Deep Learning-Based Adaptive Compression and Anomaly Detection for Smart B5G Use Cases Operation.

Area of Science:

• Telecommunications Engineering
• Network Operations
• Artificial Intelligence

Background:

• Next-generation Beyond 5G (B5G) networks demand efficient operations for high-bandwidth use cases like Digital Twins (DT).
• Digital Twins generate vast amounts of sensor telemetry data, potentially overwhelming network transport layers.
• Hierarchical data analysis can reduce centralized processing for tasks like anomaly detection.

Purpose of the Study:

• To propose a smart management system for efficient telemetry sensor data analysis in B5G networks.
• To reduce the network transport load caused by high-bandwidth use cases such as Digital Twins.
• To enable adaptive data compression and hierarchical anomaly detection for sensor telemetry.

Main Methods:

• Development of a hierarchical architecture for telemetry sensor data analysis.
• Implementation of deep autoencoder (AE)-based methods for adaptive compression (AAC) of time series data.
• Application of AE-based methods for single (SS-AD) and multiple (MS-AGD) sensor stream anomaly detection.

Main Results:

• Achieved data compression ratios of up to 64% with reconstruction errors below 1%.
• Demonstrated fast and accurate anomaly detection for both single and multiple sensor streams.
• Obtained a reduction of 50% or more in transport network capacity resources for distributed DT scenarios.

Conclusions:

• The proposed hierarchical system effectively reduces telemetry data volume and enhances anomaly detection.
• Smart use case operation significantly conserves network transport capacity for demanding B5G applications.
• Deep autoencoders offer a powerful solution for managing and analyzing large-scale sensor data in future networks.