Nonlinear Generalized Predictive Control of Nuclear Reactor Power Level With Nonlinear Disturbance Observer

Reactor load tracking control plays a vital role in enhancing the peak regulation capability and operational stability of power grids. As typical complex physical facilities, nuclear reactors feature strong coupling, time‐varying behaviors, and high nonlinearity, and their dynamic models cover the complex coupling effects among neutron kinetics, thermohydraulics, and other multiphysical processes. Most existing predictive control methods are developed based on linearized system models, which ignore high‐order nonlinear terms and further trigger model mismatch problems. Besides, iterative updates of linearized models will bring heavy computational pressure to real‐time online operation. To solve the above drawbacks, this study develops an improved control strategy integrating nonlinear disturbance observer (NDO) and parameter identification with continuous‐time nonlinear generalized predictive control (NGPC). Compared with discrete‐time control frameworks, continuous‐time–based NGPC can better capture the intrinsic nonlinear dynamics of reactors and mitigate adverse effects brought by modeling errors. Unlike traditional control methods that only adopt fixed nominal models, the proposed strategy adopts the Levenberg–Marquardt algorithm to identify time‐varying degradation parameters during reactor operation. Meanwhile, the designed NDO is applied to real‐time observation of unknown lumped disturbances, and the estimated disturbance values are fed into the NGPC loop to establish a dynamic disturbance compensation module. This design effectively improves the robustness and antidisturbance ability of the overall reactor control system. Finally, comparative load tracking experiments are conducted among PID, NGPC, and the proposed NDO–NGPC + identification method. The results demonstrate that the NDO–NGPC + identification strategy exhibits superior tracking accuracy and disturbance suppression performance in the presence of both step and sinusoidal disturbances, validating its effectiveness in improving reactor control performance under complex operating conditions.