Deep Reinforcement Learning-Based Microservices Placement Under Zero Trust Security Framework in Smart Grid

The microservice architecture has become vital in power grid systems, providing scalability, flexibility and resilience through independent deployment. However, its numerous access points introduce severe security challenges. The zero trust model, guided by the principle of “never trust, always verify,” offers a promising solution, yet integrating it with microservice deployment under strict resource and trust constraints remains difficult. This paper models the secure deployment task as a microservice placement optimization problem and formulates it as an integer linear program (ILP). To address ILP scalability limits, a hybrid linear programming (LP) approach enhanced by a Soft Actor–Critic (SAC) reinforcement learning agent and iterative rounding is proposed. The SAC agent learns adaptive rounding strategies to balance efficiency and solution quality. Results show that the proposed LP+SAC framework achieves near-optimal deployment with significantly reduced runtime and strong robustness under complex zero trust conditions, demonstrating the potential of reinforcement-guided rounding for secure and efficient scheduling in large-scale power grid systems.