针对单一预测模型忽略跨负荷关联特征导致预测效果较差的问题,提出一种基于多任务学习的组合模型对多元负荷进行预测,以提高预测的准确性。首先,量化气象因素与负荷的相关性,确定模型的输入特征;其次,构建预测模型,采用卷积神经网络（CNN）提取输入特征与负荷之间的非线性空间关联,双向长短期记忆网络（BiLSTM）模块捕获负荷的长期依赖性和时间特征,并引入时间模式注意力（TPA）机制自适应赋权突出重要时间信息;最后,采用多任务学习提取冷、热、电负荷的公共特征,捕捉其耦合关系,在任务特定分支分别预测不同类型的负荷。通过公开数据集开展实验,结果表明,冷、热、电负荷联合预测相比单任务预测模型的平均绝对百分比误差（MAPE）分别降低36.30%、23.93%和22.63%。该方法不仅可提升负荷预测的准确性和稳定性,还能更好地反映综合能源系统中多能源耦合特性。
Abstract
Aiming to address the suboptimal performance of single prediction models, which often neglect cross-load correlation features, this study proposes a multi-task learning-based combined model for multi-load forecasting to enhance prediction accuracy. Firstly, the correlation between meteorological factors and loads was quantified to determine the model's input features. Secondly, a prediction model was constructed, utilizing a convolutional neural network （CNN） to extract nonlinear spatial correlations between input features and loads, a bidirectional long short-term memory （BiLSTM） module to capture the long-term dependencies and temporal features of the loads, and a temporal pattern attention （TPA） mechanism to adaptively weight and highlight important temporal information. Finally, multi-task learning （MTL） was employed to extract shared features among cooling, heating, and electricity loads, capturing their coupling relationships, while task-specific branches were used to predict each load type separately. Experimental results based on public datasets demonstrate that the joint forecasting of cooling, heating, and electricity loads achieved reductions in mean absolute percentage error （MAPE） of 36.30%, 23.93%, and 22.63%, respectively, compared to single-task prediction models. This method not only enhances the accuracy and stability of load forecasting but also better reflects the multi-energy coupling characteristics within integrated energy systems.
关键词
预测 /
多任务学习 /
卷积神经网络 /
时间模式注意力 /
双向长短时记忆网络 /
负荷耦合特征 /
综合能源系统
Key words
forecasting /
multi-task learning /
convolutional neural network /
temporal pattern attention /
bidirectional long short-term memory /
load coupling characteristics /
integrated energy system
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 李志颖, 王致杰, 王鸿. 考虑功率可调节裕度的区域综合能源系统多时间尺度优化调度[J]. 太阳能学报, 2025, 46(2): 255-261.
Li Z Y, Wang Z J, Wang H.Multi-timescale optimal scheduling of integrated regional energy systems considering power adjustable margins[J]. Acta Energiae Solaris Sinica, 2025, 46(2): 255-261.
[2] 骆钊, 杨林燕, 王华, 等. 含碳捕集电厂-电转气-液化天然气协同的综合能源系统优化调度[J]. 太阳能学报, 2025, 46(1): 438-448.
Luo Z, Yang L Y, Wang H, et al.Optimized scheduling of an integrated energy system with carbon capture power plant-electricity-to-gas and liquefied natural gas synergies[J]. Acta Energiae Solaris Sinica, 2025, 46(1): 438-448.
[3] Wang Y, Li H M, Jahanger A, et al.A novel ensemble electricity load forecasting system based on a decomposition-selection-optimization strategy[J]. Energy, 2024, 312: 133524.
[4] Tan Q W, Cao C H, Xue G J, et al.Short-term heating load forecasting model based on SVMD and improved informer[J]. Energy, 2024, 312: 133535.
[5] 王印松, 吕率豪. 基于改进时间卷积网络的微电网超短期负荷预测[J]. 太阳能学报, 2024, 45(6): 255-263.
Wang Y S, Lyu S H.Ultra-short-term load forecasting for microgrids based on improved temporal convolutional networks[J]. Acta Energiae Solaris Sinica, 2024, 45(6): 255-263.
