Research on OTN Service Delay Estimation Method based on Machine Learning

【Objective】To meet the requirements of timeliness, accuracy, and completeness of delay data in delay-sensitive application scenarios, it is necessary to implement end-to-end service delay estimation in Optical Transport Networks (OTN).【Methods】This paper first analyzes the transmission characteristics of OTN services, and collects service routing information according to the sub-net connections. Next, it discretizes the basic data such as Network Elements (NE), links, and cross-connection in service route. Then the characteristic variables for delay estimation are obtained. Finally, the paper proposes a delay estimation model based on engineering live network, and compares the simulation results of various machine learning algorithms.【Results】The Mean Absolute Percentage Errors (MAPE) of the delay estimation results based on Support Vector Regression (SVR) and decision tree regression were 3.362 8% and 1.284 9%, respectively.【Conclusion】The OTN service delay estimation method based on machine learning and the characteristic of OTN transmission in this paper has high accuracy and wide application scenarios.