山东电力技术2026，Vol.53Issue(6)：102-116,15.DOI:10.20097/j.cnki.issn1007-9904.250241
基于多任务学习的智能建筑气温-负荷-价格联合预测
Joint Prediction of Temperature,Load,and Price in Smart Buildings Based on Multi-task Learning
摘要
Abstract
The joint prediction of temperature,load,and price is crucial for demand response and energy optimization in smart buildings.Due to environmental changes,fluctuations in energy demand,and uncertainties in pricing,single-task prediction methods struggle to fully capture the complex interdependencies among these factors,thereby limiting the accuracy and reliability of predictions.To address this issue,this paper proposes a multi-task joint prediction method for temperature,load,and price.First,the complex interrelationships among three uncertain variables—outdoor temperature,electrical load,and electricity price—are studied and verified.Then,a temporal convolutional network-weighted concatenation-bidirectional gated recurrent unit(TCN-WC-BiGRU)model is designed as a shared weight layer to extract and fuse multi-scale features.Finally,a pyramidal convolutional neural network-channel attention mechanism-bidirectional gated recurrent unit(PyCNN-CAM-BiGRU)model is constructed as a task-specific layer to achieve redundant noise suppression and deep information extraction.Experimental results show that,compared to various deep neural network models,the proposed method achieves a maximum reduction in mean absolute percentage error(MAPE)of 60.78%,49.96%,and 50.9%for temperature,load,and price prediction,respectively,as well as a maximum reduction in root mean squared error(RMSE)of 63.65%,47.98%,and 49.75%.关键词
多任务学习/时序卷积网络/双向循环门控单元/金字塔卷积网络/通道注意力机制Key words
multi-task learning/temporal convolutional network/bidirectional gated recurrent unit/pyramidal convolutional neural network/channel attention mechanism分类
信息技术与安全科学引用本文复制引用
冯媛媛,梁小姣,李鹏飞,于文嫣,耿浩文..基于多任务学习的智能建筑气温-负荷-价格联合预测[J].山东电力技术,2026,53(6):102-116,15.基金项目
国网山东省电力公司科技项目"适应新型电力系统的零碳建筑柔性互动与综合能效提升关键技术研究及应用"(520616240002).Science and Technology Project of State Grid Shandong Electric Power Company"Research and Application of Key Technologies for Flexible Interaction and Comprehensive Energy Efficiency Improvement of Zero Carbon Buildings Adapted to New Power Systems"(520616240002). （520616240002）
