Deep Reinforcement Learning for Alarm Causality Mining
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current causal mining methods rely on rule-based statistical approaches that fail to accurately represent causality due to incorrect rule representation and require expert experience, leading to invalid causality and large hypothesis spaces.
Innovation Solution
A method and device utilizing deep reinforcement learning to build a system alarm environment from system alarm information and root cause label data, enabling an agent to interact and learn an alarm causality model that accurately represents causality without manual rules.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If rule-based statistical methods are used for causality mining, then the mining process can be implemented with existing algorithms, but the causal relationship mined becomes invalid when rules cannot correctly represent causality
Solution Approach 1:
The patent replaces rule-based statistical methods with deep reinforcement learning algorithms to mine causal relationships. The DRL agent learns causal patterns from alarm data through environmental interactions, eliminating the need for manual rule construction and achieving both high reliability in causality detection and automated implementation.
2Device complexity
If co-occurrence frequency and timing are used to mine causality, then the mining process is simplified, but only correlation between variables can be mined, not causality
Solution Approach 1:
The patent substitutes simple co-occurrence-based correlation analysis with deep reinforcement learning that can distinguish causal relationships from spurious correlations. The DRL agent learns from alarm sequences and temporal patterns to accurately represent causal structures while maintaining computational efficiency.
3Adaptability or versatility
If expert experience is required to provide variable causal hypotheses, then the hypothesis space can be constrained, but difficulty in obtaining variable causal hypotheses arises
Solution Approach 1:
The patent implements self-service by enabling the deep reinforcement learning agent to autonomously generate causal hypotheses without expert intervention. The agent explores the alarm environment, learns causal patterns, and generates hypotheses automatically, eliminating the need for manual expert input while maintaining hypothesis quality.
4Adaptability or versatility
If random causal hypothesis graphs are used for verification mining, then the hypothesis space is covered broadly, but the hypothesis space becomes too large
Solution Approach 1:
The patent replaces random hypothesis generation with deep reinforcement learning-based hypothesis mining. The DRL agent systematically explores the alarm environment and learns causal patterns, generating a manageable and high-quality hypothesis space that covers essential causal relationships without the computational burden of exhaustive random sampling.
Data Source
AI summary
A method for mining an alarm causality, a device for mining an alarm causality, and a storage medium. The method for mining the alarm causality includes: building a system alarm environment (101) for deep reinforcement learning based on system alarm information and root cause label data of the system alarm information; and learning and generating an alarm causality model (102) representing the alarm causality and structure through an interaction between a deep reinforcement learning agent and the system alarm environment.