An Evolving Quantum Fuzzy Neural Network for online State-of-Health estimation of Li-ion cell

Abstract

With the rapid advancement in the battery industry, more accurate and advanced state estimation methods are required to meet the performance requirements. The State of Health (SOH) estimation is performed in the battery management system (BMS), which provides the qualitative measure of the capability of a lithium-ion battery (LIB), in terms of capacity or internal resistance. Theoretically, the cell capacity is obtained by complete charge and discharge of the cell but in practical scenario, complete charge or discharge is never the case. To address this issue of dynamic discharge, this paper presents an evolving model-based SOH estimation, predicting the capacity fade of the cell extracted from the incomplete discharge conditions as in the case of dynamic driving scenarios. The evolving algorithm uses Neural Network, which features an interval fuzzy set, with conjectural jump positions. For better identification of overlaps between the classes, the quantum fuzzy set uses graded membership function. The number of rules are automatically adjusted and evolved, in the quantum fuzzy set using Decoupled Extended Kalman Filter (DEKF) for parameter estimation. The proposed method uses voltage, current and sampling time data to estimate the capacity, over a period of 600 charging-discharging cycles of Nickel Manganese Cobalt Oxide (NMC) chemistry batteries. The dynamic discharge voltage data is obtained from the periodic characterization tests and is used for predict the complete discharge voltage. The full voltage profile has been forecasted using Long-Short Term Memory (LSTM) network and the subsequent capacity has been estimated using evolving Quantum Fuzzy Neural Network (eQFNN) with an RMSE of less than 5% making it suitable for on-board applications. The results are simulated in MATLAB 2020b and are validated using experimental verification in Battery Testing Lab (BTL), IIT Delhi.