Attention-Enhanced Double Deep Q-Network for Multi-Objective Deadline Aware Task Scheduling in Fog Computing

Abstract The growth of Internet of Things (IoT) devices has made cloud-based processing less suitable for latency-sensitive applications, prompting the adoption of fog computing as a distributed computing model to replace it. Nevertheless, the very nature of a fog environment, which is marked by the heterogeneity of the workload and challenging Quality-of-Service (quality of service) demands, creates major difficulties in the context of traditional scheduling schemes. The paper presents a framework for the Attention-Enhanced Double Deep Q-Network (ADDQN) for multi-objective, deadline-aware task scheduling in heterogeneous fog computing settings. The proposed architecture combines a node encoder and self-attention, two-way processing of global and local state representations, and a bilaterally designed multi-objective reward scheme that simultaneously optimizes response time, deadline conformity, load balancing, and energy consumption. Even a strict statistical analysis of 30 trials (independent) proves that the suggested framework attains the deadline miss rate of 4.2% and average response time of 136.33 ms, which are the 61.1% and 6.4% and 20% decrease, respectively, in the best heuristic baseline (Shortest Queue). All improvements are statistically significant (p < 0.05), with the most significant effect (p < 0.001) observed relative to the Random and Lowest Latency baselines. The outcomes also show high practical effect sizes (Cohen d > 1.0), making the suggested solution a powerful and valid remedy for the QoS-constrained fog scheduling problem.