ST-GraphRCA: A Root Cause Analysis Model for Spatio-Temporal Graph Propagation in IoT Edge Computing.

Area of Science:

• Computer Science
• Distributed Systems
• Artificial Intelligence

Background:

• Massive Internet of Things (IoT) data requires distributed microservices at the edge.
• Identifying anomaly root causes in these edge microservices is a significant operational challenge.

Purpose of the Study:

• To propose a novel spatio-temporal graph propagation model, ST-GraphRCA, for root cause analysis in IoT edge environments.
• To address challenges of time-series asynchrony and efficient fault localization.

Main Methods:

• Developed a PCA-DTW hybrid feature extraction with dynamic alignment for asynchronous time-series data.
• Constructed a stream-based forward propagation graph using flow conservation principles.
• Integrated dynamic edge weights and anomaly scores for causal reasoning and fault pathway inference.
• Employed a topology-constrained high-utility mining algorithm for filtering root cause candidates.

Main Results:

• ST-GraphRCA achieved an F1-Score of 0.89, outperforming existing methods.
• Demonstrated an average localization time of 238.8 ms in resource-constrained edge scenarios, a six-fold improvement.
• Successfully filtered unreachable service combinations to identify low-frequency, high-risk root causes.

Conclusions:

• ST-GraphRCA offers an efficient solution for anomaly fault tracing in large-scale IoT systems.
• Provides technical support for intelligent operation and maintenance of distributed microservice systems.
• Validates the effectiveness of spatio-temporal graph propagation for edge computing anomaly detection.