Published 2024-11-30
How to Cite

This work is licensed under a Creative Commons Attribution 4.0 International License.
Abstract
This study addresses the problem of low prediction accuracy of transmission time in complex network environments. A deep regression model that integrates network dynamic features is proposed. The model is based on real network states and constructs a multidimensional dynamic feature system. It includes key factors such as traffic variation, path congestion, and delay fluctuations. These features comprehensively reflect the non-stationarity and strong temporal nature of network operations. In terms of feature representation, the model introduces both time-sensitive and time-invariant structures. These features are encoded using a unified deep-learning framework. This enhances the model's ability to represent input characteristics. To further improve feature interaction and nonlinear modeling capacity, a multi-level feature fusion mechanism is introduced. It enables the integration of features from different sources across spatial and semantic levels. This enhances prediction stability and robustness. In the experimental section, the model is evaluated using the Internet2 network dataset. Its performance is compared with several mainstream models. The results demonstrate the advantages of the proposed method in error control and fitting accuracy. Ablation studies and hyperparameter sensitivity tests are conducted to verify the contributions of dynamic features and the fusion mechanism to performance improvement. Additionally, the model's robustness under abnormal network conditions is tested. The results further confirm the practicality and reliability of the proposed method in complex and dynamic network environments.