深度信念网络在建筑能耗预测领域的应用研究  

Research on Application of Deep Belief Network in Building Energy Consumption Prediction

作　　者：侯博文 吴俊峰 陈焕新[1] 徐成良[1] HOU Bowen;WU Junfeng;CHEN Huanxin;XU Chengliang(School of Energy and Power Engineering,Huazhong University of Science and Technology,Wuhan 430074,Hubei,China;State Key Laboratory of Compressor Technology(Anhui Laboratory of Compressor Technology),Hefei 230031,Anhui,China)

机构地区：[1]华中科技大学能源与动力工程学院,湖北武汉430074 [2]压缩机技术国家重点实验室(压缩机技术安徽省实验室),安徽合肥230031

出　　处：《制冷技术》2021年第5期72-77,99,共7页Chinese Journal of Refrigeration Technology

基　　金：国家自然科学基金(No.51876070,No.51576074);压缩机技术国家重点实验室(压缩机技术安徽省实验室)开放基金(No.SKL-YSJ201801)。

摘　　要：本文以河南省某办公建筑夏季制冷季的能耗数据为研究对象,建立了基于深度信念网络的建筑空调能耗预测模型,针对原始数据中存在的不合理数据,分别采用时间序列分解法和箱线图法进行异常值检测,提出了一种对深度信念网络模型参数进行优化的选择策略,采取实验和经验相结合的方法确定预测模型的最佳隐藏层层数和隐藏层节点数,然后采用布谷鸟搜索算法对预测模型中的核参数g和惩罚因子C进行寻优,得到模型最优参数。结果表明,优化后模型的平均绝对误差由1.72下降到1.05,优化效果显著。The energy consumption data of an office building in the cooling season of Henan Province are taken as the research object in this paper.A prediction model of building air conditioning energy consumption based on a deep belief network is established.For the unreasonable data in the original data,time series decomposition method and box-line graph method is used to detect outliers respectively,and a selection strategy for optimizing the parameters of the deep belief network model is proposed.A combination of experiment and experience is used to determine the optimal hidden layer number and hidden layer node number of the prediction model.The cuckoo search algorithm is used to optimize the kernel parameter g and the penalty factor C in the prediction model to obtain the optimal model parameters.The results show that the average absolute error of the optimized model decreases from 1.72 to 1.05,and the optimization effect is significant.

关 键 词：建筑能耗预测 深度信念网络 布谷鸟搜索算法 异常值检测 

分 类 号：TU834[建筑科学—供热、供燃气、通风及空调工程] TP312[自动化与计算机技术—计算机软件与理论]