Research On The Data-driven SOH Estimation And Rul Prediction Methods Of Lithium-ion Batteries

With the development of electrochemical energy storage technology,Lithium ion battery has become the main power source of electric vehicles and large-scale energy storage.Lithium ion battery will inevitably suffers performance degradation even reliability and safety problems.during cycling.Battery management system(BMS)evaluates the health status and life cycle information of battery in real time by combining the sensor data with battery health monitoring and management algorithms,which provides support for predictive maintenance of battery system and ensure safe and stable operation of energy storage.This paper carries out research on state assessment and life prediction technology of lithium ion batteries in different material systems.Focusing on the application of data-driven method in battery health management,the main research contents are as follows:(1)Aimming at capacity evaluation of batteries in different aging states,the general fading laws of batteries in different aging states are studied.Fristly,battery performance test and recycle test were designed to collect battery age data.Then,the inconsistency of battery aging state in interface state and the general law of cycle process degradation were analyzed with incremental capacity analysis method.Based on the basic law of battery decay,the peak area of capacity increment curve is selected as input feature,a multiple regression model is established with small battery data set.The capacity estimation method can evaluate the battery capacity quickly considering different aging modes with middle and small rate data.(2)Aimming at SOH estiamtion and RUL prediction in high rate and complicated conditions,a feature engineering pipeline for SOH estimation of battery was established.Firstly,based on the basic operation data of the battery,the lithium ion battery health characteristics system is constructed,which includes the full-range and sub-range statistical characteristics of the battery.Then,a feature selection method based on Filter method and Wrapper method is proposed to obtain the optimal subset of features.Based on the proposed feature engineering pipeline,three data-driven SOH estimation models for battery state of health estimation are constructed and validated with linear and accelerated fading battery data sets respectively.It is proved that support vector regression model performs best in both data sets with maximum error less than 3% and average error less than 1%.(3)Aiming at battery life prediction problem,two forecasting models were established from the perspectives of battery residual cycle number estimation and fading trend prediction with deep learning methods.The first model combine feature engineering with time series prediction model.Firstly,the life factor which is strongly related to the battery residual cycle life is extract.With the life factor sequence of a certain time scale as input and the current residual cycle number as output,the deep learning models are proposed.The results of the test set indicate that the average error of the CNN model is 36 cycles and the average error of the LSTM model is 22 cycles in the late period of battery recession.For the battery fading trend prediction problem,a sequence forecasting model based on encoder-decoder network structure is established.It has been proven that the average error of model is less than 5% and the forecast of accelerated recession trend can be achieved.Finally,this paper extend the application of the model in health index prediction which provides reference for realizing the industrial field of recession trend prediction.The proposed battery capacity evaluation method,battery health factor construction and filter method,battery SOH estimation method and life prediction method based on deep learning provide solutions to some key technical problems of lithium ion battery health management.The state of health estimation and residual life prediction of batteries can be realized by collecting some voltage interval data under actual operating condition,which has great engineering application and promotion value.