Hybrid optimized remaining useful life prediction framework for lithium-ion batteries with limited data samples

TL;DR

This paper proposes a new method for predicting the remaining useful life of lithium-ion batteries using an optimized neural network model.

Contribution

The novelty is integrating the Jellyfish optimization technique with a neural network for improved battery life prediction.

Findings

The JFO-based FNN model outperformed traditional FNN in RUL prediction accuracy.

The model achieved an MSE of 3.9494*10− 4 for LIB cell B5.

The MIT-Stanford LIB datasets validated the model's high applicability.

Abstract

This study introduces a Jellyfish optimization technique integrated with a Multi-Layer Perceptron, specifically a Feedforward Neural Network (FNN) model, for remaining useful life (RUL) prediction of lithium-ion batteries (LIBs). A multiple battery with multi-input (MBMI) profile is utilized to create 91-dimensional data features for model training. A systematic sampling approach is employed to extract relevant data features. Results show that the proposed JFO-based FNN model outperforms the traditional FNN model’s accuracy. The Mean Square Error (MSE) is used as the objective function to determine optimal model hyperparameters. The research utilizes the NASA LIB database, which includes four datasets. For LIB cell B5, the proposed model achieved an MSE of 3.9494*10− 4. The model’s accuracy and efficiency are further validated using particle swarm optimization. However, the LIBs B6 and B18 showed higher error results due to capacity regeneration issues. The MIT-Stanford LIB datasets demonstrated high applicability when validating the JFO-FNN model’s outcomes. The novelty of this work lies in using a JFO-optimized FNN model trained on systematically sampled, multi-battery LIB datasets to improve predictive accuracy, generalization, and robustness. Overall, the developed RUL prediction framework appears to be fast, effective, and yields promising results.

Figures