Battery pack state of charge estimator design using computational intelligence approaches

Abstract

This paper presents a novel design of a battery pack state of charge (SOC) estimator for electric vehicles using computational intelligence techniques. The main framework of the estimator is a three-layer feedforward neural network with four inputs and one output (estimated SOC). The inputs are the battery pack current, accumulated ampere hours, average pack temperature and minimum voltage of the battery modules. A strategy is developed to select the training data set from a large amount of the original testing data sets under different drive cycles and operating conditions. A modified particle swarm optimization (PSO) algorithm is used to train the proposed neural network. The designed SOC estimator is validated and evaluated using the testing data under different drive profiles and temperatures. The errors of the SOC estimates are well within the acceptable range compared to that obtained by using traditional mathematical models. The resulting SOC estimator is computationally efficient and can be easily implemented using low-cost microprocessors.