Research On Non-intrusive Household Appliances Recognition Method Based On Deep Learning

With the development of social economy,the speed of energy consumption among residential users has increased year by year in recent years,which leads to increasingly serious energy shortage and environmental pollution problems.To improve energy efficiency and save energy effectively has become an urgent problem to be solved.Load monitoring not only can realize energy control and fault monitoring,but also can helps users to save energy efficiently,so the realization of load monitoring is very important to the construction of an economically and environmentally friendly society.Non-intrusive load monitoring has become a research hotspot of current load monitoring because of its low cost,easy implementation and easy installation and maintenance.Under the background of non-intrusive load monitoring,this paper takes typical household appliances as the research object,and combines the deep learning technology to study the non-intrusive identification method of household appliances.Mainly from the following aspects of research:(1)Firstly,the hardware experiment platform is built to collect the voltage and current data of the steady-state operation of the electrical appliances,complete the data acquisition.The signals is analyzed and preprocessed,and then extract the voltage-current(V-I)trajectory of the electrical appliances.(2)Household appliance recognition based on deep learning.Firstly,aiming at the problem that V-I map is not available due to noise interference in the measured data,a V-I map data screening algorithm based on finite field is proposed.For the V-I map screened by different parameters,the V-I map dataset is constructed,and the deep learning network is used to find the best screening parameters and electrical dataset with the best recognition effect.Then on the basis of the classic network,the network is improved,and the training experiment is carried out from the three aspects of model depth,learning rate and iteration number,so as to design the network structure for household appliance identification(3)Household appliances recognition based on HOG feature and trajectory area feature.From the image level,the local features of HOG are extracted from V-I map of electrical appliances,and the relative pixel area features are extracted from the trajectory shape level.Combining with support vector machine,two traditional methods are used to realize the recognition of household appliances.(4)Electrical appliances recognition based on deep transfer learning.In order to solve the problem that new household appliances can not be recognized,and retraining takes a lot of time,and the sample size of V-I data is small,which leads to the unsatisfactory recognition effect,this paper proposes a method of electrical appliances recognition based on deep migration learning,which combines weight transfer with electrical appliances recognition,and achieves a higher accuracy of electrical appliances recognition.This method can not only shorten the training time,but also improve the recognition effect.