A Hybrid Attention-Augmented Convolutional Neural Network for Anomaly Detection in Industrial Wireless Networks under Dynamic Noise Conditions

الملخص

Industrial wireless networks constitute the operational backbone of mission-critical systems in manufacturing and energy sectors. Yet their exposure to cyber-physical attacks and sudden equipment failures poses a grave threat to operational continuity. This paper presents a hybrid methodology that integrates multi-head self-attention mechanisms with deep convolutional layers to address the challenge of anomaly detection under varying noise conditions—a scenario that severely degrades the performance of conventional models. We employ a benchmark dataset replicating a real industrial environment and subject it to varying Signal-to-Noise Ratio (SNR) levels ranging from 5 dB to 25 dB. Our results demonstrate that the proposed approach outperforms baseline models including Bidirectional LSTM and traditional CNNs, achieving a 12.4% improvement in recall under the harshest noise conditions while maintaining an inference latency below 50 milliseconds, rendering it viable for time-critical applications. We further discuss the inherent limitations concerning computational overhead and outline future directions involving knowledge distillation to address these shortcomings.