Grey Wolf Optimization Method Based on Dynamic Workload Allocation for Video Conferencing System

Abstract

Video conferencing systems are essential for real-time communications. Still, they can be challenging to manage due to dynamic workloads and the need to maintain high quality of service (QoS) under unpredictable network conditions and heterogeneous video content. Deep learning (DL) is used for workload prediction and optimisation; however, traditional DL-based grey wolf optimisation (GWO) often suffers from slow convergence, local optimality, and difficulty in handling uncertainty. To address these issues, the proposed workflow incorporates four novel techniques: Federated GWO with Cross-Domain Adaptability enables multiple platforms to predict allocation policies in a privacy-preserving manner collectively. Furthermore, Spatiotemporal Autoencoder with Outlier Detection learns robust latent features and removes anomalies from a videoconferencing dataset (VCD). Additionally, the Diffusion Model for QOS Prediction Generates probabilistic reports about future resource requests and quality of service metrics. Finally, Neural Evolutionary GWO (NE-GWO) employs a neural-guided evolutionary strategy to optimise server task assignment. The proposed method provides an integrated approach with efficient scalability, high QoS, adaptive workload management, and maximum resource utilization in heterogeneous video conferencing environments, achieving 94% performance.