| This dissertation focuses on the lithium-ion power battery. In order to explore the relationship between characteristic parameters of charge/discharge and state-of-health(SOH), the battery life test has been designed and implemented. The aim is to extract the characteristic parameters from battery vehiculary application, and to estabilish a model with characteristics parameters as input and SOH as output.Firstly, based on the analysis of operating principles and failure mechanisms of lithium-ion battery, the battery life test has been designed and completed in the laboratory, and then, the changes of performance during the life test has been studied. With the consideration of actual application on electric vehicle, the performance of constant current charging process is selected to analyze the relationship between change rules and SOH.Secondly, the calculation process of Capacity Incremental Analysis(ICA) is designed to extract the characteristic parameters of constant current charging process. By Wavelet Filtering, the capacity increment curve is filtered, and then two charactersitc parameters, peak height and peak position voltage, are extracted as health factors. With Gray Correlation Analysis, the correlation degree of two health factors and SOH is calculated.Thirdly, Because of the uncertainty of lithium-ion battery degradation process, a model whose output has the character of confidence level and confidence interval is reasonable. With health factor as input and SOH as output, Gaussian Process Regression which has the character of confidence interval is used to establish the model. By Conjugate Gradient, the Hyper Parameter of Gaussian Process Regression is caluated. The model performance indicators are put forward. Comparing three kinds of kernel functions, the effect of the distribution and step of training samples on the model is analyzed. The model can achieve good pefromance.Fourthly, in order to solve the problem that Conjugate Gradient can only locally solve optimization problem and ensure that Hyper Parameters can be achieved globally, Multi-island Genetic Algorithm is imported to establish the MIGA-GPR model. The effects of different health factors, different kernel functions and different steps of training samplesare analyzed.Finally, the model validation has been designed and completed, and the application of modeling method has been promoted.Under experimental conditions, the established model was validated.Based on the public data of power battery, applicability of MIGA- Gaussian process regression modeling method has been analyzed, and the results show that the modeling method is suited to SOH estimation on other types of lithium battery in effect. |
PDF Full Text Request