Integration of Gradient Descent Optimized Polynomial Regression with Neural Networks in Remaining Useful Life of Lithium ion Batteries

This research explores both Batch Gradient Descent and Stochastic Gradient Descent to optimize polynomial regression models at degrees 2 and 3, assessing the utility of this method in estimating the Remaining Useful Life (RUL) of lithium-ion batteries. After conducting an investigation, it was determined that the degree 2 polynomial regression optimized by Batch Gradient Descent had the maximum accuracy. As a result, it was chosen to be combined with Fully Convolutional Networks (FCN) in order to improve RUL prediction. Using Polynomial Feature Transformation and degree 2 polynomial regression, the 6,000 record dataset produces an R-squared (R) score of 0.9550 and a Mean Squared Error (MSE) of 0.2177.On the other hand, an MSE of 0.0741, an Explained Variance Score of 0.9847, and a R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0.9847 are obtained by the combination model of Batch Gradient Descent polynomial regression and FCN using a simple average. In comparison, the merging of the two models yields an impressive improvement in the accuracy of RUL forecasts for lithium-ion batteries, with an MSE decrease of around 66.1 percent from 0.2177 to 0.0741, and an improvement in R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of nearly 3.1 percent.