The Monitoring System Of Battery SOH Based On Feature Optimization And Mixture Kernels Function SVM

In recent years, lead-acid batteries have played a main role as the energy storage equipment and power equipment, its use ranges from the aviation navigation,transportation, military communications, power systems and etc., and now gradually applies into people's lives in all aspects . Therefore, the status of the battery health (State Of Healthy is SOH) has obtained much more attention. However, due to the factors affecting the aging of lead-acid batteries vary a lot, and the battery aging test restricts by the full charge and discharge time and the number of samples, making a small sample based on a representative feature set in status of the battery health (SOH) prediction is particularly important. Therefore, based on the analysis of battery discharge characteristics, an SOH feature selection algorithm based on unsupervised ACCA-FCM(Ant Colony Clustering Algorithm-Fuzzy C-Means Algorithm) and supervised SVM-RFE(Support Vector Machine-Recursive Feature Elimination) is proposed. The algorithm uses the ACCA(Ant Colony Clustering Algorithm) to select the effective eigenvalue clustering center from the global feature set, and overcome the clustering center sensitivity and local optimal problem of FCM(Fuzzy C-Means Algorithm),and exclude redundant features based on the correlation between features. Furthermore, the SVM-RFE feature sorting algorithm is used to eliminate the non-critical interference (low-predictive) characteristics, and finally the optimal feature subset is obtained, avoiding the complete discharge process as well as ensuring the accuracy of the premise.The SOH prediction model of the battery based on support vector machine (Support Vector Machine is SVM) is trained by using the optimal feature subset composed of the initial characteristics of the discharge. In order to improve the prediction accuracy of SOH,the kernel function of SVM is optimized. The RBF(Radial Basis Function) local kernel function with strong classification ability is combined with the classical global kernel function polynomial kernel function, which makes the mixed kernel function SVM model equips with both classify and generalize ability. And the grid search algorithm is used to optimize the parameters of the mixed kernel function SVM model based on the ten-fold cross validation, to find the global optimal parameters with the highest classification accuracy, and obtain the SOH prediction model of lead-acid battery based on mixed kernel function SVM.Finally, obtain the real-time data such as the battery voltage, temperature, current and others sent by the underlying hardware platform in the battery wireless monitoring platform, achieve dynamic display and fault warning in the android, and dump the data with the MySQL database in server. Based on the two-way communication between Matlab and MySQL database, the battery discharge data is read. Moreover, the algorithm is analyzed on Matlab and the forecasting model of mixed kernel function SVM is trained,then store the optimal feature subset selection and the SOH prediction value to the MySQL database. At last, to realize the integrity functional, accurate predicted battery SOH monitoring system for the battery health status, to provide a reliable basis and accurate monitoring.The main work of this paper focuses on the following:(1) To combine the ACCA-FCM and SVM-RFE feature selection algorithm. The advantages of supervised clustering algorithm and unsupervised feature selection algorithm are complementary to each other, and the optimal feature subset selection of battery discharge process is completed.(2) To optimize parameter of mixed kernel function SVM. The RBF kernel function is combined with the polynomial kernel function into a mixed kernel function, and the grid function is optimized by grid search algorithm.(3) To build Matlab test platform for verify the algorithm base on feature optimization and mixed kernel function SVM of SOH prediction.(4) The monitoring system of SOH for lead-acid batteries is completed, and its function include real time data of lead-acid batteries acquisition, display, storage and SOH prediction.