行波管(Traveling Wave Tube, TWT)是通信卫星中的关键部件，通常长期工作在高温高压环境中，因此，其退化程度评估对于在轨通讯卫星安全运维尤为重要。传统的机理模型和门限规则方法难以对复杂电热耦合场效应下的行波管退化程度进行有效表示。为了解决这一挑战，本文提出了一种基于多尺度注意力编码器(Multi-scale Attention Encoder, MSA-Encoder)的行波管退化评估方法。首先，本文对遥测数据的复杂特性进行了详细分析，探讨了这些特性对深度学习模型训练带来的困难及其解决方案。其次，设计了MSA-Encoder网络，其通过自相关机制和全局-局部通道注意力(Global-Local Channel Attention, GLCA)分别对多维遥测数据中的时间和通道依赖关系进行充分捕捉，从而实现对数据的精准重构。再次，本文提出了一种基于马氏距离和移动平均季节性分解的行波管退化指数计算方法，该方法通过重构差异性理论进行数据漂移指数计算，再对其进行季节性分解，实现退化指数的计算。最后，本文利用某通讯卫星全生命周期的行波管数据构建了行波管退化数据集，并在该数据集上进行了数值实验。实验结果表明，MSA-Encoder在遥测数据特征提取能力方面优于其他同参数规模基准模型。同时，实验结果验证了本文提出的退化指数计算方法的可信性。

Traveling Wave Tube (TWT) is a key component in communication satellites, typically operating in high-temperature and high-pressure environments for extended periods, making its degradation evaluation particularly important. Traditional mechanism models and threshold rule methods struggle to effectively represent degradation levels under complex electro-thermal coupling effects. To address this challenge, we propose a deep neural network degradation evaluation method based on a Multi-scale Attention Encoder (MSA-Encoder). First, the complex characteristics of telemetry data are analyzed in detail, exploring the difficulties these characteristics pose for training deep learning models and providing solutions. Second, the MSA-Encoder network is designed, which precisely captures the temporal and channel dependencies in multi-dimensional telemetry data through Auto-Correlation and Global-Local Channel Attention (GLCA) mechanisms, respectively, achieving accurate data reconstruction. Third, we propose a TWT degradation index calculation method based on Mahalanobis distance and moving average seasonal decomposition. This method calculates data drift index through reconstruction difference theory, followed by moving average seasonal decomposition to compute the degradation index. Finally, using the full life cycle TWT data from a certain communication satellite, a TWT degradation dataset (TWT Dataset) is constructed, and numerical experiments are conducted on this dataset. The experimental results show that the MSA-Encoder surpasses benchmark models of the same parameter scale in telemetry data extraction capability, proving the reliability of the degradation index calculation.