SOH Estimation And Failure Prediction Method Research For Power Batteries

Due to the rapid development of auto industry in recent years, the energy crisis and environmental pollution have become serious problems. All kinds of new energy vehicles, especially pure electric vehicles have been developing in order to solve these two problems in our country. As the power source of pure electric vehicles, battery performance directly affects the performance of electric vehicles: its capacity and internal resistance will gradually depredate, which affects the service life of power source. Estimating the state of health(SOH) of batteries can take timely measures to protect the battery, prolong the service life of power battery, and ensure the safety and reliability of equipment and prohibit serious accidents. This paper applies different methods to estimate the battery SOH according to different conditions based on laboratory data and battery online data.First, this paper introduces several kinds of power battery and their application and the SOH definition of power battery, analyzes the causes of battery capacity fade, and compares the advantages and disadvantages of various prediction methods in battery SOH predictions.Secondly, this article uses the data driven method of support vector machine(SVM) to estimate the battery SOH based on battery charge and discharge cycle data in laboratory. The C-SVR algorithm is used, which introduces penalty parameter C and slack variable factor ξ based on hard ε-SVR algorithm. Cross validation method is used to optimize the parameters in the model. Results show that SVM can estimate the battery SOH with error limitation.In order to obtain the confidential zone of estimation, Gaussian process regression(GPR) is adopted to estimate the battery SOH. The mean function and the covariance function as well as the likelihood function need to be determined in the estimating process; the key problem is the selection of the covariance function. This article proposes a selected method of the covariance function in the GPR based on the least squares curve fitting. Estimation results show that the GPR can predict the battery SOH more accurately, and acquire the confidential zone of the prediction.Finally, this article analyzes battery performance variation characteristics using knowledge based method, estimate battery SOH combined with GPR through online DC resistance. The experiences based method and the sample entropy method is used to estimate the SOH for the nickel cadmium and nickel-metal hydride battery respectively. The results show that the method based on knowledge will get the tendency of battery internal resistances. Empirical method can prognosis the battery potential failures, and give the health status of battery. The sample entropy method can calculate the complexity of battery parameters that reflect the health status of the battery.