Non-intrusive Load Identification Method Based On Improved LSTM Neural Network

With the development of smart grid technology and the spread of new consumption concept,people begin to pay more attention to energy management,energy conservation,emission reduction and renewable energy.Advanced and effective demand side management technology can provide security and reliability guarantee for smart grid development.At the same time,understanding the characteristics of residential electricity behavior is the primary prerequisite for realizing demand-side management.Compared with other methods,non-invasive load identification has the advantages of simple data collection,low cost,strong operability and convenient maintenance.Therefore,it has been gradually valued by scholars at home and abroad in recent years.Through non-invasive load identification,a large amount of user information can be obtained at a relatively low cost.From the perspective of users,it is convenient for residents to make appropriate adjustments to the household power consumption mode and reduce the electricity cost.From the perspective of power operators,non-invasive load identification can assist them in evaluating user behavior.It also helps them to manage energy distribution and integrate fluctuating energy sources.From a national perspective,load identification can provide data support for energy policy formulation and energy development plan.In previous studies,event information and load characteristics are extracted manually,which may cause unsatisfactory identification effect.In addition,most of the existing research results were based on the high-frequency data set,which has a large amount of data analysis and requires special sampling devices.This is not conducive to the practical application of non-invasive technology.Therefore,the purpose of this research is to study these problems.A non-invasive load identification model based on improved LSTM neural network is constructed in this thesis.The LSTM neural network has been improved to obtain identification system which can realize home load dentification.The effect of the load identification model is simulated by Python.The main work is as follows:(1)Firstly,the previous research results are analyzed in depth.Then,the characteristics of various algorithms are studied from three aspects:load identification event detection,feature extraction and load identification.The advantages,disadvantages and difficulties of manual load identification were summarized.(2)The basic steps of non-invasive load identification are designed,the characteristics of household load data are analyled,and the different types of electrical appliances are classified.In order to improve the effect of load identification,the wavelet denoising algorithm based on Grubbs criterion is used to deal with the outlier of data of each appliance to ensure the validity of load feature.(3)In order to make full use of the data in the database,it is necessary to extract multiple time-dependent sequences from the original load data.The parallel input of a large number of sequences into the neural network will lead to the training process is too complex,so the principal component analysis and kernel principal component analysis are used to fuse the above extracted time-dependent sequences.(4)A non-invasive load identification model is constructed.The LSTM neural network is improved by using the Back Propagation neural network,which can directly output the power data of a single electrical appliance.Then,the Dropout algorithm is added to the BP neural network to reduce the dependence of each neuron in the neural network.So as to guarantee the independence of load features extracted by neural network and prevent the occurrence of overfitting.(5)Finally,Bagging algorithm is used to integrate the identification results of LSTM neural network.This step further avoids the over-fitting of training and enhances the overall stability of the load identification model.Through experiments,the model is proved to be able to achieve high accuracy load identification.In this thesis,the identification effect of different scenes is analyzed to verify that the proposed load identification model can identify most electrical equipment in different scenes.