Attention-Based Temporal Convolutional Networks and Reinforcement Learning for Supply Chain Delay Prediction and Inventory Optimization

Accurately predicting supply chain delays is crucial for maintaining resilience in inventory management systems. This paper presents a novel framework integrating an Attention-Based Temporal Convolutional Network (ATCN) and Reinforcement Learning (RL) to address the challenges of supply chain delay prediction and inventory optimization. The ATCN model leverages convolutional layers to capture long-term dependencies in time series data, while the attention mechanism enhances prediction accuracy under high demand volatility. Additionally, the RL component optimizes inventory decisions by minimizing holding and shortage costs through multi-agent collaboration. Our proposed approach outperforms baseline models in terms of Mean Absolute Error (MAE), Mean Squared Error (MSE), R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>, and Area Under the Curve (AUC) metrics. Ablation studies confirm the importance of both the attention mechanism and RL in improving forecasting accuracy and inventory management. This work offers a robust solution for enhancing supply chain resilience and lays the groundwork for future exploration in multi-layered and real-time supply chain systems.