Intelligent Scheduling Optimization for Flexible Manufacturing Systems

Abstract

This study addresses the challenge of dynamic processing time prediction in intelligent manufacturing systems by proposing a Transformer-based time series prediction model enhanced with prior knowledge weighting. To validate the approach, an intelligent processing unit experimental platform was designed, integrating processing, logistics, and perception modules to simulate flexible manufacturing scenarios. The proposed model employs multi-head attention mechanisms and weighted feature optimization to capture complex temporal dependencies within processing sequences, emphasizing the influence of key materials in each scheduling cycle. Experimental results demonstrate that the weighted Transformer model achieves superior performance compared to conventional neural network architectures, with a coefficient of determination (R²) of 0.989 and significantly reduced mean absolute and root mean squared errors. The results confirm the model’s ability to adapt to time-varying processing conditions, improving scheduling accuracy and overall production efficiency in intelligent manufacturing environments.