Research On Resident Non-intrusive Load Identification Algorithm Based On Cloud-end Collaboration

Non-intrusive load identification technology is a kind of technology that can detect the state of load and manage energy consumption by collecting load features such as voltage and current at the user’s entrance.It can provide detailed load information for power supply company and customers,plan rationally the power system,reduce power consumption,and realize the scientific allocation of power resources.At present,most of the researches on non-intrusive load identification are carried out in the laboratory using synthetic data or public data sets,which are not applicable to engineering.Moreover,there is insufficient research on the identification of the important electric heating load in the family.The applications of fine-grained data mostly focus on the fault diagnosis,energy consumption management of residents and the demand response of power grid,and the research on social services is still blank.To solve the above problems,based on the real electricity consumption data of residential users,this paper proposes a residential non-intrusive load identification algorithm based on end-cloud cooperation.The terminal adopts the hierarchical classification algorithm,and the cloud carries out the secondary identification of the load to improve the overall accuracy.At the same time,the application of fine-grained data realizes the discrimination of abnormal user types,and provides data support for the law enforcement of relevant departments.Specific research contents are as follows:Firstly,load features and resident load models are proposed in this paper.Load features include terminal features and cloud features.Terminal features include active power,reactive power increment,active power fluctuation and slow power variation,transient active power shock,the second and third steady-state harmonic current features increment.The identification features of the cloud can be divided into two situations.The features of secondary identification of the large category of electrical appliances are obtained based on the identification results of the terminals through data analysis and processing,while the features of unknown electrical appliances are the load starting and ending features uploaded by the terminals.Finally,a variety of residential load models are established from four aspects of active power,reactive power,second and third harmonic current.Then,a non-intrusive load identification algorithm is proposed,including terminal side and cloud side load identification algorithms.Firstly,the end-cloud collaboration architecture of load identification is introduced,that is,the terminal makes first identification,and then uploads the identification results to the cloud,and the cloud makes second identification,so as to improve the overall accuracy.Hierarchical classification algorithm is adopted in terminal side,and electric heating identification algorithm based on fuzzy reasoning and new electric appliance clustering algorithm are adopted in cloud to solve the problem of insufficient research on electric heating load identification and difficult identification of new appliances at present.The experiments show that the identification accuracy of the terminal algorithm and the cloud algorithm proposed in this paper are both higher than 90%.Finally,the application of fine-grained data in social services was researched,and a house discrimination algorithm based on XGBoost was proposed.Firstly,the typical load curves of various types of houses are given,including vacant houses,group rental houses,private rental houses and online rental houses.Then,the feature extraction of users is researched,and an anomaly detection algorithm based on LOF is proposed for group renters and private renters.Finally,a user discrimination algorithm based on XGBoost is designed.The experiments show that the accuracy of the algorithm proposed in this paper is higher than Support Vector Machine(SVM)and Back-Propagation Artificial Neural Network(BP-ANN).The accuracy of the algorithm is more than 60% for group rental houses and private rental companies,and more than 50% for online rental houses.