[1]郭茂祖,于丰宁,王鹏跃,等.基于时序相关性的建筑能耗预测方法[J].智能系统学报,2026,21(1):214-225.[doi:10.11992/tis.202503013]
GUO Maozu,YU Fengning,WANG Pengyue,et al.Building energy consumption prediction method based on temporal correlation[J].CAAI Transactions on Intelligent Systems,2026,21(1):214-225.[doi:10.11992/tis.202503013]
点击复制
GUO Maozu,YU Fengning,WANG Pengyue,et al.Building energy consumption prediction method based on temporal correlation[J].CAAI Transactions on Intelligent Systems,2026,21(1):214-225.[doi:10.11992/tis.202503013]
基于时序相关性的建筑能耗预测方法
《智能系统学报》[ISSN 1673-4785/CN 23-1538/TP] 卷:
21
期数:
2026年第1期
页码:
214-225
栏目:
吴文俊人工智能科学技术奖论坛
出版日期:
2026-03-05
- Title:
- Building energy consumption prediction method based on temporal correlation
- Keywords:
- building energy consumption prediction; temporal correlation mining; building morphology features; self-supervised learning; contrastive loss function; encoder-decoder architecture; long-term dependency; temporal convolutional network
- 分类号:
- TP3-05
- 摘要:
- 建筑能耗预测对优化能源资源配置、推进节能减排措施及支撑可持续发展目标至关重要。建筑能耗数据因受季节更迭、节假日效应等因素影响，在时间序列上显现出周期性和非平稳性特征。现有方法通常采用滑动窗口建模局部时序特征，但仅能捕捉窗口内部变化，难以挖掘窗口之间潜在的长期演化趋势。此外，建筑形态对能耗具有显著影响，却在能耗预测任务中常被忽略。针对上述局限，本文提出一种基于时序相关性的建筑能耗预测方法，主要包含局部特征学习、全局特征学习及损失函数设计。针对窗口外部长期变化难以被捕捉的问题，全局特征学习模块采用编码器-解码器架构，建模滑动窗口之间的长期时序依赖。设计自监督对比损失函数，以窗口为单位构建正负样本对，进一步挖掘能耗数据的全局相关性。针对建筑形态特征未被重视的问题，通过嵌入建筑形态特征，并利用线性层捕捉滑动窗口内时间邻近能耗数据的局部相关性。实验结果表明，该方法在处理多座建筑能耗长短期预测任务中均取得了最好的预测精度，在未来24 h预测任务中，较常用能耗预测方法ARIMA、LSTM、GRU、Transformer、GWO-SARIMA-LSTM、Informer和Autoformer方法，该方法的预测精度分别提高了约17.06%、8.37%、9.79%、9.58%、9.83%、6.94%和5.55%，为建筑节能管理和用能行为优化提供科学支撑。
- Abstract:
- Building energy consumption prediction is crucial for optimizing energy resource allocation, advancing energy-saving and emission-reduction measures, and supporting sustainable development goals. Due to the influence of factors such as seasonal changes and holiday effects, building energy consumption data exhibits periodic and non-stationary characteristics in time series. Existing methods typically use sliding windows to model local temporal features, but they can only capture changes within the window and are limited in uncovering the long-term evolution trends across windows. In addition, architectural morphology significantly impacts energy consumption; however, it is often neglected in energy consumption prediction tasks. To address these limitations, this paper innovatively proposes a temporal correlation-based building energy consumption prediction method, which primarily includes local feature learning, global feature learning, and loss function design. To capture long-term changes beyond the window, the global feature learning module adopts an encoder-decoder architecture to model long-term temporal dependencies between sliding windows. Furthermore, a self-supervised contrastive loss function is designed, constructing positive and negative sample pairs at the window level to further explore the global correlations of energy consumption data. To address the neglect of architectural morphology features, building morphological features are embedded, and a linear layer is used to capture the local correlations of temporally adjacent energy consumption data within the sliding window. Experimental results show that the proposed method achieves the best prediction accuracy in multi-building energy consumption forecasting tasks, and in 24-hour ahead prediction tasks, the method improves prediction accuracy by approximately 17.06%, 8.37%, 9.79%, 9.58%, 9.83%, 6.94%, and 5.55% compared to commonly used energy consumption prediction methods including ARIMA, LSTM, GRU, Transformer, GWO-SARIMA-LSTM, Informer, and Autoformer. This method provides scientific support for building energy-saving management and energy usage behavior optimization.
