State Of Health Estimation Of Lithium-ion Battery Based On Degradation Feature Extraction And Machine Learning Algorithms

In order to alleviate environmental pollution and energy shortage,electric vehicles with low energy consumption and zero emission have gradually become the main development direction of the automobile industry.As one of the most commonly used energy sources in electric vehicles,lithium-ion battery gradually ages throughout its working life cycle,and its ability in energy storage and power supply will obviously reduce,which significantly affects the performance of the vehicle.As one of the key parameters of battery state monitoring,state-of-health(SOH)is very important for ensuring the endurance stability and the safety of vehicle system.Therefore,it is of great significance to estimate the SOH of batteries accurately.In this paper,starting with the extraction of new degradation features that can effectively characterize the degradation of battery,the research on accurate SOH estimation methods for power lithium-ion battery is carried out.The specific work is as follows:1.A lithium-ion battery aging experiment platform which can provide variable working conditions and temperatures is designed and built.Based on the analysis of the factors affecting the aging of lithium-ion batteries,two aging test schemes under different test conditions are designed and implemented to fully obtain the aging data of lithium-ion batteries under different environmental temperatures and working conditions,so as to provide the data basis for the follow-up study of battery SOH.2.Novel degradation features for lithium-ion batteries are extracted.Considering that the degradation state parameters of the battery are difficult to obtain directly in the actual use process,the degradation features,which can represent the degradation law of the battery,are proposed to be extracted from the measurable parameters such as the current and voltage of the battery.(1)The fast sample entropy extraction algorithm is adopted to extract the degradation feature from battery discharge voltage in the capacity test stage.(2)The Thevenin model is established for battery modeling and the ohmic internal resistance is identified from the model.After that,the variational mode decomposition is proposed to separate out the trend component of the identified ohmic resistance.The average value of the trend component in[30%,80%] state of charge interval is calculated to be the standard ohmic internal resistance in the current aging cycle,and then the ohmic internal resistance increment is calculated as the other degradation feature.The linear fitting and correlation analysis between the two degradation features and battery capacity degradation are carried out respectively.The analysis results indicate that the extracted degradation features can reflect battery degradation effectively.3.In order to accurately estimate the health status of lithium-ion battery,the Gaussian process regression(GPR),which can accurately process small sample and high complexity data,is introduced to build an SOH estimation framework for lithium-ion batteries.In this framework,the extracted degradation features and the corresponding capacity degradation are used as the input and target output of the GPR respectively,and the GPR is fully trained for subsequent SOH estimation.An adaptive mutation particle swarm optimizer(AMPSO)is proposed to solve the optimization problem of the hyper-parameters in the kernel function of GPR.Moreover,considering the influence of the selection of kernel function on the estimation accuracy of the framework,the Matern kernel function is selected as the optimal kernel function by testing and comparing different kernel functions.The SOH estimation framework based on the AMPSO-GPR is compared with machine learning algorithms such as back-propagation(BP)neural network and support vector machine(SVM).The results show that the proposed framework can achieve the highest SOH estimation accuracy among the comparative algorithms.The SOH estimation for batteries at different temperatures is also carried out by the proposed framework.The SOH estimation errors are approximately 1.5% when the training data and test data are from the same kind of battery.These results verify the effectiveness of the proposed framework for battery SOH estimation.4.A novel SOH estimation framework combining the metabolic extreme learning machine(MELM)and gray error compensation is proposed to realize the accurate SOH estimation for different types of batteries at multiple temperatures.This framework makes full use of the powerful generalization ability of the extreme learning machine(ELM)to build a degradation state model,which reflects the relationship between degradation characteristics and capacity degradation quantity.The degradation state model is utilized to describe the general degradation law of different batteries,and the SOH estimation of battery full life cycle is achieved by repeatedly updating the degradation state model.In addition,in order to reduce the cumulative estimation error caused by the metabolic process,the grey model is proposed to correct the SOH estimation results by mining the potential changing trend of errors and predict the subsequent error.This SOH estimation framework is applied to SOH estimation for different kinds of batteries at four temperatures,and the maximum estimation error is no more than 1.57%.The results show that the proposed method can accurately estimate battery SOH under different conditions,and it has good universality and generalization ability.