Hierarchical Graph Neural Network for Root Cause Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing root cause analysis methods are inefficient in identifying root causes in complex, interdependent systems as they focus on single networks and do not account for causation, making it difficult to diagnose system failures in large-scale, distributed systems without extensive domain knowledge.
Innovation Solution
A hierarchical graph neural network method is developed to train causal graphs from time series data, enabling the identification of interdependent causal links and propagating system failures to determine the most probable root causes through topological and individual cause scoring, without requiring prior domain knowledge.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual root cause analysis is performed in complex systems, then diagnostic accuracy can be achieved through domain expertise, but the process becomes time-consuming and labor-intensive
Solution Approach 1:
The patent replaces manual mechanical analysis processes with an automated computational system. A graph neural network model automatically processes system metrics data to identify root causes, substituting human expert manual analysis with machine learning-based automated causal inference, thereby reducing time loss while maintaining identification accuracy
Solution Approach 2:
The system enables self-service root cause analysis by automatically processing system metrics and generating causal graphs without requiring manual domain expertise intervention. The neural network model autonomously performs causal inference and identifies root causes from system data, making the diagnostic process self-executing
2Ease of operation
If existing root cause analysis methods are applied to single isolated networks, then analysis simplicity is maintained, but they cannot effectively diagnose failures in interdependent complex systems
Solution Approach 1:
The patent merges multiple isolated network analyses into a unified causal graph that captures interdependencies across the entire system. The graph neural network integrates data from multiple interconnected networks, combining their causal relationships into a comprehensive model that reliably identifies root causes in complex interdependent systems while maintaining operational simplicity
Solution Approach 2:
The causal graph neural network model serves multiple functions: it analyzes individual networks, captures inter-network dependencies, performs causal inference, and identifies root causes across diverse system types. This universal approach handles both simple and complex interdependent systems with a single framework
3Productivity
If statistical correlation methods are used for root cause analysis, then computational efficiency is improved, but causation cannot be established leading to incorrect root cause identification
Solution Approach 1:
The patent substitutes statistical correlation methods with a graph neural network-based causal inference system. The model learns causal relationships from system metrics data through automated causal discovery algorithms, replacing correlation-based approaches with causation-oriented analysis that maintains computational efficiency while improving causal relation accuracy
Solution Approach 2:
The system changes the analytical parameters from correlation coefficients to causal strength metrics. The graph neural network computes causal inference scores and causal graphs that quantify causal relationships rather than mere correlations, transforming the measurement parameters to accurately reflect causation while maintaining analysis speed
Data Source
AI summary
A method is provided for training a hierarchical graph neural network. The method includes using a time series generated by each of a plurality of nodes to train a graph neural network to generate a causal graph, and identifying interdependent causal networks that depict hierarchical causal links from low-level nodes to high-level nodes to the system key performance indicator (KPI). The method further includes simulating causal relations between entities by aggregating embeddings from neighbors in each layer, and generating output embeddings for entity metrics prediction and between-level aggregation.