Intelligence analysis of membrane distillation via machine learning models for pharmaceutical separation.

Area of Science:

• Environmental Engineering
• Chemical Engineering
• Computational Science

Background:

• Membrane distillation (MD) is a promising technology for pharmaceutical separation.
• Accurate prediction of solute concentration is vital for optimizing MD processes.
• Machine learning offers advanced tools for modeling complex separation phenomena.

Purpose of the Study:

• To computationally simulate pharmaceutical separation via membrane distillation.
• To evaluate the efficacy of Multi-layer Perceptron (MLP), Gamma Regression, and Support Vector Regression (SVR) models in predicting solute concentration.
• To optimize model hyper-parameters using the Red Deer Algorithm (RDA).

Main Methods:

• Computational simulation of membrane distillation in continuous mode.
• Implementation of mass transfer and machine learning models.
• Focus on solute concentration in the feed section of the membrane.
• Hyper-parameter optimization via the Red Deer Algorithm (RDA).

Main Results:

• The Multi-layer Perceptron (MLP) model demonstrated superior accuracy (R²=0.9955, MAE=0.0084, RMSE=0.0148).
• Gamma Regression showed acceptable performance (R²=0.9214), suitable for skewed data.
• Support Vector Regression (SVR) exhibited the lowest performance (R²=0.8710) but captured general trends.

Conclusions:

• Multi-layer Perceptron is the most effective regression model for simulating pharmaceutical separation in membrane distillation.
• Model selection and hyper-parameter optimization are critical for accurate computational analysis of membrane processes.
• Machine learning combined with computational simulation enhances the understanding and optimization of pharmaceutical separation technologies.