Anomaly Data Detection Method Based on Neural Network for Dual-link Aggregated Communication

Power grid communication is critical within the power system, directly impacting its stability and operation. To enhance anomaly detection accuracy in power grid communication networks, this paper introduces a dual-mode neural network anomaly detection algorithm. It involves real-time data collection through a dual-mode communication system, preprocessing via wavelet transformation, training and prediction using two deep convolutional neural network sub-models, and combining results via a soft voting method. Comparative experiments across seven datasets demonstrate the method's superiority in accuracy, false positives, and missed detections over traditional approaches, making it more suitable for real-world anomaly detection scenarios.