Cloud-edge collaborative data anomaly detection in industrial sensor networks.

Area of Science:

• Industrial Internet of Things (IoT)
• Sensor Networks
• Data Science

Background:

• Industrial sensor networks face challenges with large data volumes and complex spatial-temporal features.
• Centralized anomaly detection models cause heavy traffic loads, leading to communication delays and data loss.
• Existing methods struggle to comprehensively analyze both spatial and temporal features for accurate anomaly detection.

Purpose of the Study:

• To develop a cloud-edge collaborative anomaly detection approach for industrial sensor networks.
• To reduce traffic load by filtering data at the edge.
• To enhance detection accuracy by effectively extracting spatial and temporal features.

Main Methods:

• A cloud-edge collaborative architecture with edge-deployed detection models (Gaussian and Bayesian) and a cloud-deployed analysis model.
• Edge models filter non-anomalous data, reducing traffic to the cloud.
• Cloud model utilizes a Graph Convolutional Network (GCN) with Long Short-Term Memory (LSTM) for spatial-temporal feature extraction.

Main Results:

• The proposed approach significantly reduces traffic load compared to centralized methods.
• The GCN-LSTM model effectively extracts complex spatial and temporal features.
• Experimental results on public datasets show superior performance over baseline anomaly detection models.

Conclusions:

• The cloud-edge collaborative approach offers an efficient and accurate solution for anomaly detection in industrial sensor networks.
• This method addresses the limitations of centralized detection and enhances the reliability of industrial IoT systems.
• The integrated GCN-LSTM model demonstrates strong capabilities in analyzing complex sensor data features.