Enhancing Battery Prognostics Modelling with Digit Frequency Preprocessing Analysis

Abstract

Abstract To improve state-of-health (SoH) and remaining useful life (RUL) prediction in battery prognostics, a novel preprocessing method is developed that validates the input data integrity before undergoing a deep learning prognostic framework. Many of the developed deep learning models depend on a robust SoH estimation, however measured battery data are still subjected to faults stemming from physical defects to errors produced when the measured variables are logged. Hence, the preprocessing approach developed is capable of identifying anomalies such as outliers and errors within a subset of datasets. Filtering bad data from being utilized in machine learning algorithms can reduce inaccurate predictions caused from reinforced biased & skewing and inevitably reduce unwanted failures.