Modelling of a battery pack for electric vehicles using a stochastic fuzzy neural network

A dynamic simulation model for a battery pack is an important prerequisite for the simulation of electric vehicles. As the battery pack is a highly non-linear system, and their dynamic behaviour depends on different parameters, it is difficult to establish the relationship between the load voltage and the current under different temperatures and states of charge. Furthermore, the input and output data Usually contain noise; therefore, the traditional neural-netwrork-based model is affected while training the parameters. To solve the problem, a stochastic fuzzy neural network (SFNN) which has a filtering effect on the noisy input is used to model the battery non-linear dynamics in this paper. In the parameter-learning algorithm for the SFNN, a novel cost function which contains the error variables is studied. Then, the modelling test is performed on an 80 A h nickel-metal hydride (Ni-MH) battery pack and the SFNN-based model for the Ni-MH battery pack is set up. The Federal Urban Driving Schedule cycle is performed to test the model. Compared with the traditional neural-network-based model, the SFNN model can simulate the battery dynamic better and has a filtering effect on the noisy input which is more suitable for practical applications.