'Enhanced Capacity Predictive Model of Li-Ion Batteries' by Khyati Shukla

Shukla, Khyati

Abstract

As a result of its high voltage, power and better performance at low temperature, Li-Ion batteries are widely used in numerous electrical or electronics applications. One of the significant challenges for adequate battery power consumption lies in accurate capacity prediction. Another term that is necessary to evaluate is the End-of-life criteria. Generally, this is measured as the battery's energy reduces to around 80% of its beginning-of-life or the factory settings. At this point, the battery must be replaced to ensure uninterrupted operation in the application it is being used. These days machine learning algorithms are popular. They require a lot of data to process and provide a predictable outcome. Data-based predictive models are quick to deploy and give accurate capacity predictions, and thus, used extensively. However, for data-based prediction models, a fine selection of features (measurements) from the battery plays a crucial role in achieving the desired performance and accuracy.

In this project, an improvement in feature selection of an existing machine learning model for capacity prediction is proposed and validated using different machine learning algorithms (FNN, CNN, LSTM). The existing model has the feature selected limited to individual values of measured (battery terminal) voltage, current and temperature. The difference of respective voltage and current values at the battery supply source and the battery terminal is the modification considered in the new model. The modified model provides a better result as it considers both source and terminal values as a reference and maintains the uniformity of measurement.

• Online Publication Date: Jun 22, 2022
• Publication Date: Jun 22, 2022
• Deposit Date: Feb 20, 2025
• Collection Date: Jun 22, 2022