- 参考文献/References:
-
[1] SANTAMOURIS M, VASILAKOPOULOU K. Present and future energy consumption of buildings: Challenges and opportunities towards decarbonisation[J]. E-prime-advances in electrical engineering, electronics and energy, 2021, 1: 100002
[2] GIBB D, LEDANOIS N, RANALDER L, et al. Renewables 2022 global status report[EB/OL]. (2024-06-05)[2025-03-07]. https://pure.iiasa.ac.at/id/eprint/19781/1/GSR2022_Full_Report.pdf.
[3] ZHANG Jianxin, HUANG Yao, CHENG Hengda, et al. Ensemble learning-based approach for residential building heating energy prediction and optimization[J]. Journal of building engineering, 2023, 67: 106051
[4] 冯增喜, 杨芸芸, 赵锦彤, 等. 基于人工智能的建筑能耗预测研究综述[J]. 建筑节能(中英文), 2023, 51(3): 22-29 FENG Zengxi, YANG Yunyun, ZHAO Jintong, et al. A review of research on artificial intelligence-based building energy consumption prediction[J]. Building energy efficiency (Chinese and English), 2023, 51(3): 22-29
[5] 刘兆濡, 燕达, 吴如宏, 等. 基于DeST的城市建筑能耗模拟平台开发[J]. 建筑科学, 2021, 37(10): 16-23 LIU Zhaoru, YAN Da, WU Ruhong, et al. Development of an urban building energy consumption simulation platform based on DeST[J]. Building science, 2021, 37(10): 16-23
[6] CHEN Yongbao, GUO Mingyue, CHEN Zhisen, et al. Physical energy and data-driven models in building energy prediction: a review[J]. Energy reports, 2022, 8: 2656-2671
[7] 韩少锋, 吴迪, 张圣原, 等. 基于EMD与机器学习算法的近零能耗建筑负荷预测方法[J]. 暖通空调, 2024, 54(7): 82-89,97 HAN Shaofeng, WU Di, ZHANG Shengyuan, et al. EMD and machine learning algorithms-based load prediction method for near-zero energy buildings[J]. Heating ventilating & air conditioning, 2024, 54(7): 82-89,97
[8] HE Yuxuan, GONG Qijian, ZHOU Zhenxin, et al. Development of a hybrid VRF system energy consumption prediction model based on data partitioning and swarm intelligence algorithm[J]. Journal of building engineering, 2023, 74: 106868
[9] JIANG Pei, WANG Zuoxue, LI Xiaobin, et al. Energy consumption prediction and optimization of industrial robots based on LSTM[J]. Journal of manufacturing systems, 2023, 70: 137-148
[10] DOAN C. Comparing encoder-decoder architectures for neural machine translation: a challenge set approach[D]. Ottawa: University of Ottawa, 2021.
[11] HAN Mingdong, FAN Lingyan. A short-term energy consumption forecasting method for attention mechanisms based on spatio-temporal deep learning[J]. Computers and electrical engineering, 2024, 114: 109063
[12] PATRO B N, NAMBOODIRI V P, AGNEESWARAN V S. SpectFormer: frequency and attention is what you need in a vision transformer[EB/OL]. (2023-04-13)[2025-03-07]. https://arxiv.org/abs/2304.06446.
[13] KLYUEV R V, MORGOEV I D, MORGOEVA A D, et al. Methods of forecasting electric energy consumption: a literature review[J]. Energies, 2022, 15(23): 8919
[14] ZHENG Shirong, LIU Shaobo, ZHANG Zhenhong, et al. TRIZ method for urban building energy optimization: GWO-SARIMA-LSTM forecasting model[EB/OL]. (2024-10-02)[2025-03-07]. https://arxiv.org/abs/2410.15283.
[15] ZHOU Haoyi, ZHANG Shanghang, PENG Jieqi, et al. Informer: beyond efficient transformer for long sequence time-series forecasting[C]//Proceedings of the AAAI Conference on Artificial Intelligence. Virtual Event: ACM, 2021, 35(12): 11106-11115.
[16] WU Haixu, XU Jiehui, WANG Jianmin, et al. Autoformer: decomposition transformers with auto-correlation for long-term series forecasting[J]. Advances in neural information processing systems, 2021, 34: 22419-22430
[17] TONEKABONI S, EYTAN D, GOLDENBERG A. Unsupervised representation learning for time series with temporal neighborhood coding[EB/OL]. (2021-06-01)[2025-03-07]. https://arxiv.org/abs/2106.00750.
