Research On The Status Evaluation Of Lithium-ion Batteries Based On AdaBoost-Elman And MEEMD

Lithium-ion batteries are the latest generation of green high-energy rechargeable batteries.The batteries have many advantages,such as high energy density,low selfdischarge rate,long-cycle life,no memory effect,pollution free.They are widely used in transportation,aerospace,weapons,smart grid and electronic products and other applications.Lithium-ion battery is a very complex system,it is highly vulnerable to accidental failure,so it can only operate in a narrow safety zone.In order to optimize the battery operation,improve the battery life and ensure the safety of use,it is necessary to effectively manage and control the battery.The battery management system has an irreplaceable role in maintaining battery use.The battery's state of charge(SOC)estimation,state of health(SOH)and remaining useful life(RUL)prediction are collectively referred to as battery state assessment,and are the most important and critical part of the battery management system.The data-driven approach is based on data.This type of approach conducts modeling and prediction by mining the characteristics and implicit information of the data sequence,the accurate mathematical model and prior knowledge of the object system are not required.Therefore,it has become an important method in the state assessment of lithium-ion batteries.Based on the data-driven principle,this paper uses Ada Boost-Elman and MEEMD to study the SOC estimation,SOH and RUL prediction methods in the battery management system.Data-driven machine learning is an important method for SOC estimation modeling of lithium-ion batteries.A typical representative of this type of method is the neural network-based learning method.Aiming at the problems of low generalization ability,local minimization,low prediction accuracy and insufficient dynamics in the prediction process of a single feedforward neural network(BP neural network),this paper proposes a SOC estimation method for lithium-ion batteries based on Ada BoostElman algorithm.This method improved the accuracy of weak predictors by making full use of Ada Boost algorithm and the dynamic characteristics of Elman neural network,the combined strong predictor thus has strong generalization ability,prediction accuracy and dynamic characteristics.Compared with BP neural network and Elman neural network,Ada Boost-Elman neural network has high estimation accuracy and good dynamic characteristics,which provides a new approach for SOC estimation of lithium-ion batteries.Based on the data-driven method,the parameters related to battery aging are used as health indicators,which are divided into external parameters(voltage,current,temperature)and internal parameters(capacity,internal resistance)according to different health indicators.The relationship model between external parameters and SOH is too complex,and the internal parameters are difficult to be measured online,resulting in inaccurate predictive modeling methods.In addition,data-driven machine learning methods are very important for SOH and RUL prediction modeling of lithiumion batteries.Other methods,such as neural networks and support vector machines,do not factor in the local regeneration phenomenon in the battery capacity signal.The local regeneration phenomenon will affect the capacity degradation trend of lithium-ion batteries,further affect the prediction results.This paper chooses the Isobaric discharge time as health indicators that can be measured online,proposes the SOH and RUL prediction method based on the MEEMD and Ada Boost-Elman.This method can track the global degradation tendency and local regeneration tendency in the battery capacity degradation signal,and achieve a better diagnosis of the battery state of health.Compared with the prediction results of Ada Boost-Elman,this method can capture the time-varying degradation phenomenon of battery capacity degradation and reduce the influence of local capacity regeneration on SOH and RUL prediction.The validity of the proposed method for SOC estimation,SOH and RUL prediction of lithium-ion battery was proved by NASA lithium-ion battery dataset experiment verification and comparison of other methods.The proposed method provides meaningful reference for the design and development of battery management system in practical application.