[6] Kim N, Park H, Lee J, et al.Short-term electrical load forecasting with multidimensional feature extraction[J]. IEEE Transactions on Smart Grid, 2022, 13(4): 2999-3013.
[7] Lyu L L, Wu Z Y, Zhang J H, et al.A VMD and LSTM based hybrid model of load forecasting for power grid security[J]. IEEE Transactions on Industrial Informatics, 2022, 18(9): 6474-6482.
[8] Yu H F, Zhong F, Du Y J, et al.Short-term cooling and heating loads forecasting of building district energy system based on data-driven models[J]. Energy and Buildings, 2023, 298: 113513.
[9] Gong M J, Zhao Y, Sun J W, et al.Load forecasting of district heating system based on Informer[J]. Energy, 2022, 253: 124179.
[10] He D S, Zhang Y, Bin Ashab M F. Proposing hybrid prediction approaches with the integration of machine learning models and metaheuristic algorithms to forecast the cooling and heating load of buildings[J]. Energy, 2024, 291: 130297.
[11] Guo Y X, Li Y, Qiao X B, et al.BiLSTM multitask learning-based combined load forecasting considering the loads coupling relationship for multienergy system[J]. IEEE Transactions on Smart Grid, 2022, 13(5): 3481-3492.
[12] Wang S M, Wang S X, Chen H W, et al.Multi-energy load forecasting for regional integrated energy systems considering temporal dynamic and coupling characteristics[J]. Energy, 2020, 195: 116964.
[13] 张玉敏, 孙猛, 吉兴全, 等. 基于模态分解与多任务学习模型的综合能源系统多元负荷短期预测[J]. 高电压技术, 2025, 51(7): 3488-3499.
Zhang Y M, Sun M, JI X Q, et al.Short-term forecasting of multivariate loads for integrated energy systems based on modal decomposition and multi-task learning model[J]. High Voltage Engineering, 2025, 51(7): 3488-3499.
[14] Chen B Y, Wang Y F.Short-term electric load forecasting of integrated energy system considering nonlinear synergy between different loads[J]. IEEE Access, 2021, 9: 43562-43573.
[15] Zhu J Z, Dong H J, Zheng W Y, et al.Review and prospect of data-driven techniques for load forecasting in integrated energy systems[J]. Applied Energy, 2022, 321: 119269.
[16] 沈赋, 刘思蕊, 徐潇源, 等. 基于多尺度特征提取的IES多元负荷短期联合预测[J]. 高电压技术, 2024, 50(7): 2918-2930.
Shen F, Liu S R, Xu X Y, et al.Short-term joint forecasting of IES multivariate loads based on multi-scale feature extraction[J]. High Voltage Engineering, 2024, 50(7): 2918-2930.
[17] SONG C R, YANG H D, CAI J Y, et al.Multi-energy load forecasting via hierarchical multi-task learning and spatiotemporal attention[J]. Applied Energy, 2024, 373: 123788.
[18] 倪建辉, 张菁, 张昊立, 等. 基于时序特征灰度图与多任务学习的综合能源负荷短期预测[J]. 电气工程学报, 2024, 19(2): 186-199.
Ni J H, Zhang J, Zhang H L, et al.Short-term forecasting of integrated energy loads based on time-series feature grayscale maps and multi-task learning[J]. Journal of Electrical Engineering, 2024, 19(2): 186-199.
[19] Gao B X, Huang X Q, Shi J S, et al.Hourly forecasting of solar irradiance based on CEEMDAN and multi-strategy CNN-LSTM neural networks[J]. Renewable Energy, 2020, 162: 1665-1683.
[20] Cheng F, Liu H.Multi-step electric vehicles charging loads forecasting: an autoformer variant with feature extraction, frequency enhancement, and error correction blocks[J]. Applied Energy, 2024, 376: 124308.
[21] Wang C, Zhao H S, Liu Y, et al.Minute-level ultra-short-term power load forecasting based on time series data features[J]. Applied Energy, 2024, 372: 123801.
PDF(1730 KB)
