Nonlinear model identification using soft neural networks

According to the normal fuzzy reasoning methodology [26], a soft neural network is designed to perform an adaptive compensatory fuzzy reasoning [25]. A soft learning algorithm is used to intelligently initialize all parameters to train the soft neural network efficiently. Compared with some frequently used methods, the soft neural network is capable of modeling a nonlinear system effectively. A lot of simulations for the gas furnace model identification have indicated that the soft neural network is an effective soft-computing system with the ability of discovering fuzzy knowledge from numerical data and applying trained fuzzy rules to predict complex nonlinear behaviors.