Identification And Repair Of Abnormal Data In Energy Consumption Monitoring Platform

With the continuous development of urbanization and energy conservation of public buildings,many cities across our country have gradually established energy consumption monitoring platforms for large public buildings with national network.The following methods and rules,such as energy consumption statistics,energy audits,energy efficiency announcements,energy consumption quota and fare increase of over-quota,can improve the level of energy-saving management on large-scale public buildings and prepare for further energy-saving renovation of high energy consumption buildings.As the basis for guiding the operation management,energy-saving diagnosis and energy efficiency evaluation of buildings,the correctness and reliability of energy consumption data are essential.The monitoring meters of building energy monitoring system are usually installed in an environment with strong electromagnetic interference such as substations.The target buildings are widely distributed and the network environment is complex and variable.Based on the above reasons,energy consumption data are easily affected during collection,communication,transmission and storage,which cause the quality of data on the building energy monitoring platform is generally low,and generate a large number of problem ones.To solve the above problems,based on the survey results of several public building energy consumption monitoring platforms across the country,this paper firstly studied the data quality problems among the underlying network,data transmission and data processing of existing platforms,identified the definition and characteristics of various types of abnormal data in such platforms.Secondly,this paper proposes a method for identifying invisible abnormal data which is based on power consumption characteristics.This method uses the improved k-means++ clustering algorithm to identify visible abnormal ones.According to the energy characteristics under different working conditions,this method firstly classifies the energy usage mode,then cleans the historical hourly electricity consumption data and forms a historical electricity characteristic line.Through comparing the difference of line slope characteristic between timely data and historical data in the same energy usage mode,the invisible abnormal data is further identified.Finally,by using this method to a real building experimentally,it was verified effective on identifying abnormal energy consumption data with high accuracy and wide applicability.Thirdly,based on the knn algorithm,this paper proposes a new missing data repair method by power consumption feature,namely knn-slope algorithm.The data repair accuracy,the relationship between data missing rate and repair accuracy and real-time repair effects under different methods are studied.In addition,for the knn-slope algorithm,this paper further explores its advantages,analyzes the repairing effects of different k-values,different historical dataset sizes,and the way to reduce the relative error of electrical feature points.By using it in real case,the method shows fast responding speed and high precision.Finally,this paper develops a new software to evaluate and repair data of building energy consumption monitoring platform,and applies the identification and repair algorithm in this paper to actual energy consumption monitoring platform.This paper provides theoretical guidance and technical methods for solving data quality problems in energy consumption monitoring systems.It has important theoretical and practical significance for improving the application and development of building energy consumption monitoring technology in China.