A dynamic scheduling framework for co-working space resources optimized by multi-agent reinforcement learning

Efficient resource allocation in co-working spaces is essential for optimizing utilization and user satisfaction. Traditional scheduling methods often lead to suboptimal resource use and conflicts, particularly during peak hours. Standard approaches and single-agent reinforcement learning techniques are inadequate for adapting to dynamic multi-user environments and coordinating multiple resources simultaneously. This study develops a dynamic intelligent framework for real-time scheduling that balances diverse resource demands and user preferences. The proposed Multi-Agent Proximal Policy Optimization Actor-Critic Scheduling (MAPPACS) integrates multi-agent proximal policy optimization with actor-critic mechanisms, enabling collaborative decision-making among agents representing different resources. Agents model desks, rooms, and equipment, each learning allocation policies through environmental interaction. A sample of 2300 items from the dynamic co-working space resource dataset was examined, which includes information on resource availability, booking duration, current occupancy, peak hours, and previous disputes, is used in this research to examine how well different prediction models perform. One-hot encoding for categorical features and outlier detection for numerical features ensure stable and meaningful input representations. Independent Component Analysis (ICA) extracts independent patterns from correlated usage features to enhance agent learning efficiency. Coordination is facilitated via multi-agent communication and shared value estimation. Python frameworks are employed for model development and simulation. With an MSE (28.47), RMSE (5.33), R2 (0.87), and MAE (4.12), MAPPACS performs well, showing minimal prediction error and excellent accuracy across all measures. The method offers a robust, adaptive, and scalable solution for dynamic co-working space scheduling, supporting efficient and user-centric resource management.