A Parameters Self-adjusting ANN-PI Controller Based on Homotopy BP Algorithm

Abstract

Back-Propagation neural network can record the fuzzy control rules efficiently, and utilize these experiences according to associative memory. But, all existing feedforward net learning algorithms have a local minimum problem inevitably. To solve above problem, a Homotopy continuation BP algorithm is adopted in this paper, which provides an effective method for BP network’s global convergence and is of very fast convergent speed. For some complex nonlinear control systems, a parameters self-adjusting fuzzy-PI controller is ever adopted effectively. Because ANN has strongly nonlinear mapping power, so we can use ANN based on Homopy BP algorithm to replace fuzzy segment to reconstruct a new ANN-PI controller, which has a faster dynamic response, higher control accuracy, better disturbance-resisting ability, less sensitive to parameter changes, and robustness.