Network traffic prediction based on transformer and temporal convolutional network

Area of Science:

• Computer Science
• Artificial Intelligence
• Network Engineering

Background:

• Current network traffic prediction models struggle with effectively capturing both long-term and short-term temporal dependencies.
• Accurate prediction of network traffic is crucial for efficient network management and resource allocation.

Purpose of the Study:

• To propose a novel hybrid model that integrates Transformer and Temporal Convolutional Network (TCN) for enhanced network traffic prediction.
• To address the limitations of existing methods in modeling complex temporal dependencies in network traffic data.

Main Methods:

• A hybrid model combining the strengths of Transformer and Temporal Convolutional Network (TCN).
• Transformer module utilizes multi-head self-attention to capture global temporal relationships.
• TCN module employs dilated convolutions to model local and long-term dependencies.

Main Results:

• The proposed hybrid model significantly outperforms current mainstream methods across all time steps on PeMSD4 and PeMSD8 datasets.
• Superior performance is particularly evident in long-term network traffic step prediction.
• Ablation studies confirm the significant contribution of both Transformer and TCN modules to the overall prediction accuracy.

Conclusions:

• The hybrid Transformer-TCN model offers a robust and effective solution for network traffic prediction.
• The integration of self-attention and dilated convolutions provides a powerful mechanism for capturing diverse temporal dependencies.
• This approach represents a significant advancement in the field of network traffic forecasting.