Trainable fuzzy and neural-fuzzy systems for idle-speed control

The authors describe the use of a neural-network-based procedure to train fuzzy or hybrid neural-fuzzy systems as vehicle idle-speed controllers. Simulation with a nonlinear model containing a significant delay was used, and an attempt was made to simulate the effects of realistic sampling and controller update frequencies. The present treatment may be regarded as a step toward online training with an actual system. The fuzzy system has a parameterized form similar to that described previously, allowing use of methods identical to those used for training neural networks. The results of training are illustrated by imposing various torque disturbances and showing the controller actions and the response of the system.<>