Nonlinear multi-node reactor core models can describe spatial power redistribution, thermal feedback, xenon dynamics, but their high dimensionality and nonlinear characteristics make them difficult to use directly in real-time MPC design. To address this problem, this paper proposes a reduced-order pDMDc-MPC strategy for nuclear reactor power control. Excitation data are first generated from a fractional multi-node axial core model, and local DMDc models are identified in different operating regions. These local models are then fused through a scheduling-dependent pDMDc weighting strategy to construct a low-order prediction model suitable for online MPC optimization. Based on the pDMDc model, a constrained MPC controller is designed to coordinate total-power tracking and axial-offset regulation while considering actuator effort and smoothness. The proposed method is evaluated under reference power step changes, inlet coolant-temperature disturbances, and load-following scenarios. Compared with an optimized PID controller, the pDMDc-MPC strategy achieves improved tracking accuracy, disturbance rejection, axial-offset suppression, and actuator smoothness. The results show that pDMDc provides an effective data-driven bridge between nonlinear fractional multi-node reactor modeling and real-time predictive control.
 

Fractional nodal reactor model; reduced order model; dynamic mode decomposition with control; model predictive control