Research On Annual Air Conditioning Load Forecasting Of Shopping Malls In Pearl River Delta Region Based On Machine Learning Method

Reasonable architectural design scheme is very important to reduce building energy consumption.However,it is generally difficult to consider the comparison of building energy consumption of different schemes.Hence there are the congenital shortages of energy saving in the design of shopping malls,and it is difficult to reduce the energy consumption through the selection of the envelope structure.It is necessary to quickly and accurately predict the annual air conditioning load in the architectural design of shopping malls in order to assist the decision-making of architectural design schemes.This study took the shopping malls in Pearl River Delta region as the research object,the on-the-spot investigation was conducted and the internal heat source work model was established at first,in order to realize the modularization and parameterization of the air conditioning load calculation model of the shopping malls.Then,it was to compare the accuracy and applicability of the prediction results of various machine learning methods.And the rapid forecasting model for the annual air conditioning load in shopping malls was established.Furthermore,it was to develop software for predicting and analyzing building load of shopping malls in Pearl River Delta region.Firstly,the design data collection,indoor temperature,humidity measurement and questionnaire survey were used to investigate the actuality and characteristics of influencing factors of architectural characteristics,the distribution ratio of the business,thermal parameters of the envelope structure,internal heat source parameters,air conditioning interior design parameters and so on for air conditioning load in the shopping malls in Pearl River Delta region.And the range of the factors affecting the air conditioning load of the shopping malls in Pearl River Delta region were determined.Then according to the characteristics of various floors of shopping malls,nine combination modes with typical business were proposed,and the concept of modular design was introduced to use them as basic modules.The accurate description of building model for every shopping mall was realized by calling the corresponding modules and superimposing.Secondly,based on the data statistics of the heat sources in the shopping malls,the schedule of the personnel,lighting and electrical equipment of the typical format was established,and the accuracy was verified by the actual operational data of the two malls.Consequently,the establishment of the internal heat source model of the typical shopping mall format can improve the accuracy of the annual air conditioning load simulation and realize the refined simulation of the energy consumption of the shopping malls.Thirdly,the parameterization method of the air-conditioning load calculation model of the shopping mall was proposed and verified.The Latin Hypercube sampling method was used to generate 80,000 random combinations of different factors affecting the air conditioning load.Then the R language and EnergyPlus were used to generate and simulate 80,000 sets of calculation models.After that,the simulation results collected in the unit of the floor were used to form the sample database of annual air conditioning load corresponding to nine combination modes with typical business.The sample database was used for subsequent machine learning methods for training and testing.Finally,by comparing the accuracy of 11 different machine learning method models,the machine learning method with the highest model quality was Gradient Boosting.The annual air conditioning load forecasting model of the shopping malls in Pearl River Delta region based on Gradient Boosting was established.And it was to develop forecasting analysis software of building load for the shopping malls in Pearl River Delta region.Then sensitivity analysis was introduced to determine the influence of various influencing factors on the annual air conditioning load of the shopping malls.The accuracy of the annual air conditioning load forecasting model of different input variable simplified schemes was analyzed,and the simplified principle of input variables was proposed.Finally,aiming at the problem that the missing values in the actual operation data of the shopping malls can not be used as complete samples to further optimize the forecasting model,the EM algorithm and MICE are selected as the best missing values filling algorithm,so as to facilitate the application of actual operation data to modify the forecasting model.