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

Area of Science:

• Battery Health Monitoring
• Machine Learning for Prognostics
• Artificial Intelligence in Engineering

Background:

• Accurate prediction of remaining useful life (RUL) is crucial for the reliable operation and maintenance of lithium-ion batteries (LIBs).
• Traditional RUL prediction models often face challenges with accuracy and generalization across diverse battery datasets.
• The integration of advanced optimization algorithms with neural network architectures offers potential for enhanced predictive performance.

Purpose of the Study:

• To develop and evaluate a novel Jellyfish Optimization (JFO) based Feedforward Neural Network (FNN) model for improved RUL prediction of LIBs.
• To investigate the effectiveness of systematic data sampling and a multiple battery with multi-input (MBMI) profile for feature extraction.
• To compare the performance of the proposed JFO-FNN model against traditional FNN models using established LIB datasets.

Main Methods:

• Utilized a Feedforward Neural Network (FNN) model integrated with a Jellyfish Optimization (JFO) technique for hyperparameter optimization.
• Employed a multiple battery with multi-input (MBMI) profile to generate 91-dimensional features from LIB data.
• Applied a systematic sampling approach for relevant data feature extraction and Mean Square Error (MSE) as the objective function.

Main Results:

• The JFO-based FNN model significantly outperformed the traditional FNN model in RUL prediction accuracy.
• Achieved a low Mean Square Error (MSE) of 3.9494*10^-4 for LIB cell B5 on the NASA LIB dataset.
• Model performance was validated using particle swarm optimization and demonstrated high applicability on MIT-Stanford LIB datasets, though capacity regeneration issues affected specific LIBs (B6, B18).

Conclusions:

• The JFO-optimized FNN model offers enhanced predictive accuracy, generalization, and robustness for LIB RUL prediction.
• The proposed framework provides a fast and effective solution for battery prognostics, leveraging systematic data sampling and advanced optimization.
• The study highlights the potential of JFO-FNN for reliable battery health management and predictive maintenance applications.