Neural networks and adaptive finite-time state observer-based preassigned-time fault-tolerant control of load following for a PWR-SMR under CRDM faults and sensor noises

Abstract

Load Following (L-F) control of Nuclear Power Plant (NPP) under actuator faults and measurement noises remains challenging both in theory and in practice. To solve this challenge, a globally preassigned-time stable fault-tolerant control strategy of rapid L-F for a Pressurized Water Reactor based Small Modular Reactors (PWR-SMR) under control rod drive mechanism (CRDM) faults and sensor noises is first proposed. From a practical standpoint, considering some states of PWR-SMR cannot be measured directly and the sensor noises in practice, a finite-time tracking differentiator and a nonlinear adaptive practical finite-time state observer are tactfully constructed, which can guarantee that the observation error system converges to a small residual set within a finite-time. Under the framework of Filippov solution, the differential inclusion technique is used to tackle the power error system which may be discontinuous. In addition, to compensate the adverse effects of the CRDM faults, a radial basis function neural network (RBFNN) is employed with a preassigned-time convergent updated law. Then two preassigned-time stable controllers are proposed to guarantee that the prescribed performance of L-F for PWR-SMR can be realized within a preassigned-time. At last, simulation studies, involving performances of L-F, tracking differentiator, finite-time observer and uncertainty of parameters, demonstrate the effectiveness and feasibility of the theoretical results.