基于长期记忆的智能体长程任务一致性保持方法
By generating memory snapshots and quantifying context drift entropy during long-term task execution by an agent, and using a risk prediction model to calculate risk probability values, the problem of memory promiscuity and drift after external asynchronous interruption is solved, thus improving the stability and reliability of task recovery.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHENZHEN LAIDA SIWEI INFORMATION TECH CO LTD
- Filing Date
- 2026-05-12
- Publication Date
- 2026-07-17
AI Technical Summary
In existing technologies, when an agent performs long-term tasks, the historical memory and the current target node are easily mixed and drifted after an external asynchronous interruption. This makes it difficult to accurately match the effective context during the recovery phase, and it can easily cause state judgment deviations and task execution deviations in complex scenarios, reducing the stability and reliability of long-term tasks of the agent.
During the execution of long-term tasks by the intelligent agent, it responds to external asynchronous interruption events, obtains task execution logs and current state context to generate memory snapshots, converts them into snapshot feature vectors and stores them in a vector memory bank, retrieves historical memory snapshots through similarity, quantifies context drift entropy, calculates risk probability values using a risk prediction model, and generates control commands to drive state transitions or trigger memory cleaning processes to ensure consistency.
It achieves consistency and traceability of task execution during the interruption recovery phase. By quantifying the effect of context drift and entropy increase, it solves the promiscuity and drift problems of agents performing long-term tasks in existing technologies, and improves the stability and reliability of agents performing long-term tasks.
Smart Images
Figure CN122174867B_ABST