A Hybrid Heuristic Approach Using Deep Q-Network for Crowdsourcing Resource Scheduling Optimization

The Crowdsourcing Resource Scheduling Problem (CRSP) is an important application scenario in the Multi-Skill Resource Constrained Project Scheduling Problem (MS-RCPSP). Since crowdsourcing resource scheduling involves key challenges such as resource constraints and skill matching, its core issues are highly similar to those of MS-RCPSP. This paper proposes a hybrid heuristic deep Q-network (HH-DQN) method based on deep reinforcement learning to solve the resource scheduling problem in the crowdsourcing test environment. This method utilizes the deep Q-network to construct an intelligent decision-making model, integrates task processing time, resource availability and dependency relationships into the state representation, and combines heuristic algorithms, especially the shortest Processing Time (SPT) strategy, to guide the exploration of the complex scheduling solution space. The experiments demonstrate that relative to traditional heuristics and deep reinforcement learning approaches, HH-DQN exhibits stronger robustness and effectiveness on task sets of different scales, significantly shortens the project completion time, and improves the computing efficiency and resource utilization rate.