贫乏信息下基于深度迁移学习的智慧建筑负荷预测方法

摘要
图/表
参考文献
相关文章 (15)

全文: PDF (894 KB) HTML (1 KB)
输出: BibTeX | EndNote (RIS)

摘要精确的负荷预测可以显著优化建筑设备运行策略,释放建筑节能潜力,其中基于数据驱动的负荷预测模型得到了广泛关注。然而,一些具有新建数据平台的建筑能够为模型提供的数据非常有限。为了提高在贫乏信息状态下的预测性能,本文建立一种基于因果卷积神经网络结构的深度迁移学习负荷预测模型。首先,引入迁移学习机制,将与目标建筑类型一致的建筑作为源域建筑为模型提供数据;其次,引入全卷积神经网络和因果逻辑约束增强时间序列特征表达,以增加输出神经元的接受域并提升预测精度;最后,以实际工程中的办公建筑为例,对所提出的方法和所建模型进行验证,并将预测结果与长短期记忆网络模型的预测结果相比。结果显示,本文所建模型将预测值与实际值的平均绝对百分比误差和方均根误差系数分别平均降低了22.88%和0.372 0。

	服务

	把本文推荐给朋友
	加入我的书架
	加入引用管理器
	E-mail Alert
	RSS
	作者相关文章
	江晶晶
	窦真兰
	杨海涛
	赵敏

关键词 ：负荷预测, 迁移学习, 深度学习, 卷积神经网络（CNN）

Abstract：Accurate load forecasting can significantly optimize the operation strategy of equipment and release the energy-saving potential of buildings. With the advancement of computer science and smart meters, data-driven load forecasting models have become popular because of good forecasting accuracy. To improve the forecasting performance under poor information, this paper proposes a deep-transfer-learning predictive method based on the causal convolutional neural network. Taking three office buildings of the same type as an example, one of them is set as the target building and the other two are set as the source building. The proposed method and the built model are verified, and the forecasting results are compared with the long short-term memory network model. The forecasting results show that the model built in this paper reduces the average percentage error and the root mean square error coefficient of the predicted value and the actual value by 22.81% and 38.85%, respectively. Finally, this paper selects a building with better forecasting accuracy and compares and analyzes its annual cold, heat, and electrical load forecast results. The results show that the forecast accuracy of electricity and heat load is better than that of cooling load.

Key words： load forecasting transfer learning deep learning convolutional neural network (CNN)

收稿日期: 2021-09-29

基金资助:国家电网公司总部科技项目（5400-202017201A-0-0-00）

作者简介: 江晶晶（1982—）,女,浙江省宁波市人,硕士,工程师,主要从事电能计量装置安装与故障处理方面的研究工作。

引用本文:

江晶晶, 窦真兰, 杨海涛, 赵敏. 贫乏信息下基于深度迁移学习的智慧建筑负荷预测方法[J]. 电气技术, 2022, 23(5): 55-61. JIANG Jingjing, DOU Zhenlan, YANG Haitao, ZHAO Min. Load forecasting method of smart building based on deep transfer learning under poor information. Electrical Engineering, 2022, 23(5): 55-61.

链接本文:

