State of health estimation for Li-ion battery using characteristic voltage intervals and genetic algorithm optimized back propagation neural network

Abstract

The accurate estimation of lithium-ion battery state of health (SOH) is important for the battery safety condition and range. However, in most cases, the operating conditions of lithium batteries are highly random. The data length of each cycle during the actual use of lithium-ion batteries is highly random and does not satisfy the input conditions of existing SOH estimation methods. To deal with the randomness of battery data, this paper proposes a SOH estimation scheme based on BP neural network optimized by a genetic algorithm (GA-BP) and fixed characteristic voltage interval. The method selects the effective charging cycle data on random length data through fixed characteristic voltage interval as the model training set, constructs 3 different feature parameters on this training set, and trains the GA-BP model with it to achieve SOH estimation on random length data. Experiments with the Monte Carlo method on 2 publicly available datasets validated the accuracy of the method at different levels of randomization.