Research On Data Preprocessing Methods Of Electric Energy Consumption For Air Conditioning In Public Buildings

In the process of collecting the original operation data,the public building energy consumption monitoring platform will be affected by many external factors,resulting in the lack or mutation of the data.Therefore,before the analysis of public building energy consumption data mining,data preprocessing had become an indispensable step,and improving the filling accuracy of operation problem data had become a key problem to be solved urgently.The energy consumption of air-conditioning in public buildings was relatively large,about 50%.The energy consumption data of air-conditioning in public buildings can reflect the real situation of the operation of air-conditioning in buildings.Analyzing and mining this type of data was an important link to reduce the energy consumption of public buildings in the future.At present,only a single algorithm was used to fill all the problem data in the data preprocessing of building air conditioning in China,and there was no more complete processing system,so the accuracy of filling the problem data was relatively low.According to the characteristics and influencing factors of the actual operation data of air conditioning,this paper proposes a reasonable and effective pretreatment scheme,and a variety of algorithms are introduced to compare and verify the prediction accuracy of the algorithm.The main research contents and results were showed:(1)Identify the energy consumption data of air conditioning with missing and abnormal conditions.R language was used to identify missing data directly with simple sentences;The abnormal data of buildings were divided into continuous repeated data and abrupt data.The method of direct recognition and K-means algorithm was used to identify all abnormal data.(2)Identify the operation mode of air conditioning and its correlation with the cooling and heating loads of buildings,and propose the data filling strategies for different operation modes of air conditioning.The air-conditioning operation mode was divided into three modes of heating,cooling and fresh air.Based on the original daily average energy consumption dataof air conditioning of the per month,according to K-means clustering results and climate characteristics,the air-conditioning operation mode of each month in each year is determined.Through the theoretical method and correlation coefficient method,the correlation degree between energy consumption data of air conditioning of each air-conditioning mode and outdoor environmental parameters was determined,and the correlation between each air conditioning operation mode and the cooling and heating load of the building was obtained.According to the correlation between two,the energy consumption data of air-conditioning was divided into air-conditioning power consumption data related to building cooling and heating load and air-conditioning power consumption data unrelated to building cooling and heating load.Based on this,the filling algorithm of each type of problem data was determined,instead of simply using one algorithm to directly solve all kinds of problem data.(3)Aiming at filling the data of air-conditioning power consumption related to the building cooling and heating load,a BP-POS·SVR-BP hybrid algorithm was proposed.This algorithm can take advantage of BP network to better find the optimal weight coefficient combination between the prediction results of BPNN and SVM.The prediction accuracy of single algorithm(BPNN,SVR)and hybrid algorithm was verified by experiments,and it was found that hybrid algorithm has higher filling accuracy.Aiming at filling in the problems data of power consumption of air-conditioning which are not related to the cooling and heating load of the building,comparing and verifying the prediction accuracy of KNN regression algorithm,mean value interpolation and multiple interpolation based on mice,it was found that KNN regression algorithm has higher filling accuracy.This research provided a certain reference for the selection of data preprocessing scheme and problem data filling algorithm of air conditioning power consumption data in the public building energy consumption monitoring platform.On the basis of the filling algorithm,slightly modify the problem data that can be applied to other sub item power consumption data of the building energy consumption monitoring platform.At the same time,the accuracy of data mining knowledge of air conditioning energy consumption,as well as building energy conservation provides the basis and guarantee.