Research On Prediction And Sorting Technology Of Retired Lithium Batteries Based On Integrated Learning Algorithm

At present,the increasing energy shortage crisis and environmental pollution problems need to be solved urgently,and energy conservation and emission reduction have become an important content of energy policies of all countries in the world,so as to alleviate a series of problems such as huge resource consumption and environmental pollution caused by vehicle exhaust emissions associated with fuel vehicles.With the rapid growth of new energy vehicles and the increasing demand for batteries,a large number of retired power batteries are bound to be produced.It is one of the most effective treatment methods to sort and reuse retired batteries by step utilization.In engineering application,the cascade utilization can’t be achieved without the state evaluation or prediction of retired batteries.The state of charge and health are the important technical indicators of batteries,and also the key parameters for the cascade utilization and sorting of retired batteries.In view of this,this paper aims at the estimation of the state of charge,the estimation of the state of health and the consistent sorting and reuse of retired lithium batteries.The main work and innovations are as follows:Firstly,this paper fully reviews the existing methods of lithium battery state of charge estimation,state of health estimation and sorting into groups.Aiming at the inconsistency of retired lithium-ion power batteries,the same batch of retired lithium iron phosphate batteries is used as the research object,and a battery test platform is set up to carry out the aging experiment and mixed pulse power characteristic experiment of retired lithium batteries.The changes of voltage,capacity and internal resistance during the aging process of retired lithium batteries are analyzed,and a batch of single batteries that can be used for subsequent sorting are preliminarily screened,which is the research of state of charge estimation,state of health estimation and battery sorting for subsequent lithium batteries.Secondly,the state of charge estimation of retired lithium battery based on Ada Boost.Rt-RNN algorithm is studied.The traditional neural network method has the problems of narrow application range,weak generalization ability and low accuracy,and it is difficult to meet the problem of low SOC prediction accuracy caused by the dramatic change of discharge nonlinearity during the echelon reuse of retired lithium batteries.This paper proposes an ensemble learning algorithm based on Ada Boost.Rt recurrent neural network model to achieve accurate prediction of SOC of retired lithium batteries.This method uses a chain-connected recurrent neural network model to overcome the problem of correlation adaptability of sample data in spatial and temporal dimensions,and uses the integrated learning method to use the RNN model as the base learner to form the Adaboost.Rt-RNN strong learning model can improve the SOC prediction accuracy and improve the generalization performance of the model.The simulation results show that the Ada Boost.Rt-RNN model has high prediction accuracy in estimating the state of charge of retired lithium batteries.Thirdly,aiming at the problem that recurrent neural network can’t handle long sequences,an Ada Boost.Rt ensemble learning model algorithm based on LSTM is proposed to realize the joint estimation of SOC and SOH of lithium batteries.Cyclic neural network can give full play to the advantages of its feedback neural network and improve the prediction accuracy when predicting the short-term state change SOC of lithium batteries.However,RNN can only store partial sequences,and it is easy to produce large errors when dealing with long sequence dependence problems,such as health state estimation based on the whole life cycle of lithium batteries.Long-term memory neural network not only has the advantages of circulating neural network,but also can solve the problems of gradient disappearance and gradient explosion.In view of this,this paper uses Ada Boost.Rt-LSTM model algorithm to jointly estimate the SOC and SOH of lithium batteries.Simulation and comparison experiments show that the proposed Ada Boost.Rt-LSTM algorithm is effective,and the simulation results have high accuracy,which can provide effective parameters for battery sorting in the following.Finally,in order to solve the influence of inconsistency between retired lithium battery cells on battery group performance,this paper proposes a battery comprehensive characteristic sorting method based on mean shift clustering algorithm.The static and dynamic parameters of multiple batteries are selected as the input parameters of the decommissioned lithium battery sorting model.Through the analysis of the principal components of the sample data,the mean shift clustering algorithm is used to realize the comprehensive characteristic sorting of the single sample of the decommissioned lithium battery.The simulation and experimental results show that the capacity attenuation of retired lithium batteries is consistent,and the consistency between batteries is good,which verifies the accuracy and feasibility of the proposed sorting method.