Snapshot SoC identification with Pulse Injection Aided Machine Learning

Abstract

Machine learning, in the form of a fully connected feedforward network, is applied to predict the state of charge (SoC) of lithium ion batteries injected with a series of pulses applied at different SoCs and states of health (SoHs). A snapshot of the normalized voltage response to these pulses is the only required input. Neither previous data nor Coulomb counting are needed. The Pulse Injection Aided Machine Learning (PIAML) algorithm is able to predict the SoC to better than 1% error on average for fresh, unaged cells and to below 2% error on average for a dataset of both fresh and aged cells. It provides SoC estimates without the need for rest periods, knowledge of capacity, or other equivalent parameters found in other methods. This algorithm can be used as a standalone estimator or as a periodic adjuster to other SoC estimation methods whose results may drift over time. PIAML is validated with constant discharge and drive cycle data.