# Surrogate Assisted Evolutionary Multi-objective Optimisation applied to a Pressure Swing Adsorption system

**Authors:** Liezl Stander, Matthew Woolway, Terence L. Van Zyl  
**arXiv:** [2204.12585](https://arxiv.org/abs/2204.12585)  
**Date:** 2022-03-28  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2204.12585v1/

## Key points

- Chemical plant design and optimization often require computationally expensive mathematical formulations and simulation models
- Machine learning surrogate models can be used as substitutes for these expensive models during optimization processes
- This paper explores Surrogate Assisted Genetic Algorithms (SA-GA) in more complex variants of chemical plant design and optimization problems
- The study focuses on including parallel and feedback components in the original plant design, adding to the complexity of the system
- A novel extension called Surrogate Assisted NSGA-2 (SA-NSGA) is proposed to address this challenge
- The effectiveness of this approach is tested on a well-known case, the Pressure Swing Adsorption (PSA) system
- Genetic Algorithms provide an effective framework for optimizing complex chemical plant systems in both single and multi-objective scenarios
- Random Forest surrogate assisted Evolutionary Algorithms can handle increasingly complex chemical systems with parallel and feedback components
- Combining a Genetic Algorithm framework with Machine Learning Surrogate models leads to significant computational efficiency improvements
- The proposed approach achieves a speedup of 1.7 - 1.84 times for increased complexity examples and a 2.7 times speedup for the Pressure Swing Adsorption system
- This research highlights the computational advantages obtained by using a GA combined with Machine Learning Surrogate models for chemical plant design and optimization
- Random Forest SA-GA and SA-NSGA algorithms are flexible and robust, suitable for handling more complex multi-objective systems
- This surrogate-assisted optimization approach should be further verified and tested within industry settings

## Summary

Chemical plant design and optimization are complex tasks that often require computationally expensive mathematical formulations and simulation models. However, recent research has shown the potential benefits of using machine learning surrogate models as substitutes for these expensive models during optimization processes. This paper builds upon previous studies by exploring Surrogate Assisted Genetic Algorithms (SA-GA) in more complex variants of chemical plant design and optimization problems. The study specifically focuses on including parallel and feedback components in the original plant design, which adds to the complexity of the system. To address this challenge, the authors propose a novel extension to the original algorithm called Surrogate Assisted NSGA-2 (SA-NSGA). The effectiveness of this approach is tested on a well-known case in literature, the Pressure Swing Adsorption (PSA) system. Extensive experimentation is conducted to compare various meta-heuristic optimization techniques and machine learning models as surrogates. The results demonstrate that Genetic Algorithms provide an effective framework for optimizing complex chemical plant systems in both single and multi-objective scenarios. Additionally, Random Forest surrogate assisted Evolutionary Algorithms can be successfully scaled to handle increasingly complex chemical systems with parallel and feedback components. One of the key findings of this study is that combining a Genetic Algorithm framework with Machine Learning Surrogate models leads to significant computational efficiency improvements. The proposed approach achieves a speedup of 1.7 - 1.84 times for increased complexity examples and a 2.7 times speedup for the Pressure Swing Adsorption system. Overall, this research highlights the computational advantages obtained by using a GA combined with Machine Learning Surrogate models for chemical plant design and optimization. The flexibility and robustness of algorithms like Random Forest SA-GA and SA-NSGA make them suitable for handling more complex multi-objective systems. The success demonstrated across various chemical systems suggests that this surrogate-assisted optimization approach should be further verified and tested within industry settings. By enabling faster decision-making processes, this approach has the potential to greatly enhance the optimization of chemical plant design and operations.