A novel approach for prognosis of lithium-ion battery based on geometrical features and data-driven model

Abstract

Lithium-ion (Li-ion) batteries are widely used in such devices as today’s electrical vehicles, consumer electronics, and unmanned aerial vehicles, and will play a key role in the future. Unexpected Li-ion battery abnormities may result in serious inconvenience and enormous replacement costs. Thus, the diagnostic and prognostic methods play important roles in battery replacement scheduling, maintenance strategy development, and battery failure precaution, However, many published methods are unsuitable for both battery capacity and end-of-discharge. In this paper, a hybrid ensemble approach, integrating k-fold cross validation (k-CV) and genetic algorithm with back-propagation neural network (GA-BPNN), is proposed for capacity and end-of-discharge of Li-ion battery prognostics combined with geometrical features. Geometrical features extracted from charge-discharge cycles of Li-ion batteries are set as the inputs of the neural network. K-fold cross validation is introduced to determine the number of BPNN hidden layer neurons, genetic algorithm is used to initialize and optimize the connection weights and thresholds of BPNN. By the critical geometrical feature extraction and the ensemble BPNN model with k-fold cross validation and genetic algorithm, accurate battery capacity and end-of discharge are accomplished, making the proposed model can potentially be used for real-time estimate for the conditions given in this article. The performance of the proposed approach is demonstrated by using actual Li-ion battery data, which is supplied by the NASA Ames Research Center database.