Data Tech 2026

Effective inventory management across multiple regions is often hindered by stochastic demand, inter-regional dependencies, and lead-time uncertainties. In this study, we introduce a spatially-aware reinforcement learning framework (SAR-Q-Learning) that integrates region-specific demand forecasting with multi-agent Q-learning to optimize inventory decisions across a network of warehouses. Using the UCI Online Retail dataset, we estimate autoregressive demand dynamics for each region and incorporate spatial correlations to guide policy learning. Empirical results show that the SARQ-Learning agent successfully balances service levels, holding costs, and stockouts, achieving higher mean rewards and more stable final inventories compared to naive or independent policies. Visual analyses of weekly sales, AR(1) parameters, training convergence, and forecasted inventory trajectories demonstrate the model’s capability to capture complex temporal and spatial patterns. These findings underscore the potential of spatially-informed reinforcement learning for robust, data-driven inventory control in multi-regional supply networks