Dirty-data-based alarm prediction in self-optimizing large-scale optical networks.

Area of Science:

• Optical Network Engineering
• Machine Learning Applications
• Network Reliability and Management

Background:

• Large-scale optical networks generate performance and alarm data that is often incomplete, inconsistent, and contains errors, especially from older equipment.
• This 'dirty' data, even after preprocessing, presents challenges like unbalanced feature distributions, limiting the effectiveness of standard machine learning algorithms for alarm prediction.
• Accurate alarm prediction is critical for network administrators to implement preventive measures and ensure network stability.

Purpose of the Study:

• To develop and demonstrate a novel machine learning-based method for predicting alarms in large-scale optical networks despite data quality issues.
• To address the challenge of dirty, unbalanced datasets in optical network alarm prediction.
• To improve the reliability and self-optimizing capabilities of optical networks through accurate, timely alarm prediction.

Main Methods:

• Development of a Dirty-data-based Alarm Prediction (DAP) method specifically designed for optical networks.
• Utilizing machine learning techniques to handle incomplete, inconsistent, and unbalanced datasets.
• Testing the DAP method on a commercial large-scale field topology comprising 274 nodes and 487 links.

Main Results:

• The proposed DAP method demonstrated high accuracy in predicting various types of alarms.
• The method proved effective even when dealing with inherently 'dirty' and unbalanced data typical of optical network environments.
• Validation was performed on a substantial real-world network topology, confirming practical applicability.

Conclusions:

• The DAP method offers a robust solution for alarm prediction in large-scale optical networks, even with challenging data quality.
• This approach enhances the potential for self-optimizing optical networks (SOONs) by enabling proactive issue resolution.
• The findings highlight the importance of tailored machine learning strategies for managing complex network data and improving operational efficiency.