Method and system for recognizing abnormal behavior of multiple agents
By performing trajectory fragmentation and feature extraction on the operational data of multi-agent systems, a posterior probability distribution is generated to predict the probability of action reward and calculate the risk deviation index. This solves the problem of insufficient robustness of multi-agent systems in identifying abnormal behavior in highly dynamic environments and achieves accurate anomaly warning and diagnosis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGXI TRANSPORTATION SCI & TECH GRP CO LTD
- Filing Date
- 2026-03-16
- Publication Date
- 2026-06-16
AI Technical Summary
Existing multi-agent systems struggle to effectively distinguish between normal exploratory behavior and potential risks in highly dynamic and adversarial environments, exhibiting insufficient robustness and failing to effectively identify abnormal behavior using traditional methods.
By collecting operational data from a multi-agent system, performing trajectory fragmentation processing, extracting latent features using a permutation-invariant encoder, generating a posterior probability distribution, combining it with a distributed reinforcement learning value network to predict the probability distribution of action rewards, calculating a risk deviation index, constructing an anomaly judgment comprehensive vector, and realizing hierarchical early warning for multiple types of abnormal states.
In complex collaborative environments, it can accurately identify abnormal behaviors caused by communication errors, decision-making logic errors, or actuator failures, improve fault diagnosis accuracy, and provide reliable early warnings for road traffic safety monitoring and autonomous driving systems.
Smart Images
Figure CN122221092A_ABST