Integrated Modeling of Transfer Learning and Intelligent Heuristic Optimization for Steam Cracking Process.

Area of Science:

• Chemical Engineering
• Process Optimization
• Computational Chemistry

Background:

• Steam cracking plants face challenges with computational expense and time for model optimization due to plant expansion and feedstock diversification.
• Existing numerical models for molecular refining and intelligent manufacturing are limited in industrial application by high computational costs and data demands.

Purpose of the Study:

• To develop a high-performance optimization process for steam cracking furnaces adaptable to various feedstocks.
• To overcome the limitations of existing models in terms of computational expense and time consumption.

Main Methods:

• A transfer learning structure based on reaction network motif features was designed for effective prediction.
• A hybrid genetic algorithm and particle swarm optimization (GA-PSO) method was employed for coil outlet temperature (COT) curve optimization.
• The optimization process integrates transfer learning with heuristic algorithms for enhanced performance.

Main Results:

• The proposed method successfully optimizes furnaces for diverse feedstocks, considering different product pricing policies.
• The optimization results are influenced by the weight coefficients of product prices.
• Yield distribution patterns and reaction mechanisms provide further explanation for the obtained results.

Conclusions:

• The integrated transfer learning and heuristic algorithm approach offers a computationally efficient solution for steam cracking furnace optimization.
• This method facilitates molecular refining and intelligent manufacturing in industrial settings.
• The approach provides a framework for optimizing complex chemical processes based on economic factors and mechanistic understanding.