【摘要】 与固态器件相比，真空电子器件在高频段可以提供更大的功率、更宽的带宽。输能窗作为其中的重要部件，用以隔绝真空电子器件内外环境，保持管内的高真空度，传输高能电磁波。目前，传统输能窗的性能已经制约宽带高频真空器件性能。基于此，本文提出了一种利用神经网络算法设计超材料窗片的方案，实现了低反射和驻波比的宽带输能窗。本文利用神经网络的非线性拟合效应对窗片结构与其频谱之间建立映射关系，从而避免了求解麦克斯韦方程组的繁复过程，提高了频谱性能仿真的速度，简化了设计过程。通过设计双向神经网络并进行训练，实现反向设计窗片的目标，输入目标频谱特性参数即可生成满足指标要求的结构参数。更多还原

【Abstract】 Compared with solid-state devices, vacuum electronic devices can provide higher power and wider bandwidth. As an important component, the microwave window is used to isolate the internal and external environment of the vacuum electronic device, maintain the high vacuum inside the device, and transmit high-energy electromagnetic wave. At present, the performance of traditional microwave window constrains the performance of broadband high-frequency vacuum devices. Therefore, a design method of metamaterial structure using neural network algorithm is proposed which can realize a broadband microwave metamaterial window with low reflection and low voltage standing wave ratio. The nonlinear fitting effect of neural network is used to establish a mapping relationship between the window structure and its spectrum, thus avoiding the complicated solving process of Maxwell’s equations, improving the simulation speed of spectrum performance and simplifying the design process. By designing and training the bidirectional neural network, the reverse design of window is achieved, by which the structural parameters can be generated only by inputting the corresponding target spectral characteristic parameters.更多还原