A Real-Time Adaptive Machine Learning Charging and Neural Network Balancing Mechanism of Lithium-Ion Battery Pack

Abstract

In this article, a real-time novel adaptive deep neural network (A-DNN) charging scheme is proposed which increases the life of the batteries by controlling the heating impact inside the battery. The input variables used in the charging algorithm are state of charge (SoC), state of health (SoH), voltage (V), current (I), and temperature (T) which makes the algorithm adaptive toward the temperature deviation and reduces the peak overshoot of the temperature at different SoH of the batteries. The parameters of the battery 1-RC model are estimated by the forgetting factor recursive least square (FF-RLS) method. The SoC and SoH are estimated by the dual-particle filter (D-PF) algorithm. Furthermore, a DNN balancing mechanism sensitive to SoC and SoH is developed to avoid the fault in the battery during the charging process. The A-DNN charging algorithm is compared with the constant current constant voltage (CC-CV), constant current pulse charging (CC-PC), and deep neural network (DNN) charging algorithms at 40°C, 45°C, and 50°C. The A-DNN outperforms in terms of peak temperature, incremental life, and charging time of the batteries at 45°C. The proposed charging methodology reduces the economic cost of the EVs by increasing the life of the battery by 34.41% at 45°C as compared to the other algorithms.