Simulation-Based Optimization over Discrete Spaces Using Projection to Continuous Latent Spaces.

Area of Science:

• Computational Chemistry
• Process Systems Engineering
• Machine Learning

Background:

• Simulation-based optimization of complex systems with discrete decision spaces presents significant computational challenges due to combinatorial explosion.
• Existing methods struggle with the vast number of possible alternatives, making exhaustive simulation infeasible.
• Efficiently navigating these discrete spaces is crucial for designing complex chemical processes.

Purpose of the Study:

• To develop a novel computational approach for optimizing complex systems with discrete decision spaces.
• To reduce the computational burden associated with simulation-based optimization.
• To enable efficient navigation of complex design spaces for chemical process engineering.

Main Methods:

• Transformation of discrete decision spaces into a continuous latent space using Variational AutoEncoders (VAEs).
• Integration of VAEs with Bayesian optimization (BO) for efficient exploration of the latent space.
• Application of the proposed method to design complex distillation systems.

Main Results:

• The proposed VAE-based transformation effectively projects discrete decision spaces into a continuous latent space.
• Bayesian optimization in the latent space significantly reduces the number of required expensive simulations.
• Successful application demonstrated in case studies for designing distillation systems, including caprylic acid recovery and extractive dividing wall columns.

Conclusions:

• The integration of Variational AutoEncoders and Bayesian optimization offers a powerful and efficient solution for simulation-based optimization of complex systems with discrete decision spaces.
• This approach significantly alleviates computational challenges, paving the way for improved design of chemical processes.
• The method shows practical utility in designing challenging separation systems, highlighting its potential for broader applications in process engineering.