Abstract
In modern smart grids and the industrial Internet of Things, ensuring safety and the real-time issuance of control instructions for metering automation systems poses severe challenges. Traditional static encryption protocols often cause significant communication delays and control interruptions due to channel packet loss and data expansion when operating in bandwidth-limited environments with high-frequency electromagnetic interference. To overcome the inherent barrier between high-strength cryptographic protection and real-time interaction with short packets, this paper proposes an adaptive encryption and fault-tolerant transmission architecture that integrates the SM4 block cipher algorithm and a genetic algorithm. To bridge the structural gap in existing static security models, this study introduces a Spatiotemporal-Aware Adaptive Encryption framework. Specifically, a multi-layer Transformer encoder extracts latent spatiotemporal network features to serve as real-time environmental inputs for a Genetic Algorithm (GA), which dynamically optimizes SM4′s key rotation period and forward error correction (FEC) redundancy. Simulations conducted on the NS-3 platform involving 50 terminal nodes and 60,000 real-world power control logs demonstrate the framework's robustness. Compared to the standard 'SM4-Static' baseline, the proposed SM4-GA architecture improves comprehensive transmission reliability by over 20% in high-interference environments with up to 30% packet loss. Quantitative analysis shows that the strategy achieves a 20.9% reduction in average processing latency, consistently converging end-to-end delay to within 40 ms while maintaining a malicious stream interception rate above 99% against replay and brute-force attacks.