A Production-Logistics prediction method integrating Spatial-Temporal features in flexible production workshop for buffer allocation problem

To meet the demands of personalized manufacturing, characterized by customized production with varying batch sizes, logistics equipment such as Automated Guided Vehicles (AGVs) play a critical role in the manufacturing process. However, the distribution of multiple batches is influenced by various factors, with buffer zone capacity allocation emerging as a key challenge. Optimizing buffer zone allocation necessitates a thorough consideration of both spatial characteristics (e.g., shop floor layout and workpiece pathways) and temporal characteristics (e.g., the sequence of material distribution) to enhance resource allocation, reduce bottlenecks, and improve efficiency. This research proposes a novel logistics prediction method for flexible production plants, utilizing a graph attention network that integrates spatial–temporal features. The method first applies a multi-head attention mechanism to capture significant temporal information. Then, a graph convolutional network is employed to model the workshop layout topology and workpiece processing paths, thereby extracting the spatial features of logistics. This spatial information is sequentially processed through a gated recurrent unit and the multi-head attention mechanism to capture the dynamic temporal features of logistics. The proposed model is ultimately employed to predict production logistics in a flexible manufacturing workshop. The experimental results of the MA-T-GCN (Multi-head Attention Temporal Graph Convolution Network) model on production logistics prediction demonstrate an improvement over the best-performing baseline methods on standard benchmark metrics under varying experimental conditions.