DC TO DC CONVERTER WITH NEURAL NETWORK CONTROL FOR ON-BOARD ELECTRICAL ENERGY MANAGEMENT

摘要

Neural network (ANN) methodology is proposed to control DC/DC static converters. These converters are used to adapt the voltage levels and currents between sources in parallel (battery, ultracapacitors and fuel cells) and loads (DC bus bar) in an electric or hybrid vehicle. The update of the parameters is carried out using the method of Levenberg-Marquardt; the training and the validation of the network by the model assumption NARX (Nonlinear Autoregressive with exogenous input). The system is split in two parts: the first is composed of a battery (12V), a DC/DC boost converter and a toad (DC bus bar 42V). The second consists on a pack of 8 ultracapacitors in series, a two quadrants DC/DC buck-boost converter and the load (DC bus bar 42V). The system is designed for power up to 10 KW on the level DC bus bar. Mathematical models for the battery, the ultracapacitors, the two converters, the classical command and the neural command, are developed using MATLAB/SIMULINK~® software. Comparisons with experimental results are presented in order to validate these models. Simulations give very satisfactory results in term of stability.