Feedforward neural networks using RPROP algorithm and its application in system identification

By comparative study of some typical improved algorithms of Back Propagation (BP) algorithm, this paper points out that most improved algorithms are difficult to use because the computational complexity is too depending on concrete application in a wide. range. Moreover, through analysis combined with experimental research, a good method (RPROP) of partly self-adapting learning rate has been brought forth, which has been testified to have the qualities of currency, fleetness and good robustness learning. We have got many satisfactory results since we applied this method to the identification of process control objects.