http://dqjs.cesmedia.cn/CN/Y2022/V23/I5/55

[1] 王健, 阮应君. 建筑分布式能源系统设计与优化[M].上海: 同济大学出版社, 2018.
[2] JRADI M, RIFFAT S.Tri-generation systems: energy- policies, prime movers, cooling technologies, configu- reations and operation strategies[J]. Renewable & Sustainable Energy Reviews, 2014, 32: 396-415.
[3] 廖丽娜. 面向建筑的天然气三联供与可再生能源的耦合优化设计[D]. 上海: 上海电力学院, 2015.
[4] 潘毅群, 郁丛, 龙惟定, 等. 区域建筑负荷与能耗预测研究综述[J]. 暖通空调, 2015, 45(3): 33-40.
[5] KAVGIC M, MAVROGIANNI A, MUMOVIC D, et al.A review of bottom-up building stock models for energy consumption in the residential sector[J]. Building and Environment, 2010, 45(7): 1683-1697.
[6] KONTOKOSTA C E, TULL C.A data-driven predi- ctive model of city-scale energy use in buildings[J]. Applied Energy, 2017, 197: 303-317.
[7] 燕达, 陈友明, 潘毅群, 等. 我国建筑能耗模拟的研究现状与发展[J]. 建筑科学, 2018, 34(10): 130-137.
[8] 罗凤章, 张旭, 杨欣, 等. 基于深度学习的综合能源配电系统负荷分析预测[J]. 高电压技术, 2021, 47(1): 23-32.
[9] 周念成, 廖建权, 王强钢, 等. 深度学习在智能电网中的应用现状分析与展望[J]. 电力系统自动化, 2019, 43(4): 180-191.
[10] HAN Fujia, PU Tianjiao, LI Maozhen, et al.Short- term forecasting of individual residential load based on deep learning and K-means clustering[J]. CSEE Journal of Power and Energy Systems, 2021, 7(2): 261-269.
[11] RUNGE J, ZMEUREANU R.A review of deep learning techniques for forecasting energy use in buildings[J]. Energies, 2021, 14(3): 1-26.
[12] WEI Nan, LI Changjun, PENG Xiaolong, et al.Conventional models and artificial intelligence-based models for energy consumption forecasting: a review[J]. Journal of Petroleum Science and Engineering, 2019, 181: 106187.
[13] RUAN Yingjun, LIU Qingrong, LI Zhengwei, et al.Optimization and analysis of building combined cooling, heating and power (BCHP) plants with chilled ice thermal storage system[J]. Applied Energy, 2016, 179: 738-754.
[14] 徐萍, 吴超, 胡峰俊, 等. 基于迁移学习的个性化循环神经网络语言模型[J]. 南京理工大学学报, 2018, 42(4): 401-408.
[15] YU Chongchong, CHEN Yunbing, LI Yueqiao, et al.Cross-language end-to-end speech recognition research based on transfer learning for the low-resource Tujia language[J]. Symmetry, 2019, 11(2): 179.
[16] 亢洁, 李佳伟, 杨思力. 基于域适应卷积神经网络的人脸表情识别[J]. 计算机工程, 2019, 45(12): 201-206.
[17] WANG Shanshan, ZHANG Lei, ZUO Wangmeng, et al.Class-specific reconstruction tansfer learning for visual recognition across domains[J]. IEEE Transa- ctions on Image Processing, 2020, 29: 2424-2438.
[18] GUO Li, LIU Wenjian, CAI Jiejin, et al.A two-stage optimal planning and design method for combined cooling, heat and power micro-grid system[J]. Energy Conversion and Management, 2013, 74: 433-445.
[19] 曾竞. 区域供冷供热管网系统优化设计研究[D]. 长沙: 湖南大学, 2017.
[20] FANG Tingting, LAHDELMA R.Genetic optimization of multi-plant heat production in district heating networks[J]. Applied Energy. 2015, 159: 610-619.
[21] 陈振祥, 林培杰, 程树英, 等. 基于K-means++和混合深度学习的光伏功率预测[J]. 电气技术, 2021, 22(9): 7-13, 33.
[22] 洪华伟, 王峰, 郑鹏, 等. 基于模糊综合评价与主成分分析的客户用电能效研究[J]. 电气技术, 2020, 21(10): 29-34, 40.
[23] 王东, 史晓霞, 尹交英. 不同核函数的支持向量机用于空调负荷预测的对比研究[J]. 电工技术学报, 2015, 30(增刊1): 531-535.
[24] 刘念, 张清鑫, 刘海涛. 基于核函数极限学习机的微电网短期负荷预测方法[J]. 电工技术学报, 2015, 30(8): 218-224.
[25] 张帅, 杨晶显, 刘继春, 等. 基于多尺度时序建模与估计的电力负荷数据恢复[J]. 电工技术学报, 2020, 35(13): 2736-2746.
[26] 储晨阳, 秦川, 鞠平, 等. 基于优化稀疏编码的超短期负荷滚动多步预测[J]. 电工技术学报, 2021, 36(19): 4050-4059.
[27] KRIZHEVSKY A, SUTSKEVER I, HINTON G E.Imagenet classification with deep convolutional neural networks[J]. Communicatios of the ACM, 2017, 60(6): 84-90.