一种基于残差神经网络与时空特征重塑的涤纶工艺参数配置与优化方法
By combining residual neural networks and spatiotemporal feature reshaping with Bayesian optimization algorithms, the problems of long process parameter configuration cycles and low accuracy in polyester production were solved, achieving rapid and safe process parameter optimization and significantly improving production efficiency and product quality.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHEJIANG SCI-TECH UNIV
- Filing Date
- 2026-04-20
- Publication Date
- 2026-07-17
AI Technical Summary
The current polyester production process parameters rely on human experience, resulting in long R&D cycles, high costs, and low control precision. Furthermore, conventional deep learning algorithms, due to their large hysteresis characteristics, extract spurious correlation features and lose spatial topological information during continuous chemical production, outputting dangerous parameters that violate physical common sense.
A method based on residual neural networks and spatiotemporal feature reshaping is adopted. By using time-series backtracking offset and two-dimensional feature matrix reshaping, a surrogate model is constructed. Then, a Bayesian optimization algorithm with physical penalty terms is used for reverse iterative optimization to generate the optimal process parameters.
It has enabled rapid and safe optimization of polyester process parameters, shortened the R&D cycle by 97%, reduced production costs by 30%, improved parameter configuration accuracy to within 0.1%, and significantly improved product quality stability.
Smart Images
Figure CN122065689B_ABST