ANN Based Modelling of Optimal Passive Cell Balancing

Abstract

Batteries cost nearly 30-40% of overall vehicle cost, making them the most crucial part of electric vehicles. Among different types of batteries, Li-ion batteries are mostly preferred because of their high energy as well as power density, low self-discharge rate, no memory effect, large life cycles and low cost. Batteries are the combination of many cells connected in series and parallel depending upon the voltage and current ratings. Each cell differs in its cell behaviour, creating cell imbalance. This unbalancing needs a balancing circuit to properly utilize the cell’s capacity. In this paper, Multi-layer Feed Forward Neural Network-based passive cell balancing has been done, and used ANN model is trained by Levenberg-Marquardt backpropagation algorithm using Neural network toolbox of MATLAB. An analytical method for bleed resistor calculation considering the parameters such as maximum power loss, balancing time and voltage difference has been proposed. Equalization of the SoC level shows the successful completion of cell balancing. In the ANN-based model, cell voltages show no ripples during balancing than in the Logic-based Stateflow model. During the change of bleed resistor range, ANN-based models again show no ripple in battery voltage and current.