SYNTHESIZE A NEURAL NETWORK PARAMETER OPTIMIZER FOR AN ADAPTIVE PID CONTROLLER
Аннотация
Wide application of proportional-integral-differential (PID)-regulator in industry requires constant improvement of methods of its parameters superstructuring. In the paper, the questions of optimization of PID-regulator parameters with application of methods of neural network technology are considered. A methodology for selecting the architecture of neural network optimizer designed to determine the tuned parameters of PID regulator is proposed. The algorithm of training of the neural network, with the set on the basis of the method of inverse gradient propagation is offered. The proposed improved PID-neural regulator allowed to provide stabilization of neural network operation and its trainability in the control loop in real time. Neural network optimizer, is able to reconfigure the parameters of the PID-regulator during the transient process and provide the required quality of the transient process when varying the parameters of the object. The main changes of the proposed approach for neural network control of a dynamic object, is to develop a methodology for changing the speed and direction of training of the neural network, as well as the rules of training the output neurons of the neural network, which are the parameters of the PID-regulator. The results of computational experiments have shown that the created neural add-on in the form of a neurooptimizer can become a prototype of an industrial controller for tuning the parameters of the PID controller.