Investigation of Adaptive Neuro-Fuzzy Inference System for Battery Degradation on a Charge-Sustaining Hybrid Electric Vehicle

In this study, adaptive neuro-fuzzy inference system is investigated for battery degradation. Ten widely used drive cycles are selected to obtain a dataset for six different initial state-of-charge levels between 30%-80%. Torque demand, state-of-charge and battery capacity fade are used as the inputs of adaptive neuro-fuzzy inference system and internal combustion engine torque is selected as the output. Equivalent consumption minimization strategy is used as the dataset algorithm. Dataset size is increased by aging the battery to investigate the algorithm in further aging conditions. Four different datasets are used to investigate the algorithm for different aging conditions. Initial dataset size is increased by running each drive cycle in the dataset for 25, 50 and 100 times respectively. Results are obtained for Worldwide Harmonised Light Vehicle Test Procedure (WLTP) and New European Driving Cycle (NEDC). The WLTP case results for 100 cycles indicate 19.71% capacity fade reduction with 1.42% increase of fuel consumption. The NEDC case results for 100 cycles reveal a 23.31% reduction in capacity fade and 12.48% increase in fuel consumption..