Predicting pellet metallurgical performance is critical for optimizing industrial smelting processes,yet traditional methods face computational bottlenecks when handling large-scale material datasets.This study propos...Predicting pellet metallurgical performance is critical for optimizing industrial smelting processes,yet traditional methods face computational bottlenecks when handling large-scale material datasets.This study proposes an HPC-optimized hybrid model integrating XGBoost and multilayer perceptron(MLP)architectures.By implementing batch-optimized memory hierarchies and cache-aware data partitioning,we efficiently process a large amount of feedstock ratio data and metallurgical performance metrics from industrial production cycles.Experimental results demonstrate superior accuracy in predicting RDI,ΔT,RI,and RSI indices compared to single-model approaches.The proposed framework provides a scalable solution for real-time performance prediction in smart manufacturing systems,reducing computational overhead through dynamic load balancing across HPC nodes.展开更多
文摘Predicting pellet metallurgical performance is critical for optimizing industrial smelting processes,yet traditional methods face computational bottlenecks when handling large-scale material datasets.This study proposes an HPC-optimized hybrid model integrating XGBoost and multilayer perceptron(MLP)architectures.By implementing batch-optimized memory hierarchies and cache-aware data partitioning,we efficiently process a large amount of feedstock ratio data and metallurgical performance metrics from industrial production cycles.Experimental results demonstrate superior accuracy in predicting RDI,ΔT,RI,and RSI indices compared to single-model approaches.The proposed framework provides a scalable solution for real-time performance prediction in smart manufacturing systems,reducing computational overhead through dynamic load balancing across HPC nodes.
