This research proposes an SLA-aware deep reinforcement learning framework for adaptive task scheduling and load balancing in edge–cloud environments. It models scheduling as a sequential decision problem and introduces an explicit SLA Violation Risk Score (SVRS) to capture per-task violation likelihood. The scheduler uses a hybrid Conv1D–GRU DRL policy with tier-aware reward shaping and SLA-aware action pruning to avoid high-risk allocations during congestion. A closed-loop monitoring mechanism updates penalties using real-time feedback. Simulation-based experiments show reduced SLA violations, lower latency, improved completion ratio, and better energy–efficiency trade-offs.
Yamsani et al. (Wed,) studied this question.