A TSENet Model for Predicting Cellular Network Traffic.

Area of Science:

• Computer Science
• Electrical Engineering
• Telecommunications

Background:

• Wireless sensor networks (WSNs) are increasingly vital for adaptable, configurable, and flexible network communication.
• Predicting future network traffic in WSNs is crucial for efficient resource management and performance optimization.
• Existing temporal sequence models offer potential but require enhancement for complex cellular network dynamics.

Purpose of the Study:

• To introduce TSENet, a novel method for accurate cellular network traffic prediction.
• To leverage transformer and self-attention mechanisms for enhanced traffic forecasting.
• To improve the understanding and management of network traffic in wireless sensor networks.

Main Methods:

• TSENet integrates a temporal transformer module to extract temporal traffic flow features at near-term and periodical intervals within network grids.
• A spatial transformer module amalgamates spatial features from correlated grids to generate spatial predictions.
• Self-attention aggregation captures dependencies between external factors and cellular data features.

Main Results:

• TSENet demonstrated high accuracy in predicting cellular network traffic.
• Empirical assessments on a real-world cellular traffic dataset validated the model's efficacy.
• The method successfully captured complex temporal and spatial dependencies in network traffic.

Conclusions:

• TSENet offers a robust and accurate solution for cellular network traffic forecasting.
• The integration of temporal and spatial modeling with self-attention significantly enhances prediction capabilities.
• This approach holds promise for optimizing wireless sensor network performance and resource allocation.