冷凝器是动力和发电系统中的关键热力设备之一，其性能的优劣对动力系统的循环热效率和运行经济性具有重要影响。然而由于冷凝器中动辄上万根管的排布，涉及参数多，结构复杂，传统的热力设计方法已无法有效提升冷凝性能。随着计算流体力学（CFD）方法的进步，在设计过程中结合CFD计算结果将显著提升冷凝器设计水平。然而受到CFD计算量大、结果与设计参数交互性不足等限制，耦合CFD结果的参数优化设计方法仍然是缺失的。为此，本文基于有限体积法自编程开发了冷凝器的二维双流体CFD计算模型，进一步构建了利用粒子群优化算法驱动CFD数值计算并实现设计参数自动寻优的设计方法；实现了不同热负荷条件下冷凝器运行特性的神经网络快速预测。本研究取得如下成果：（1）基于MATLAB平台自编程开发了二维单相及两相CFD计算模型，可以合理描述冷凝器中的流动、传热和相变过程。并通过与标准算例及文献报道实验数据的比对，验证了所开发二维CFD数值方法的准确性。通过参数敏感性分析，发现液膜热阻中的参数取值对冷凝器性能预测具有重要影响。提出了一种新的冷凝液溢流效应修正项的计算方法，消除了对管排数的依赖，该方法使计算冷凝器热流密度分布的平均相对误差由24.73%降低至19.20%。（2）基于CFD计算模型，实现了粒子群优化算法耦合CFD求解的参数自动寻优设计方法。并基于多目标粒子群算法（MOPSO）建立了管束选型的优化设计方法，可为工程师提供从换热强度、功耗角度考虑的最优方案；在管束排布设计优化过程中，面对管束区全网格随机优化孔隙率所带来的计算量过大问题，首次提出了一种基于单目标粒子群（PSO）的分区域孔隙率优化思路，优化结果相比于标准的均匀管束布置方式，冷凝率提高了0.11%，壳侧气阻下降7.35%。该方法可在实际管排布置设计过程中，为设计师提供管排疏密布置的建议。（3）基于两相CFD模型，得到了实际工程上所关心的不同热负荷条件下冷却水入口流速变化曲线。并基于浅层前馈神经网络，对85%热负荷条件下的速度场和热流密度场进行了预测，预测的相对误差均小于5%，结果的相关系数均大于0.99。且与CFD计算相比，神经网络预测所耗时间大幅降低。训练得到的神经网络满足实际工程需要。

Condenser is one of the key thermal devices in power and generation systems, and its performance has a significant impact on the cycle thermal efficiency and operational economy of power systems. However, due to the arrangement of tens of thousands of tubes in the condenser, involving numerous parameters and complex structures, traditional thermal design methods are no longer effective in improving condenser performance. With the advancement of Computational Fluid Dynamics (CFD) methods, combining CFD calculation results in the design process will significantly enhance the level of condenser design. However, due to limitations such as the large computational workload of CFD and insufficient interaction between results and design parameters, a coupled parameter optimization design method based on CFD results is still lacking. Therefore, this paper develops a two-dimensional two-phase CFD calculation model of the condenser based on the finite volume method, further constructs a design method that utilizes the Particle Swarm Optimization (PSO) algorithm to drive CFD numerical calculation and achieve automatic parameter optimization of design parameters. The paper also realizes rapid prediction of condenser operating characteristics under different thermal load conditions using neural networks. The main achievements in this research are as follows:(1) A two-dimensional single-phase and two-phase Computational Fluid Dynamics (CFD) calculation model was self-programmed and developed on the MATLAB platform, which can reasonably describe the flow, heat transfer, and phase change processes in the condenser. The accuracy of the developed two-dimensional CFD numerical method was validated by comparison with standard cases and experimental data reported in the literature. Through parameter sensitivity analysis, it was found that the parameter values in the liquid film thermal resistance have a significant impact on the prediction of condenser performance. A new calculation method for correcting the condensate overflow effect was proposed, which eliminates the dependence on the number of tube rows. The average relative error of the method for calculating the condenser heat flux density distribution is reduced from 24.73% to 19.20%.(2) Realized the parameter automatic optimization design method coupled with Particle Swarm Optimization (PSO) algorithm based on CFD calculation model. Based on the Multi-Objective Particle Swarm Optimization (MOPSO) algorithm, an optimization design method for bundle selection was established, which can provide engineers with the optimal solution considering heat transfer intensity and power consumption. In the optimization process of bundle layout design, facing the problem of excessive computational workload caused by random optimization of grid porosity in the bundle area, a partitioned porosity optimization approach based on single-objective Particle Swarm Optimization (PSO) was proposed for the first time. Compared with the standard uniform bundle arrangement, the condensation rate increased by 0.11% and the shell-side gas resistance decreased by 7.35%. This method can provide suggestions for the sparse and dense arrangement of tube bundles in the actual tube layout design process.(3) Based on the two-phase CFD model, the variation curve of cooling water inlet velocity under different thermal load conditions of practical engineering was obtained. Based on the shallow feedforward neural network, the velocity field and heat flux density field under 85% thermal load conditions were predicted, and the relative errors of the predictions were all less than 5%, and the correlation coefficients of the results were all greater than 0.99. Moreover, compared with CFD calculation, the time consumed by neural network prediction was significantly reduced. The trained neural network meets the needs of practical engineering.