高速列车尾流区通常具有复杂的三维流动结构，由剪切层、脱落涡、分离与再附区、以及一对大型的反对称旋转流向涡等构成，强度不同的旋涡迅速地生成与脱落，对高速列车产生了非常不利的影响。这不仅增大了头尾车的压差阻力，涡对的周期性变化也严重影响尾车的升力和侧向力，对乘客乘坐的舒适性和列车运行的稳定性、安全性等都是一大挑战。此外，列车运行时诱导的最大列车风往往发生在尾流中，会产生显著的诱导压力，对平台上等待的乘客以及铁路沿线作业工人来说是非常危险的。因此，非常有必要对高速列车的尾流流动机理进行深入研究，对尾涡的生成与演化机理进行剖析，并在此基础上有针对性地提出合理有效的控制策略，以改善高速列车尾流场。本文以此为出发点，主要的研究内容和取得的研究进展如下：

1) 通过数值模拟实验和网格无关性验证，确定了适合于高速列车的精细数值计算方法；实现了两种流场分解方法——本征正交分解方法和动力学模态分解方法，并应用于高速列车尾流场研究中，对列车尾流进行了模态分析和特征识别；实现了涡参数分析方法，对列车尾涡结构及其发展规律进行了定量化描述。以上方法丰富了高速列车气动特性研究的分析手段。

2) 采用流场分解方法对高速列车尾流场开展模态分析，并对模态结构、模态频率及增长/衰减率等重要流场特征进行讨论，得到了高速列车尾流场在演化过程中的重要相干结构和主要规律；利用流场分解方法对流动频率的高解析能力，探索了尾流场的速度相似律问题，发现了尾涡脱落频率与列车运行速度的线性关系；通过涡参数分析方法，对列车尾涡涡旋中心的位置及其运动路径、涡核尺寸、涡强度等进行了识别与量化。进一步地，对三种不同简化程度的三编组列车尾流场的非定常特性开展研究，证明了转向架结构对高速列车尾流场的强干扰作用；并结合模态分解及流场重构，对列车尾涡对的两处产生位置、关键诱导因素和整个生成过程等进行了分析与阐述。以上成果可为高速列车尾流流动控制提供目标和思路。

3) 在明晰了列车尾涡的产生位置和关键诱导因素后，提出了三种流动控制策略，分别为：安装底部导流装置，安装尾涡扰流装置和设置喷流。通过在列车底部安装导流装置，可以实现三编组列车减阻约12%，车身长度上及尾流中列车风明显减小，并对比分析了底部导流装置的安装位置和结构参数对优化效果的影响；以减小尾车的气动升力为优化目标设计了三种尾涡扰流装置，对扰流装置的有效性和作用机理进行了分析，其中最优装置可实现减小尾车升力约31%；通过在尾车排障器末端设置喷流，可以大幅减小尾车的正升力并明显减小尾流场强度，此外还讨论了喷流压力和喷流角度对列车气动力及尾流场参数的影响规律。

4) 开展机器学习方法在高速列车中的应用研究。以TensorFlow为平台搭建了全连接神经网络和径向基函数神经网络两种人工神经网络模型，并在高速列车的气动力预测、表面压力分布预测等方面进行了应用研究；利用机器学习的数据驱动优势和流场分解方法的降阶技术，建立了本征正交分解+神经网络的高精度流场重构方法，并以高速列车表面压力分布、圆柱绕流非定常流场和高速列车非定常尾流场三个算例为例，对该方法进行了验证及拓展。

The wake region of high-speed trains usually has complex three-dimensional flow structures, consisting of shear layers, shedding vortices, separation and reattachment regions, and a pair of large anti-symmetric rotating streamwise vortices, etc. Vortices with different strengths are rapidly generated and shed, especially in the near wake area, which has adverse effects on high-speed trains. It not only increases the pressure drag, but also seriously affects the lift and side forces of the trailing car because of the periodic variation of vortex pair, causing discomfort to the passengers and affecting the stability and safety of the train. Moreover, a strong slipstream is always induced by the highly unsteady wake flow, causing significant induced pressure and potentially endangering passengers waiting at platform and track-side workers. Therefore, it is imperative to conduct an in-depth research on the flow mechanism of wake flow of high-speed trains, analyze the generation and evolution mechanism of the wake vortices, and put forward some reasonable and effective flow control strategies. Based on the above demands, the main contents and innovation progresses obtained in this study are as follows:

1)   A systematic numerical analysis method for high-resolution simulations of high-speed trains is established, through numerical simulation tests and grid resolution studies. Two flow field decomposition methods, namely the proper orthogonal decomposition method and the dynamic mode decomposition method, are realized and applied to the researches of wake flow of high-speed trains, including the modal analysis and feature recognition of the wake flow. Additionally, a vortex parameter analysis method is realized as well, and the structure of wake vortex and its evolution law can then be described quantitatively. With the above progresses, the analysis methods of aerodynamic characteristics research of high-speed trains are expanded.

2)   The modal analysis of wake flow of high-speed trains is carried out adopting the flow field decomposition methods. The important flow characteristics of wake flow, such as modal structures, modal frequencies and growth/decay rates, are discussed, revealing the important coherent structures and main laws of the wake flow in its evolution process. In addition, the velocity similarity law of wake flow field of high-speed trains is explored by using the highly analytical ability of flow frequency of the flow field decomposition methods, and the linear relationship between the shedding frequency of wake vortex and the train speed is found. The vortex parameter analysis method is successfully applied to identify and quantify the position of the vortex center and its motion path, the size of vortex core, and vortex intensity, etc. in wake flow of high-speed trains. Further, the unsteady characteristics of the wake flows of three train models with different simplifications are studied, revealing that bogies have a strong interference effect on the wake flow of high-speed trains. Based on modal decomposition and flow field reconstruction, the two generation positions, the key inducing factors and the detailed generation process of the wake vortex pairs are analyzed and explained. These results can provide design goals and ideas for wake flow control.

3)   After clarifying the generation locations and key inducing factors of the wake vortices, three flow control strategies are proposed, namely installing bottom deflectors, installing wake vortex spoilers and setting up jet flows. Firstly, by installing deflectors at the bottom of the train, the drag reduction of the three-carriage train can be achieved up to 12%, and the slipstream intensities along the train length-direction and in the wake are significantly reduced as well. The influences of the installation positions and structural parameters of the bottom deflectors on the optimization effect are also analyzed. Secondly, three types of wake vortex spoilers are designed to reduce the aerodynamic lift force of the trailing car, and the effectiveness and mechanism of the spoilers are analyzed. It is found that the best type can reduce the lift force of the trailing car by about 31%. Finally, by setting jet flows at the end of the tail cowcatcher, the positive lift force of the trailing car is reduced to negative lift force, and the strength of the wake is significantly reduced as well. Moreover, the effects of the rate of jet flow and jet angle on aerodynamics forces and wake field parameters are discussed.

4)   The machine learning method is applied to the research of high-speed trains. Two artificial neural network models, namely the fully connected neural network and the radial basis function neural network, are built based on the TensorFlow platform and applied to the predictions of aerodynamic forces and surface pressure distributions of high-speed trains. Using the data-driven advantages of machine learning and the reduced order technique of flow field decomposition method, a high-precision flow field reconstruction method combining proper orthogonal decomposition and neural network is established. Taking the predictions of surface pressure distribution of a high-speed train, unsteady flow field around a cylinder and unsteady wake flow of a high-speed train as three examples, this method is verified and extended.