| In recent years,against the backdrop of fossil energy shortages and environmental degradation,the automotive industry is accelerating the transformation,upgrading and development from traditional fuel vehicles to clean and low-carbon new energy vehicles.Vehicle power batteries are the core power source of new energy vehicles and an important component of the vehicle’s overall cost structure.Accurate assessment and scientific prediction of their service life are of great significance for improving the safety,economy,reliability,and durability of new energy vehicles.This article focuses on the study of lithium-ion power batteries for pure electric vehicles,and addresses the needs and concerns of users regarding battery life prediction and cost anxiety in the process of promoting the industrialization and application of electric vehicles.To address technical challenges such as poor adaptability and interpretability of battery health characteristics extracted from battery data collected from real vehicles,the method of incremental capacity analysis is used to extract battery health characteristics from local charging segments.Based on these health characteristics,a neural network-based battery State of Health(SOH)estimation model is established,followed by the establishment of a Remaining Useful Life(RUL)prediction model for batteries using Gaussian Process Regression(GPR).This enables accurate prediction of the remaining useful life of vehicle lithium-ion batteries.The specific research contents of this article are as follows:Firstly,the basic structure,working principle,and aging mechanism of lithium ion batteries are briefly introduced.The internal reaction mechanism of lithium ion batteries during the charging and discharging process and aging process is analyzed from both theoretical and experimental aspects.Based on the open source lithium ion battery experimental dataset,the factors affecting battery capacity and battery aging were studied,providing theoretical support for subsequent battery health estimation.Secondly,the battery capacity was calculated using the data collected from operating pure electric vehicles online.Extract the battery data of the parking charging process from the data collected by the real vehicle and divide the charging data into segments.After processing the abnormal value and missing value of the charging data,the battery capacity is calculated based on the ampere-hour integral method;the box diagram method is used to eliminate the capacity outliers,and then the moving average filter method is used to denoise the capacity data to obtain a reliable battery capacity sequence.Then,real-time estimation of the health of lithium-ion batteries for vehicles is realized based on the data collected from real vehicles.The multi-dimensional health features that can represent the health of the battery are extracted from the real vehicle charging data,and used as the input of the neural network model to estimate the SOH.The relative error of the SOH estimation on the test set is within 4%.The SOH estimation method is verified based on the NASA lithium-ion battery experimental dataset.The results show that the Mean Absolute Error(MAE)and Root Mean Square Error(RMSE)of the battery SOH estimation on the test set are within 0.0042 and 0.0064,respectively.Finally,a prediction study was carried out for the remaining useful life of lithium-ion batteries for vehicles.Taking the vehicle mileage as the battery RUL indicator,a RUL prediction model is established based on the Gaussian process regression algorithm.The model takes the mileage as input and SOH as output to perform long-term prediction of battery SOH,and finally obtains the RUL prediction result.The model was verified based on real vehicle data,the Gaussian process model was trained using real vehicle historical mileage and SOH data,the GPR kernel function was configured according to the characteristics of the real vehicle battery SOH degradation curve,and used the conjugate gradient descent algorithm to solve the maximum likelihood estimation of the marginal function to obtain the optimal model parameters.The relative error between the final battery life prediction value and the actual life is within 10%,and the battery RUL single point prediction result can be calculated based on the predicted battery life.In addition,the RUL prediction algorithm was verified using the NASA data set and the Oxford data set respectively.The results showed that the relative error between the predicted battery life and the real life at 70% of the battery life cycle was within 10%.This paper proposes a method based on the combination of SOH estimation and RUL prediction based on the battery charging data collected by real vehicles.Predict its RUL.Compared with the traditional method based on laboratory data,the method based on real vehicle battery data does not require high data set quality,and can adapt to complex battery operating conditions,so it is closer to the actual application of power batteries;in addition,the method is based on Compared with the number of cycles,the driving mileage is more feasible and practical as a battery life indicator.This paper proposes a method based on the combination of SOH estimation and RUL prediction based on the battery charging data collected by real vehicles.Predict its RUL.Compared with the traditional method based on laboratory data,the method based on real vehicle battery data does not require high data set quality,and can adapt to complex battery operating conditions,so it is closer to the actual application of power batteries;in addition,the method is based on Compared with the number of cycles,the driving mileage is more feasible and practical as a battery life indicator. |
PDF Full Text Request