Design of direct controllers of PID type by receptive field neural networks

A direct controller of PID type using receptive field neural networks is suggested. This method does not put too much restriction on the type of the plant to be controlled and it has a stable performance for the type of inputs it has been trained for. Unlike backpropagation or other supervised methods of training, this approach does not require the knowledge of the appropriate form of controller output for each given input and neither does it require identification of the plant or its inverse model. The parameters of the PID controller in this method are determined by weight perturbation. Simulation results show a satisfactory performance.