Research On State Estimation And Remaining Useful Life Prediction Of Lithium-Ion Batteries

Faced with increasingly urgent energy depletion and environmental pollution issues,countries have vigorously supported the new energy vehicle industry,thus electric vehicle technology has received widespread attention from many researchers.Lithium-ion batteries have been widely used in devices such as electric vehicles,mobile phones,airplanes,and satellites due to their high power and energy density,long cycle life,and low self-discharge rate.The state of charge(SOC),state of health(SOH),state of energy(SOE)estimation,and remaining useful life(RUL)are important parameters for battery systems,which directly affect the safety status of related equipment.However,batteries are a nonlinear system with complex internal chemical reactions,and obtaining the aforementioned battery states is very challenging.This article focuses on lithium-ion batteries and uses meta heuristic optimization algorithms,Kalman filtering algorithms,machine learning,and deep learning algorithms to study battery status.The main achievements achieved are as follows:(1)Introduce fractional calculus theory to establish a first-order RC battery model,and use nonlinear functions,adaptive weight coefficients,and Gaussian mutation improved grey whale optimization algorithm to identify the model parameters.A dual scale adaptive multi-innovation fractional order square root unscented Kalman filter-square root Kalman filter algorithm is proposed for model parameter updates,battery SOC,and capacity estimation.After adding different amplitude drift values to the current and voltage,the algorithm can still achieve high estimation accuracy and robustness in dynamic operating conditions at different temperatures.The maximum MAE of SOC is 3.829%,the maximum MAE of voltage is 45.313 m V,and the maximum MAE of capacity is 0.493%.(2)To address the issues of significant differences in battery model parameters under different operating conditions,inconsistent SOC OCV relationships at different temperatures,and poor estimation performance under noisy conditions,a gated cyclic unit neural network with an activation layer function layer is proposed to establish a battery model for SOC estimation.The experimental results indicate that the proposed network model does not rely on the battery model,and the trained network model can achieve online SOC estimation under different operating conditions.Despite the presence of noise in the measurement data,it still has high estimation accuracy and robustness,and the MAE and RMSE of SOC estimation results are both within 2%.(3)A transfer learning scheme is proposed to estimate battery SOC and SOE for data-driven methods that require a large number of datasets.Firstly,multi-scale convolutional networks are used to extract features from different channels,and attention mechanisms are used to enhance useful channel features to improve the estimation accuracy of the model.Then,the unscented Kalman filtering algorithm is used to further optimize the battery SOC and SOE output from the deep learning model,resulting in more accurate and smooth estimation results.Train a deep learning model using source domain data to estimate the SOC and SOE of other battery models through transfer learning.In the case of a small amount of data in the target domain,the method proposed in this paper can still achieve good accuracy in battery SOC and SOE estimation.(4)To address the issue of low accuracy in battery SOH estimation using some machine learning algorithms,this chapter proposes an improved grey whale optimization algorithm to optimize the online extreme learning machine with variable forgetting factor and particle filter algorithm for battery SOH estimation and RUL prediction.Firstly,features are extracted from the data during the charging phase,and extreme random trees are used to rank the importance of data features.Using the improved grey whale optimization algorithm to optimize the parameters of an online limit learning machine with variable forgetting factors,in order to obtain accurate battery SOH.Conduct simulation experiments on adding Gaussian noise to input features,combining different features,and comparing with other methods.The results show that the proposed method can achieve high accuracy in battery SOH estimation and has a certain degree of noise anti-interference ability.Finally,the grey whale optimization algorithm is used to optimize particle filtering and complete the prediction of battery RUL.The results of long-term battery RUL prediction can provide important reference basis for battery life prediction and fault diagnosis.(5)To address the issue of noise interference in voltage and current,the CVVariable model is used to reconstruct the voltage curve during the constant current charging stage,and the limit learning machine algorithm is used to reconstruct the current curve during the constant voltage stage.The reconstructed voltage curve can make the capacity increment curve smoother and easier to extract peaks,peak positions,and peak areas.Using a hybrid model of time convolutional network and bidirectional gated recurrent unit neural network to achieve SOH estimation on different battery datasets.The SOH estimation results of the PF model battery indicate that the hybrid network model still has good estimation results even under changes in environmental temperature.Finally,analyze the impact of feature combinations within different voltage windows,different correlation feature combinations,and early multi-step estimation on battery SOH estimation.Estimating the RMSE of SOH 25 steps in advance can be controlled within 2.6%,which can improve the efficiency and reliability of equipment,and reduce energy waste and safety risks.