From data to durability: Evaluating conventional and optimized machine learning techniques for battery health assessment

Abstract

In the electronic era, the demand for efficient storage systems has rapidly increased, making the health and durability of batteries crucial. This research investigates the performance of distinct Machine Learning (ML) techniques—namely, Logistic Regression (LR), Convolutional Neural Network (CNN), and CNN performance tuning using Particle Swarm Optimization (PSO)—for Battery Health Analysis (BHA). The dataset comprises various parameters related to battery health, with Remaining Useful Time (RUL) as the target variable. The proposed work is evaluated using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R-squared (R2) scores. Initially, the basic LR Model is employed for BHA, followed by the CNN Model to capture complex data patterns. Subsequently, the CNN Model's performance is optimized using the PSO algorithm, aiming for improved performance. Experimental results demonstrate that the CNN Model significantly outperforms the LR approach in terms of accuracy, lower RMSE and MAE, and higher R2 scores. The conventional CNN model significantly outperformed the LR approach, resulting a lower RMSE of 20.11, MAE of 15.26, and higher R2 score of 0.996; whereas, the PSO-Optimized-CNN further enhanced the performance metrics with RMSE of 14.97, MAE of 8.03 and R2 score of 0.998. Henceforth, the PSO-optimized CNN Model exhibits further improved performance metrics compared to the standalone CNN Model. The findings offer valuable insights into ML approaches for BHA and suggest methods for optimizing battery management systems in various applications, including renewable energy systems, electric vehicles, and portable electronics.