年第5期662-674,共13页
Journal of Global Energy Interconnection
关键词
综合能源系统
多元负荷预测
组合预测模型
多任务学习
变分模态分解
integrated energy system
multivariate load forecasting
combined forecasting model
multi-task learning
variational mode decomposition
