Abstract: To accurately reflect the nonlinear function approximation abilities of improved algorithms of six typical BP networks, this paper elaborates on improved algorithm learning processes of the six typical BP networks. And the training function of MATLAB toolbox is briefly introduced which is used for BP network design. Finally a specific network is designed on MATLAB platform to conduct approximation test for a given nonlinear function. At the same time, a comparison between the performance differences of the six BP networks is made. The simulation result indicates that for a small scaled network, LM optimization algorithm has the best approximation ability, followed by quasi-Newton algorithm, conjugate gradient method, resilient BP algorithm, adaptive learning rate algorithm and momentum BP algorithm.

Key words: BP neural network, improved algorithm, function approximation, MATLAB