一种基于LLM推理架构的作业调度优化方法
By employing a scheduling optimization method based on an LLM inference architecture, and leveraging cognitive blueprints and multi-agent collaboration, the limitations of existing technologies in scheduling and cross-task generality in complex industrial scenarios are addressed, achieving efficient adaptive scheduling and cross-task knowledge transfer.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGDONG UNIVERSITY OF FOREIGN STUDIES
- Filing Date
- 2026-06-22
- Publication Date
- 2026-07-17
AI Technical Summary
Existing scheduling optimization techniques suffer from limitations in structural feature representation, lack of cross-task versatility, and low efficiency of black-box search when facing complex and ever-changing industrial scenarios, thus failing to meet the adaptive scheduling needs of diverse industrial scenarios.
A job scheduling optimization method based on LLM inference architecture is adopted. A generalized scheduling cognitive architecture search system is built using labeled scheduling problem instances. Through cognitive blueprint construction module, intelligent candidate generation module, cognitive deliberation module and architecture evaluation module, adaptive scheduling for complex scenarios is achieved.
It achieves the discovery of the optimal graph neural network architecture within 50 iterations, significantly improving search efficiency, possessing cross-task generalization ability, and enabling zero-sample or few-sample transfer in unseen scenarios, thus reducing search costs.
Smart Images
Figure CN122414769A_ABST