State-of-Charge Estimation of Li-Ion Battery in Electric Vehicles: A Deep Neural Network Approach
Dickshon N. T. How,Mahammad A. Hannan,Molla S. Hossain Lipu,Khairul S. M. Sahari,Pin Jern Ker,Kashem M. Muttaqi
DOI: https://doi.org/10.1109/tia.2020.3004294
IF: 4.079
2020-09-01
IEEE Transactions on Industry Applications
Abstract:The state of charge (SOC) is a crucial parameter of a battery management system for Li-ion batteries. The SOC indicates the amount of charge left in the battery of electric vehiclesakin to the fuel gauge in combustion vehicles. An accurate SOC knowledge contributes largely to the longevity, performance, and reliability of the battery. However, the SOC of Li-ion batteries cannot be easily measured by any apparatus. Furthermore, the SOC can also be influenced by numerous incalculable factors such as battery chemistry, ambient environment, aging factor, etc. In this article, we propose an SOC estimation model for a Li-ion battery using an improved deep neural network (DNN) approach for electric vehicle applications. We found that a DNN with a sufficient number of hidden layers is capable of predicting the SOC of the unseen drive cycles during training. We developed a series of DNN models with a varying number of hidden layers, and its training algorithm was to investigate their respective performance when evaluated on different drive cycles. We observe that the increasing number of hidden layers in the DNN (up to four hidden layers) decreases the error rate and improves SOC estimation. An additional increase in the number of hidden layers beyond that increases the error rate. In this study, we show that a four-hidden-layer DNN trained on Dynamic Stress Test drive cycle is capable of predicting SOC values unexpectedly well of other unseen drive cycles such as Federal Urban Driving Schedule, Beijing Dynamic Stress Test, and Supplemental Federal Test Procedure, respectively.
engineering, electrical & electronic, multidisciplinary