LAUSR

Accurate state of health estimation of lithium-ion batteries provides scientific basis for secure operation and stepwise utilization in on-board powertrain. However, the variable discharge depths inevitably reduce the elasticity and precision of the estimation method in prevalent partial discharge situations. In this work, multiple candidate health indicators are extracted from the peaks and valleys of the partial incremental capacity curves and screened first. Specifically, the fine-tuning process of deep belief network based on particle swarm optimization are elaborated and synthetic comparison in terms of error and time consumption with three classical deep networks is performed. To better accommodate practical scenarios, three datasets of the LiFePO$_{4}$ cells under different discharge depths are applied to verify the proposed framework. The experimental results indicated that the presented framework is feasible and the prediction error can be minimized to less 2%.