Recurrent Neural Network Based Adaptive Variable-Order Fractional PID Controller for Small Modular Reactor Thermal Power Control

Abstract

This paper presents the synthesis of an adaptive PID type controller in which the variable-order fractional operators are used. Due to the implementation difficulties of fractional order operators, both with a fixed and variable order, on digital control platforms caused by the requirement of infinite memory resources, the fractional operators that are part of the discussed controller were approximated by recurrent neural networks based on Gated Recurrent Unit cells. The study compares the performance of the proposed neural controller with other solutions, which are based on definitional fractional-order operators exploiting an infinite memory buffer and a classical adaptive PID controller. The proposed neural approximations of variable-order fractional operators applied to a PID-type controller provide a viable solution that can be successfully implemented on present-day digital control platforms. The research presented here focuses on the aspects of accuracy of approximators in simulated operating conditions within the thermal power control system of the challenging plant such as Small Modular Nuclear Reactor.