[18] YUE Zhihan, WANG Yujing, DUAN Juanyong, et al. TS2Vec: towards universal representation of time series[C]//Proceedings of the AAAI Conference on Artificial Intelligence. Vancouver: AAAI, 2022, 36(8): 8980-8987.
[19] WOO G, LIU Chenghao, SAHOO D, et al. CoST: contrastive learning of disentangled seasonal-trend representations for time series forecasting[EB/OL]. (2022-02-03)[2025-3-77]. https://arxiv.org/abs/2202.01575.
[20] WANG Zhiyuan, XU X, ZHAGN Weifeng, et al. Learning latent seasonal-trend representations for time series forecasting[J]. Advances in neural information processing systems, 2022, 35: 38775-38787
[21] KO?UCH A, CYWICKA D, ADAMOWICZ K, et al. A comparison of artificial neural network and time series models for timber price forecasting[J]. Forests, 2023, 14(2): 177
[22] EMAMI P, SAHU A, GRAF P. BuildingsBench: a large-scale dataset of 900K buildings and benchmark for short-term load forecasting[EB/OL]. Advances in neural information processing systems, 2023, 36: 19823-19857.
[23] LU Xingyu, LIU Zhining, GUAN Yanchu, et al. GreenFlow: a computation allocation framework for building environmentally sound recommendation system[C]//Proceedings of the Thirty-Second International Joint Conference on Artificial Intelligence. Macao: IJCAI, 2023: 6103-6111.
[24] 姜之点, 杨峰. 街区尺度城市形态因子对建筑能耗影响的模拟研究[J]. 建筑科学, 2022, 38(6): 140-149 JIANG Zhidian, YANG Feng. Simulation study on the impact of neighborhood-scale urban morphological factors on building energy consumption[J]. Building science, 2022, 38(6): 140-149
[25] BOSE S, LI Yijiang, VAN SANT A, et al. From RNNs to foundation models: an empirical study on commercial building energy consumption[EB/OL]. (2024-11-21)[2025-03-07]. https://arxiv.org/abs/2411.1442.
[26] POTO?NIK P, ?KERL P, GOVEKAR E. Machine-learning-based multi-step heat demand forecasting in a district heating system[J]. Energy and buildings, 2021, 233: 110673
[27] 晏雄锋, 袁拓, 杨敏, 等. 建筑物形状特征分析表达与自适应化简方法[J]. 测绘学报, 2022, 51(2): 269-278 YAN Xiongfeng, YUAN Tuo, YANG Min, et al. Analysis and expression of building shape features and adaptive simplification method[J]. Acta geodaetica et cartographica sinica, 2022, 51(2): 269-278
[28] JENA M, DEHURI S. An integrated novel framework for coping missing values imputation and classification[J]. IEEE access, 2022, 10: 69373-69387
[29] CHUNG J, GULCEHRE C, CHO K, et al. Empirical evaluation of gated recurrent neural networks on sequence modeling[EB/OL]. (2014-12-11)[2025-03-07]. https://arxiv.org/abs/1412.3555.
[30] WANG Huiqiang, PENG Jian, HUANG Feihu, et al. MICN: multi-scale local and global context modeling for long-term series forecasting[C]//International Conference on Learning Representations. Kigali: ILCR, 2023.
[31] MILLER C, KATHIRGAMANATHAN A, PICCHETTI B, et al. The building data genome project 2, energy meter data from the ASHRAE great energy predictor III competition[J]. Scientific data, 2020, 7: 368
备注/Memo
收稿日期:2025-3-7。
基金项目:北京市自然科学基金项目(4232021).
作者简介:郭茂祖，教授，博士生导师，主要研究方向为机器学习、时空数据分析、智能建造与智慧城市、生物信息学。主持国家重点研发计划课题、国家自然科学基金重点和面上项目、北京市自然科学基金面上项目等多项，牵头获教育部自然科学奖、吴文俊人工智能自然科学奖、黑龙江省自然科学奖。发表学术论文300余篇。E-mail：guomaozu@bucea.edu.cn。;于丰宁，硕士研究生，主要研究方向为深度学习、计算性设计。E-mail：13563037298@163.com。;王鹏跃，讲师，博士，主要研究方向为机器学习、城市计算、计算性设计。E-mail：wangpengyue@bucea.edu.cn。
通讯作者:王鹏跃. E-mail：wangpengyue@bucea.edu.cn
更新日期/Last Update:
2026-01-05