基于模式划分的空调能耗混合填补方法  

Mode division based hybrid filling method of air conditioning energy consumption

作　　者：孙鸿昌 周风余[1] 单明珠 翟文文 牛兰强 SUN Hongchang;ZHOU Fengyu;SHAN Mingzhu;ZHAI Wenwen;NIU Lanqiang(School of Control Science and Engineering,Shandong University,Jinan 250061,Shandong,China;Institute of Intelligent Buildings,Shandong Dawei International Architecture Design Co.,Ltd.,Jinan 250101,Shandong,China)

机构地区：[1]山东大学控制科学与工程学院,山东济南250061 [2]山东大卫国际建筑设计有限公司机电智能化设计研究院,山东济南250101

出　　处：《山东大学学报（工学版）》2022年第1期9-18,共10页Journal of Shandong University（Engineering Science）

基　　金：国家自然科学基金资助项目(61773242);住房和城乡建设部科技计划资助项目(2020-K-083);山东省住房和城乡建设科学技术计划资助项目(2020-K3-10)。

摘　　要：针对公共建筑能耗监测平台采集的空调能耗数据存在缺失、异常等问题,提出一种将无监督学习与监督学习结合的基于模式划分的空调能耗混合填补方法。利用k-means聚类算法将空调能耗数据划分至制冷、制热和独立新风3种运行模式中。在制冷及制热模式下,提出了一种BP神经网络(back propagation neural network,BPNN)和改进粒子群优化算法(amelio-rate particle swarm optimization,APSO)相结合的混合填补策略,进行能耗填补,采用随机森林(random forest,RF)作为特征提取策略,用改进的惯性权重和速度更新方程的APSO优化BPNN的初始参数;在独立新风模式下,采用k最邻近算法(k-nearest neighbor,kNN)填补能耗。青岛市某商场空调能耗试验数据分析结果表明,与RF-APSO-BPNN算法、BPNN算法、小脑神经网络算法(cerebellar model articulation controller,CMAC)相比,本研究方法填补空调能耗的平均百分比误差分别减少了53.44%、69.39%、62.15%。RF-APSO-BPNN-kNN混合方法填补空调能耗更优于其他算法。Aiming at the missing and abnormal air conditioning energy consumption data collected by the energy consumption monito-ring platform of public buildings,a hybrid filling method of air conditioning energy consumption based on mode division,which com-bined unsupervised learning and supervised learning,was proposed.The k-means clustering algorithm was employed to divide air con-ditioning energy consumption data into refrigeration,heating and independent fresh air operation modes.In the refrigeration and heating modes,a hybrid filling strategy combining back propagation neural network(BPNN)and ameliorate particle swarm optimization(APSO)was proposed to fill energy consumption data.In this hybrid model,the random forest(RF)was adopted as feature extrac-tion strategy,and the APSO algorithm with improved inertia weight and velocity update equation was utilized to optimize the initial pa-rameters of BPNN.In the independent fresh air mode,the k-nearest neighbor(kNN)algorithm was used to fill the energy consump-tion data.The experimental analysis results of air conditioning energy consumption data in a shopping mall in Qingdao showed that compared with the RF-APSO-BPNN algorithm,BPNN algorithm and cerebellar model articulation controller(CMAC)algorithm,the average percentage error of the proposed method was reduced by 53.44%,69.39%and 62.15%respectively.Therefore,the RF-APSO-BPNN-kNN hybrid method was superior to other algorithms to fill air conditioning energy consumption.

关 键 词：公共建筑 空调能耗数据 运行模式划分 RF-APSO-BPNN-kNN混合模型 机器学习 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]