Deep Reinforcement Learning Resource Scheduling Algorithm for Intelligent Computing Platform

As intelligent computing platforms face the dual challenges of surging dynamic demand for heterogeneous resources and real-time evolution of task priorities, traditional resource scheduling methods based on static rules are difficult to effectively coordinate the contradictions between computing power allocation efficiency, energy consumption cost and service quality. To address this problem, this study proposes a dynamic hierarchical scheduling framework based on deep reinforcement learning (DRL-HSF). First, a multi-dimensional state representation model is designed and a dynamic resource topology map is constructed. Second, a hierarchical attention mechanism is introduced, and a competitive architecture Dual Deep Q-Network (Dueling DDQN) is used in the global scheduling layer to achieve cross-node resource matching. Finally, a transfer learning mechanism is integrated to enhance the algorithm’s ability to quickly adapt to new task load patterns. Experimental results show that: in terms of basic scheduling performance, DRL-HSF shortens the task completion time to 94.4 s and reduces the resource fragmentation rate to 14.5%; in burst load scenarios, the service quality compliance rate is always maintained above 74%; in the policy migration experiment, the performance recovery rate reaches 96%, showing strong generalization ability and scheduling stability, verifying the practicality and adaptability of the proposed method in complex dynamic environments.