Li-S Battery Outlier Detection and Voltage Prediction using Machine Learning

State estimation is essential for enabling battery management systems (BMSs) to monitor, control, and optimize battery performance. The first step toward this is the ability to predict a battery's behavior given its input current and present state. In the early stages of new battery chemistry development, prior to commercialization, lab-fabricated battery cells might be used for characterization and BMS development. Such custom-fabricated batteries are often more prone to anomalies in their cycling behavior, including loss of connectivity and instantaneous internal shorts. The use of battery cycling data containing such anomalies can negatively affect modeling accuracy and system predictability. This paper uses the K-means clustering method to detect outlier patterns in battery cycling, thereby enabling the extraction of sanitized cycling data. A feedforward neural network is then trained to predict battery voltage one step ahead, given the input current and prior voltage history. The paper demonstrates this machine learning approach for cycling data from laboratory-fabricated lithium-sulfur (Li-S) cells. This demonstration highlights both the accuracy of the proposed voltage prediction algorithm and the degree to which the proposed outlier detection algorithm helps improve this accuracy. Copyright (c) 2023 The Authors. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/)