Dynamic model of a lithium-ion cell using an artificial feedforward neural network with dynamical signal preprocessing

Abstract

Abstract A common approach in lithium-ion cell modeling is based on an electrical equivalent circuit model. The advantage of the electrical model is its simple structure. However, estimation of its parameters usually requires several steps in post-processing of the measured data in order to achieve satisfying accuracy. In this paper, we propose a solution with an artificial feedforward neural network and dynamical signal preprocessing, which does not require complex estimation procedures and has good accuracy. In contrast to the common conviction that a neural network requires many tests in a training data set, we show that only a few tests are enough to train the neural network. In the paper, we present practical aspects of the training process including methods to overcome obstacles related to measurement inaccuracy. Finally, the results of the artificial neural network model are validated and compared with those from an electrical equivalent circuit model.