Cloud Gaming Resource Management Platform Based on Edge Intelligence

This study thoroughly explores the rapid development of edge intelligence, emphasizing the synergy between cloud computing and edge computing to significantly enhance data processing efficiency. It highlights the advantages of edge intelligence-based cloud gaming platforms over traditional cloud gaming platforms. Traditional resource pooling techniques perform poorly and incur high costs during fluctuating user demands. To address this, we introduce edge intelligence to cloud computing and, employing the LSTM algorithm, construct a predictive model for resource pooling, demonstrating its efficiency and adaptability. The innovation of this paper lies in proposing a wireless communication traffic prediction model based on federated learning within a distributed architecture. Individual grid traffic prediction models are trained synchronously, and the central cloud server uses Jensen-Shannon (JS) divergence to select grid traffic models with similar distribution. It utilizes a federated averaging algorithm to merge parameters of grid traffic models with comparable distribution, aiming to enhance model generalization while accurately characterizing local traffic patterns. Additionally, the paper elaborates on optimizing resource caching through PID automatic control algorithms in the context of pooling strategies, addressing sudden spikes and drops in traffic.