文章旨在解决光传送网（OTN）网络性能劣化（即误码）预测难题。OTN误码类性能值依赖于OTN帧开销中的比特交叉奇偶校验（BIP）字节检测，并由网络管控系统周期性统计计算得出。在OTN网络正常运行的绝大多数情况下，误码类性能值保持为零，这无疑为传统方法及最新人工智能（AI）技术预测OTN误码相关性能带来了挑战。

文章提出了一种利用OTN光层与电层之间对应关系进行误码概率预测的方法。首先，借助深度学习技术预测光信道误码率（BER）的变化趋势；随后，基于预测的光信道BER，运用文章所提机器学习模型进一步预测OTN误码概率。

通过仿真实验验证，该方法的预测准确性超过90%。

文章所提方案达到了工程化应用的要求，为OTN网络性能劣化预测提供了新的有效方法，并为OTN网络预测性维护提供了有力依据。

This paper aims to address the challenge of predicting performance degradation (frame transmission errors) in Optical Transport Network (OTN). Frame error performance metrics in OTN rely on the detection of Bit Interleaved Parity (BIP) bytes in OTN frame overhead, which are periodically calculated by network management systems. In the vast majority of cases where the OTN network operates normally, the error-related performance values remain zero, which undoubtedly poses a challenge for both traditional methods and the Artificial Intelligence (AI) technologies in predicting OTN error-related performance.

This paper proposes a creative approach to predict error probability by leveraging the correspondence between the optical and electrical layers in OTN. Firstly, deep learning techniques are used to predict the trend of Bit Error Rates (BER) in optical channels. Subsequently, based on the predicted BER in optical channels, the proposed machine learning models are employed to further predict the frame error probability in OTN.

Verified through simulation experiments, the prediction accuracy of this method exceeds 90%.

The proposed solution meets the requirements for engineering applications, providing a new and effective method for predicting performance degradation in OTN networks. It also provides a strong basis for predictive maintenance of OTN networks.