Proactive Failure Management Methods for Coherent Optical Networks

点击翻译按钮获取中文摘要

Current fault management strategies are essentially reactive and consist in implementing in the network control**system algorithms for network reconfiguration which take action only after the failures. Proactive approaches**that allow forecasting equipment failures in optical networks have been explored in the last few years.**Equipment failure prediction methods using machine learning (ML) and time series in optical networks have**been proposed. Different ML techniques have been explored in the last two years to provide cognitive solutions**for failure detection in optical networks. A failure identification and localization tool based on Bayesian**inference, as well as methods for bit error rate (BER) degradation detection and failure detection in optical**networks, have been proposed recently. However, most of the research on cognitive optical networking is at a**very early stage of exploration and based on synthetic or lab performance data in the absence of field**performance data. Therefore, the applicability of these new proactive fault management strategies in real**network deployment conditions is yet to be demonstrated. The development of predictive system analysis tools**would provide network operators with an advantage in making their optical network more reliable and would**also provide an opportunity to reduce their operational expenditures through automated network control**system.**The proposed project will explore proactive fault management strategies in coherent optical networks based on**machine learning techniques. The project will leverage the research activities on cognitive optical networking**at the Network Technology Lab in the last two years. The project will consist in evaluating different ML**algorithms, with the objective to determine the most promising approaches for fault prediction. The innovative**ML methods will be evaluated according to the following metrics: prediction accuracy, computing time and**scalability. Synthetic BER data will be used for evaluating the performance of the ML-based fault predictors.