Simulation,Prediction And Regional Identification Classification Of Energy Consumption Of VRF System In University Dormitory Building

University buildings accounted for no more than 7%of urban building.As to the total building energy consumption,the energy consumption o f university buildings accounted to 22%.Variable refrigeration flow system always was the optimal air-conditioning system scheme of small-sized or medium-sized building,because of its high energy efficiency,flexible configuration,energy saving and low operating cost.University dormitory building and VRF system were the objects of the research.The purposes of the study were testing the reliability of simulated results and the feasibility of combining the machine learning and the simulated results.Simulating VRF system energy consumption of university dormitory and referring to the investigation results of dormitory’s air conditioning energy consumption,the simulated outcomes of air conditioning energy consumption were reasonable.The regional identification classification models were respectively based on the CART decision tree algorithm and the Na?ve Bayesian C lassifier.The effect had been anatomized which feature selection or parameter optimization had on the model results.The results showed parame ter optimization significantly enhanced the performance of regional identification model.The precision of the CART had been improved from 85.76%to 87.19%,the AUC of each category came between 0.9959 and 0.9972.At the same time,the precision of the Na?ve Bayesian C lassifier had been improved from 37.83%to 48.36%and the AUC came between 0.6512 and 0.7146.Generally speaking,the CART is more suitable than the Na?ve Bayesian C lassifier when it comes to regional identification of energy consumption of VRF system in university dormitory.In this research,the results of BPNN air-conditioning energy consumption prediction models had been compared,which consisted of different neural network structures,different activation functions and different training a lgorithms.Other being equal,it showed BPNN1 and BPNN4 had great performance,which respectively based on Bayesian regularization algorithm and Levenberg-Marquardt algorithm.Comparing them with Linear R,the run duration of BPNN4 and BPNN1 were more than 1 seconds,which of Linear R was 0.1 seconds,but the fitting performance and anti-interference ability of them were better than Linear R.At last,the generalization performance of BPNN1 and BPNN4 was checked by the chiller energy consumption of the hotel.The R~2 and RMSE of BPNN1 were 0.9914 and 2.2244 respectively,which of BPNN4 were 0.9917 and 2.1273respectively.However,the run duration of BPNN1 was about five times than the run duration of BPNN4.The results suggested the BPNN4 could be widely used in energy consumption prediction of air-conditioning system in other kind of building.