基于PCA-LSTM神经网络的建筑空调负荷预测方法研究  

Research on Load Forecasting Method of Building Air Conditioning Based on PCA-LSTM Neural Network

作　　者：顾春锋 罗其华 奚培锋 张少迪 胡桐月 GU Chunfeng;LUO Qihua;XI Peifeng;ZHANG Shaodi;HU Tongyue(Lingang Energy Service Center of State Grid Pudong Power Supply Company,Shanghai 200120,China;School of Automation Engineering,Shanghai University of Electric Power,Shanghai 200090,China;Shanghai Elctrical Apparatus Research Institute(Group)Co.,Ltd.,Shanghai Key Laboratory of Smart Grid Demand Response,Research and Development(Experiment)Center of Electrical Equipment for National Energy Smart Grid User Side,Shanghai 200063,China)

机构地区：[1]国网浦东供电公司临港能源服务中心,上海200120 [2]上海电力大学自动化工程学院,上海200090 [3]上海电器科学研究所(集团)有限公司,上海市智能电网需求响应重点实验室,国家能源智能电网用户端电气设备研发(实验)中心,上海200063

出　　处：《现代建筑电气》2021年第10期1-7,共7页Modern Architecture Electric

基　　金：国网上海浦东供电公司2021年区域虚拟电厂体系建设及示范目(6409212000ND);上海市科学技术委员会科研计划项目(19DZ1206700)。

摘　　要：针对传统的机器学习模型无法同时处理空调负荷数据的时序性和非线性问题,提出一种基于主成分分析(PCA)和长短记忆(LSTM)神经网络空调负荷预测方法,利用主成分分析法对影响空调负荷的多元数据进行降维,得到主成分数据序列;然后建立基于PCA-LSTM神经网络的空调负荷预测模型;最后以上海某建筑空调负荷的数据进行仿真验证,并通过与传统BP预测模型以及不进行主成分分析的LSTM预测模型进行对比分析,仿真结果显示所提预测方法具有更高的精度和更好的泛化性。Due to the reason of traditional machine learning model can not deal with the time-series and nonlinear of air conditioning load data at the same time,a forecasting model of air conditioning load based on principal component analysis(PCA)and long short term memory(LSTM)neural network is proposed.The principal component data sequence is obtained by reducing the dimension of the multivariate data which affecting the air conditioning load by principal component analysis,and the main influencing factors are obtained.Then,the prediction model of air conditioning load based on PCA-LSTM neural network is established.Finally,taking the air conditioning load data of a building in Shanghai to verify,the traditional prediction BP model and the LSTM prediction model without principal component analysis are compared.The simulation results show that the proposed prediction method has higher accuracy and better generalization.

关 键 词：空调负荷预测 主成分分析法 长短记忆神经网络 传统模型 

分 类 号：TU201.5[建筑科学—建筑设计及理论]