控制棒水压驱动系统（CRHDS）是一种新型内置式控制棒驱动技术，驱动泵是控制棒水压驱动系统的关键动力设备，现有驱动泵旁路因调节阀门节流会带来回路振动噪声问题。该文完成了驱动系统旁路变流阻流固耦合性能实验，获得了可调流阻部件的减振消能特性，通过物理信息神经网络（PINN）结合非支配排序遗传算法（NSGA-Ⅱ）进行了减振消能特性的建模和多目标优化，对改进型减振消能部件性能进行实验验证并完成了高温工况流场消能机理分析。结果表明，驱动回路振动加速度随关阀角度先上升后下降，多级阀门受末级阀门关闭角度影响较大。PINN模型相比于纯数据驱动的模型具有更好的泛化性能和预测精度，优化后的减振部件加速度较变流阻流固耦合实验最优结果降低84.9%，并满足驱动回路额定压降需求。研究成果为控制棒水压驱动系统旁路消能设备的设计和应用提供指导。

Objective: The control rod hydraulic drive system (CRHDS) is an innovative internal control rod drive technology developed by Tsinghua University for the low-temperature nuclear heating reactor NHR200-Ⅱ, in which the drive pump serves as the critical hydraulic power equipment. Its operational reliability is directly linked to reactor safety. However, the bypass circuit of the drive pump suffers from significant vibration and noise induced by the throttling of the control valve. Consequently, developing a high-efficiency multi-stage bypass energy dissipation and vibration reduction component is essential. This study aims to reduce pipeline vibration and enhance the operational reliability and safety of the CRHDS bypass loop through an optimization approach, thereby providing robust design criteria and theoretical guidance for bypass energy dissipation equipment. Methods: This study employed an integrated approach combining fluid-structure interaction (FSI) experiments, machine learning, multi-objective evolutionary optimization, and computational fluid dynamics (CFD) simulations. First, an FSI experimental loop simulating the CRHDS bypass was constructed, comprising a water tank, a centrifugal pump, adjustable flow resistance components, and pipe supports. Three series-connected ball valves were used to simulate a multi-stage flow resistance component, with system parameters recorded at a sampling frequency of 4 000 Hz using four fast-response pressure sensors, a three-axis pipe accelerometer, and an ultrasonic flow meter. A 125-group full-factorial FSI experiment was conducted using characteristic valve closing angles of 20°, 30°, 40°, 50°, and 60°. Second, a physics-informed neural network (PINN) surrogate model, evaluated via 5-fold cross-validation, was developed to predict the total pressure drop and synthetic vibration acceleration using the three-stage resistance coefficients. To improve prediction accuracy, a physical constraint stipulating that the total pressure drop increases monotonically with the sum of the resistance coefficients was integrated into the network's loss function. Third, the PINN surrogate was coupled with the non-dominated sorting genetic algorithm Ⅱ (NSGA-Ⅱ) to perform multi-objective optimization, targeting minimized synthetic vibration acceleration under rated pressure drop constraints. Based on the resulting Pareto optimal front, an optimized design was selected to fabricate a physical three-stage orifice component for experimental validation. Finally, CFD simulations were carried out to analyze the internal flow fields of the optimized component under high-temperature operating conditions. Results: The experimental, optimization, and simulation results indicated the following: 1) The loop vibration acceleration exhibited a non-monotonic trend, initially increasing and subsequently decreasing with increasing valve closing angles, with the final-stage valve angle exerting a dominant influence; 2) The prediction accuracy of the PINN model for the total pressure drop and synthetic acceleration improved by 27.9% and 29.4%, respectively, compared with the conventional radial basis function model; 3) The optimized multistage orifice component achieved an 84.9% reduction in synthetic vibration acceleration compared with the initial FSI tests while successfully satisfying the loop pressure-drop requirements; 4) Under the 230 ℃ high-temperature condition, the minimum pressure in the flow field was 3.58 MPa, which was considerably above the saturation vapor pressure of water (2.7 MPa), thereby ensuring an ample cavitation margin. Conclusions: By implementing an integrated framework of FSI experiments and multi-objective optimization, the optimal flow resistance configuration of multi-stage throttling components can be systematically determined to achieve vibration reduction and energy dissipation, thereby mitigating pipeline vibration induced by single-stage throttling. The optimized multi-stage orifice component achieved an 84.9% reduction in loop vibration acceleration under rated driving pressure and was verified to possess a sufficient cavitation margin under high-temperature reactor conditions.