Research On Non-intrusive Residential Electricity Identification Method

In recent years,the increasing global population has led to an increasingly tense energy situation.The proportion of coal and oil in the energy consumption terminal has continued to decline,while the demand for clean energy such as electricity has been growing rapidly.At the same time,the arrival of intelligent era makes the types of power load and total power load consumption of residential users increase year by year.In order to promote the construction of smart power grid,improve the interaction between supply and demand between power grid and users,and guide users to develop more reasonable energy use habits,power grid needs to further strengthen the identification and analysis of residential load.Load identification technology can be divided into intrusive load monitoring and non-intrusive load monitoring according to different data collection methods.The traditional intrusive load monitoring realizes data acquisition by installing independent sensors and other hardware on each electrical device.Its monitoring accuracy is high,but the input of manpower and material resources is large.Under the background of the gradual increase of the number of home appliances,the flexibility and scalability of intrusive load monitoring is poor,and it is no longer applicable to most residents.Compared with the intrusive load monitoring,the non-intrusive load monitoring technology eliminates a lot of sensing equipment,and realizes load monitoring by extracting the features of electrical equipment based on the only bus data.Due to its high reliability,flexibility and low economic cost,NILM is gradually replacing the invasive load monitoring.The non-intrusive electric quantity monitoring method for residential users proposed in this thesis is based on the optimization modeling of deep learning algorithm,which uses steady-state features such as one-dimensional active power sequence to carry out load monitoring and electric quantity monitoring.Overcome the high cost of high frequency transient data in the early acquisition and data preprocessing stage.In this thesis,a nonlinear regression fitting interpolation method based on sliding window is established based on the characteristics of NILM public data set.Then,various network structures of deep learning algorithm are deeply studied,and a compound neural network oriented to one-dimensional time series is built to optimize the monitoring effect of residential load.Finally,on the basis of load monitoring,a data denoising and reconstruction method is proposed,and a non-intrusive electric quantity monitoring model is built based on the deep codec mechanism,and the simulation verification is completed.The load monitoring algorithm in this thesis has strong applicability to different household data,which can help the power grid to understand residents’ power consumption behavior and provide users with a more perfect power consumption strategy.