An Improved SoC Estimation Algorithm Based on Artificial Neural Network

The state of charge(SoC)´s real time estimation plays an essential role in effective energy management, and has great significance to efficient operation and safe of electric vehicles(EV). Many methods, such as current integration, open circuit voltage, support vector machine(SVM), Kalman filter, artificial neural network(ANN) and so on, are used to estimate SoC, but these methods don´t work perfectly. To ANN and SVM methods, traditional methods don´t fit diverse work situation and must take some error when current change acutely. This paper put forward a new training set and voltage correction algorithm to improve above problems. The traditional method was tested by different experimental data, and root-mean-square-error(RMSE) is 7.14%. After the ANN model was trained by a new training set, the RMSE is 2.50%. In the last, voltage correction algorithm decrease the RMSE to 1.36%.