Online Hybrid Prognostic Health Management Prediction Using a Neural Network and Smooth Particle Filter for Lithium-ion Batteries

Accurate real-time prognostic health management (PHM) estimation is essential to lithium-ion battery safety and efficiency. Recent work on developing a framework to predict remaining useful life (RUL) has primarily focused on the traditional empirical degradation model due to its simplicity. Although this model works well under specific operational conditions, for online RUL prediction it may lack the ability to describe capacity degradation, given the variability in decline between cells and others under different operational conditions. As such, this can result in inaccurate RUL prediction. Therefore, this work proposes a hybrid approach to improve the accuracy of online forecasting in the existing framework by integrating data-driven and model-based approaches. The proposed framework utilises the neural network (NN) to model and track battery degradation trends, and it also degrades the initial values of the degradation model’s transactions under different operating conditions. Furthermore, the proposed hybrid framework includes smooth particle filter (SPF) algorithm, which continuously updates the degradation NN model. Lithium-ion battery capacity degradation datasets from the Centre for Advanced Life Cycle Engineering (CALCE) were used to evaluate the proposed paradigm. The results show that the proposed hybrid framework is more accurate and improves the convergence rate compared to the traditional capacity prognostic framework