Battery State of Health Prognostic based on Neural Hierarchical Interpolation

The precise State of Health (SOH) estimation of Lithium-ion Batteries (LIB) is critical for analyzing the dependability and endurance of battery-powered systems. SOH estimate using data driven techniques has gained popularity in recent years. The approach envisions the possibility to harness battery information and also supports in cloud-based battery management techniques. In this proposed work, novel data driven SOH estimation approach was carried out and performance metrics including RMSE, MAPE, and coefficient of determination (R-square) were calculated, to evaluate the efficacy of the proposed approach. To build and validate the model, experimental data of LIB were taken from NASA website. The performance metrics of the proposed model is compared against similar deep learning and machine learning models to analyze the improvisation of accuracy. The proposed Linear-NHI model achieves the lowest error inclusive of RMSE as 0.004, MAPE as $\mathbf{0. 4 9 \%}$ and attains highest coefficient of determination as $\mathbf{9 5 \%}$ for $\mathbf{B 0 0 0 5}$ model. For B0018 model, the performance metrics are similarly evaluated: RMSE, MAPE and $R^{2}$ are obtained as $\mathbf{0. 0 0 6 5}, \mathbf{0. 7 5 5}$ and $\mathbf{7 7. 6 \%}$ respectively thereby securing improved accuracy and reliability in SOH estimation. The findings emphasise the potential of advanced modelling techniques and is helpful to improve the operational efficiency and maintenance of Lithium-ion battery.