Balancing Security and Communication Performance in 5G‐A Industrial IoT With CES‐Based TD3

Abstract

5G‐Advanced (5G‐A) industrial Internet of Things (IIoT) requires security mechanisms and communication resources to be coordinated under stringent reliability and ultra‐low‐latency constraints. Existing security‐oriented or communication‐oriented optimization methods usually treat protection strength and resource allocation separately, making it difficult to capture the nonlinear coupling among authentication, encryption, deep packet inspection, access control, latency, residual risk, and quality of service under dynamic traffic and threat conditions. To address this problem, this paper proposes a security‐aware CES‐TD3 framework that integrates a unified latency‐risk model, a constant elasticity of substitution (CES) utility function, and a twin delayed deep deterministic policy gradient (TD3) algorithm for continuous joint security–communication control. The framework explicitly quantifies security‐induced transmission, processing, and queueing delays, penalizes weak security or communication dimensions through CES utility, and incorporates threat‐guided exploration, violation‐aware prioritized replay, twin‐Critic learning, and feasible‐action projection to improve safe adaptation. Experiments with 20 industrial terminals, mixed load and threat scenarios, and five independent random seeds show that the proposed method achieves the highest reward of 0.8457±0.0059$$ 0.8457\pm 0.0059 $$ and the highest CES utility of 0.8584±0.0061$$ 0.8584\pm 0.0061 $$ , improving utility by 4.7%$$ 4.7\% $$ and reward by 4.9%$$ 4.9\% $$ over the Risk‐aware heuristic, outperforming DDPG, SAC, and PPO in joint utility, reducing packet loss by 23.1%$$ 23.1\% $$ , and maintaining a zero risk‐violation rate. Ablation, Pareto‐frontier, adaptive‐behavior, and CES‐parameter analyses further confirm that the proposed method provides an interpretable and adjustable route for balancing security protection and communication performance in 5G‐A industrial private networks.