State of charge estimation based on evolutionary neural network

Abstract Based on clonal selection theory, parallel chaos immune evolutionary programming (PCIEP) is presented. Compared with classical evolutionary programming (CEP) and evolutionary algorithms with chaotic mutations (EACM), experimental results show that PCIEP is of high efficiency and can effectively prevent premature convergence. A three layer feed forward neural network is designed to predict the state of charge (SOC) of Ni–MH batteries. Initially, partial least squares regression (PLSR) is used to select input variables. Then, five variables, battery terminal voltage, voltage derivative, voltage second derivative, discharge current and battery temperature, are selected as the inputs of the neural network (NN). In order to overcome the weakness of the back propagation (BP) algorithm, PCIEP is adopted to train the weights. Finally, under the state of a dynamic power cycle, the estimated SOC from the NN model and the measured SOC from experiments are compared, and the results confirm that the proposed approach can provide an accurate estimation of the SOC.