As intelligent manufacturing systems increasingly pursue mass customization and rapid market responsiveness, efficient resource allocation has become a critical determinant of production competitiveness. Dynamic resource scheduling in such systems involves complex challenges stemming from highly variable market demand and complex production conditions. This study extends a multi-modal data fusion approach that incorporates deep learning methods to solve dynamic resource scheduling with variable demand scenarios. The approach involves the fusion of sensor, production, and market data using intermediate fusion methods, the adoption of an LSTM with attention model to predict demand, and Double Deep Q-Network (DDQN) methods to control resource scheduling decisions. Simulation outcomes show that the system achieves 82.6% resource utilization and 89.1% on-time delivery performance in nominal scenarios; however, it performs poorly during peak hours, leading to severe delays accounting for 12.8% of total deliveries. The multi-modal fusion approach outperforms single-source methods by 9.6% in prediction accuracy but exhibits limitations in cross-industry generalization (67.9% efficiency retention) and computational efficiency (18–35 min for large-scale scenarios). These findings provide realistic insights into the practical deployment of AI-driven scheduling systems in manufacturing environments.
Xiaobo Wu (2026) studied this question.