Application of Virtual Sample Generation and Screening in Process Parameter Optimization of Botanical Medicinal...

Area of Science:

• Chemical Engineering
• Biotechnology
• Process Modeling

Background:

• Small sample size is a persistent challenge in chemical, biological, and medical industries, hindering accurate process modeling and optimization.
• Difficulty in data acquisition exacerbates small sample problems, reducing predictive accuracy for critical parameters.

Purpose of the Study:

• To address the small sample size problem in process modeling using virtual sample generation.
• To optimize extraction rate models for botanical medicinal materials by improving parameter prediction accuracy.

Main Methods:

• Proposed a virtual sample generation method utilizing full factorial design (FFD) combined with initial experiments from Box-Behnken Design (BBD).
• Developed a virtual sample screening technique employing an extreme learning machine (ELM) to refine generated data.

Main Results:

• Virtual sample data effectively increased sample size, leading to improved model precision.
• Models trained on semi-synthetic samples (experimental and virtual) demonstrated higher accuracy than those using only experimental data.

Conclusions:

• The proposed virtual sample generation and screening methods successfully resolve small sample modeling challenges.
• Optimizing models trained on semi-synthetic data enables the reliable determination of process parameters.