Dynamic Task Prioritization for Edge AI in Smart Cities: Balancing Latency and Energy Efficiency

The growth of latency sensitive smart city applications is rapid nowadays. Therefore deploying microservice architecture in the heterogeneous edge cloud continuum has become a mainstream choice. However a structural challenge arises in orchestrating these coupled services modeled as Directed Acyclic Graphs. Exact algorithms like Branch and Bound struggle with computation in high concurrency scenarios. Meanwhile Deep Reinforcement Learning methods face the challenges of excessive training overhead and a lack of zero shot adaptation capability for topology changes. Recently popular quantum inspired algorithms often fail to satisfy strict predecessor constraints. This makes them unsuitable for use in dependent workflows. To alleviate this dilemma this paper proposes a Dependency Aware Quantum Inspired Scheduler. This framework utilizes a topological quantum coding scheme and dynamic dependency masks to integrate DAG constraints into the quantum search process. It also introduces an entropy weighted evolutionary rotation mechanism to accelerate the convergence of critical paths. After conducting extensive simulation experiments in city level environments we found that the scheduling success rate of DAQ Scheduler is 100% while the success rate of standard quantum inspired algorithms is only 58.1%. Compared with leading multi objective DRL baselines this method reduces the average makespan by 9.9%. This method provides near optimal scheduling solutions with millisecond level inference latency. It builds an efficient and lightweight paradigm for real time edge intelligence and balances theoretical optimality with engineering feasibility.