Analysis And Research On Battery Status By Machine Learning In Network Framework

In recent years,with the increasing awareness of people's environmental protection and the strong support of governments of various countries for the development of the new energy vehicle industry,the production of electric vehicles in China has increased year by year.To ensure that electric vehicles can run safely and stably,electric vehicles need to effectively monitor and manage their various indicators.The battery status is an important monitoring indicator of the battery pack.It is necessary to accurately estimate the battery pack health(SOH)and battery charge state(SOC)to provide a reliable basis for self-test and diagnosis of the electric vehicle.At the same time,extend the battery life and increase the electric vehicle's Economic benefits.Therefore,it is of great practical significance to accurately estimate the health status of the power lithium battery pack and accurately estimate the battery charge status based on real historical data.In the past,the research on the battery status was mainly on the monomer research in the experimental environment.Compared with the entire pack of batteries in actual production,the research object gap is larger.The more important thing is that the problem is not analyzed in conjunction with the actual situation,and end-to-end learning cannot be achieved,Not time-sensitive,unable to meet the needs of real production.This article takes the real historical data of electric vehicles accumulated on the cloud platform as the research object,establishes a battery pack health estimation model and battery state of charge estimation model,and preprocesses the data,including deletion,interpolation and deduplication,and divides charging The discharge event summarizes the user's usage behavior data and lays the foundation for the subsequent model establishment.In order to accurately estimate the health status of battery packs,this article references the literature and combines the battery information that can be mined in the data to explore the various influencing factors of the health status of lithium battery packs,and performs a detailed correlation analysis to explore the coupling between features performance,remove highly relevant input features,and simplify the model input.Secondly,in order to solve the problem of platform data,only the charging / discharging mark is made,and the problem of fast and slow charging events cannot be distinguished.Based on some observable samples,this article uses the SOMTE resampling method to solve the problem of uneven distribution of the number of fast and slow charging events,and builds a SVM classification model of fast and slow charging events,which separately counts the cumulative charge under fast charging conditions and the cumulative charge under slow charging conditions as input features of the final model.After determining the model input parameters,this paper uses a gradient boosting tree model(GDBT)to establish a battery health state estimation model.After training the model and optimizing the parameters such as model learning rate,maximum depth,and number of iterations,the model is finally established.Experimental results show that the gradient-based tree model has higher estimation accuracy than other models and can meet the actual needs.In order to prove the feasibility of the model in practical applications,this article builds a Flask network framework based on python language,simulates the client to send data requests to the cloud,and the cloud uses the trained algorithm model and historical data to return the SOH value to the client.This can reflect the superiority of using historical data combined with machine learning to build a battery health estimation model,and the feasibility of implementing the model in actual scenarios through a network framework.Finally,by comparing different SOC estimation methods,a method using LSTM time series model to estimate the battery SOC during movement is proposed,and the ideal accuracy is achieved.