State of Charge Imbalance Classification of Lithium-ion Battery Strings using Pulse-Injection-Aided Machine Learning

Abstract

Lithium-ion battery strings are important modules in battery packs. Due to cell variation, strings may have imbalanced state of charge levels, reducing pack capacity and exacerbating degradation. While much research has been devoted to individual cells, string diagnostics using pulse-injection-aided machine learning can reduce sensing requirements and simplify computations. Experimental voltage response data from pulse perturbation of battery cells is used to generate virtual cell strings and ‘design’ the state of charge imbalance within the string. A feedforward neural network is trained on thousands of unique virtual string voltages and can distinguish between the balanced and imbalanced strings with up to 95% accuracy. Verification is performed using different string configurations and state of charge levels. The proposed technique has high promise and could be used to localize or regress the degree of imbalance.