Back Propagation Neural Network-Enhanced Generative Model for Drying Process Control

Abstract

To improve the control precision and stability of the drying process, this work investigates a drying process control model based on a Back Propagation Neural Network (BPNN). It constructs a data generation model to address the issue of insufficient sample space for process parameters in drying machines. This model includes a complete data generation model structure, integrating the discriminator and generator network structures of the BPNN, with optimized loss functions. The model's performance is validated through experiments, including fit analysis of the generated results and the model’s reliability analysis. The results show that the composite R² value of the data generation model reaches 0.93915 in the fit analysis. This consistency validates the model's ability to accurately fit the global data distribution, reflecting its generalization capability. Additionally, significance analysis reveals that the H values of the process parameters in the datasets generated by the data generation model and the original datasets are all 0, with p-values greater than 0.05. This indicates no significant statistical difference between the two, and confirms the reliability of the data generation model in filling the insufficient sample space. It suggests that the model can effectively enhance the completeness of the dataset without affecting the data distribution characteristics. The findings of this work provide theoretical and practical guidance for optimizing control in the drying process, contributing to improved control precision and stability in industrial drying operations.