Integration of topological data analysis and entropy features for the prediction of lithium-ion battery state-of-health

Abstract

Lithium-ion batteries are now widely used in many devices due to their high performance. However, their use in electric vehicles poses challenges such as limited driving range and variable cycle life. A data-driven approach can be useful to better understand the aging mechanisms of batteries. Predicting a battery’s State-of-Health (SoH) accurately is crucial to improve battery technology/life. A machine learning algorithm, combined with features extracted from voltage/temperature during a charging cycle, can be used to predict the SoH of a battery. Creating new feature vector by integrating known features which can help machine learning (ML) algorithms in solving a particular problem will be a novel work. A Topological Data Analysis (TDA) technique and entropy features are utilized to create a feature vector that can predict the SoH of a battery through a machine-learning model called long short term memory neural networks. The TDA features represent the functions of one- and two-dimensional holes in the data which appear/disappear as a tolerance value is increased. The entropy features represent the amount of information present in the dataset. There are multiple ways to define entropy of a time series data. In this article, we carefully selected entropies suitable for the battery datasets. Oxford battery degradation dataset, which is publicly available, was used to apply a Long Short-Term Memory (LSTM) model. The average Mean Absolute Error (MAE) of the model with topological data analysis features is 0.02045 (2.56 %), while the average MAE of the model with entropy features is 0.02241 (2.77 %). However, the average MAE of the model with integrated entropy-TDA features is only 0.02025 (2.54 %). The low MAEs in the models suggest that the feature set created by integrating topological-entropy features will be helpful in predicting the SoH of a Lithium-ion battery.