Optimized Data-Driven Network for Remaining Useful Life Prediction of Lithium-ion Batteries

Predicting the remaining useful life (RUL) of a lithium-ion battery (LIB) is important for assessing battery reliability to identify the potential for failure and minimize battery risks. The current RUL prediction techniques deliver inaccurate outcomes due to the inappropriate utilization of battery parameters, especially the capacity degradation profile. In this work, a novel Jellyfish optimization (JFO) technique based on Feedforward Neural Network (FFNN) is proposed to evaluate the RUL of LIBs. A multi-channel multi-input (MCMI) profile is applied for developing a 91-dimensional data features to train model. The extraction of suitable data features was conducted with a systematic sampling (SS) technique. The outcomes suggested that the proposed JFO-based FFNN delivers accurate results compared with the conventional FFNN model. The Mean Square Error (MSE) was considered an objective function for obtaining suitable model hyperparameters. The NASA battery database consisting of 4 battery datasets namely B5, B6, B7 and B18 is considered for conducting the proposed research. For battery B5, it is observed that the proposed JFO-FFNN model delivered MSE as 3.9494*10-4 when trained with a training ratio of 70:30. Furthermore, batteries B6 and B18 delivered less accurate results due to capacity regeneration. Overall, the developed JFO-FFNN model is fast, effective and delivers satisfactory outcomes.