Graph Neural Network Root Cause Analysis for Large-Scale Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing root cause analysis (RCA) algorithms require significant human intervention and domain knowledge, are not suitable for large-scale networks, and struggle with predicting the effects of faults accurately.
Innovation Solution
A method utilizing a graph neural network (GNN) to perform root cause analysis in a network by generating feature vectors from KPI data, creating relationship data, and classifying nodes as candidate root cause or victim nodes based on graph information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deterministic RCA algorithms (DT, SVM) are used, then performance on high dimensional space is improved, but requirement for pre-defined rules and manual labelling increases
Solution Approach 1:
The system performs self-service by automatically generating feature vectors from raw KPI data and relationship data from network topology, eliminating the need for manual labelling and pre-defined rules. The GNN model learns patterns autonomously from the data, making the RCA process self-sufficient without human intervention for feature engineering or rule creation.
Solution Approach 2:
The patent replaces the mechanical system of manual rule definition and labelling with an intelligent system using GNN. Instead of manually creating deterministic rules for RCA, the system uses neural network-based learning to automatically discover patterns and relationships in the data, substituting human-driven mechanical processes with automated intelligent computation.
2Reliability
If bayesian networks are used, then ability to deal with uncertainty is improved, but requirement for priori knowledge and tree-structured BNs increases
Solution Approach 1:
The GNN-based system performs self-service by automatically learning probability distributions and dependencies from the data without requiring pre-defined priori knowledge. The model autonomously captures uncertainty and relationships through its neural network architecture and training process, eliminating the need for manual Bayesian network construction and prior specification.
Solution Approach 2:
The patent changes the parameters from fixed priori knowledge in Bayesian networks to learnable parameters in the GNN model. The neural network dynamically adjusts its internal parameters during training to capture uncertainty and dependencies, replacing the static parameter requirement of Bayesian networks with adaptive parameter learning.
3Measurement precision
If clustering based RCA algorithms are used, then performance on small datasets with well-defined labels is improved, but applicability to large-scale networks without label data deteriorates
Solution Approach 1:
The GNN-based RCA system achieves universality by functioning effectively across different network scales and data conditions. It can handle both small and large-scale networks, with or without label data, by learning from the inherent structure and relationships in the KPI and topology data. The same model architecture adapts to various network sizes and data availability scenarios.
Solution Approach 2:
The patent replaces the mechanical system of clustering-based algorithms that require well-defined labels with a neural network-based GNN that learns patterns from raw data. This substitution enables the system to handle large-scale networks without label data by automatically discovering relevant features and relationships through the graph neural network's message-passing mechanism.
Data Source
AI summary
A method for root cause analysis in a network comprising a set of nodes Ni for i=1 to N, where N>2. The method includes obtaining N sets of KPI data, each one of the N sets of KPI data being for one of the N nodes. The method also includes, for each one of the N nodes, using the set of KPI data associated with the node to generate feature vectors for the node. The method also includes generating relationship data using the feature vectors, the generated relationship data, indicating relationships between the nodes within the set of N nodes. The method also includes inputting to a GNN the generated relationship data and the feature vectors. The method also includes obtaining from the GNN information indicating that at least node Nj is a candidate root cause node and at least node Nk is a candidate victim node, where k≠j. The method further includes using the relationship data to i) determine whether to indicate the candidate root cause node Nj as a predicted root cause node and/or ii) determine whether to indicate the candidate victim node Nk as a predicted victim node.