Soft Sensing of LPG Processes Using Deep Learning

Area of Science:

• Chemical Engineering
• Data Science
• Industrial Process Control

Background:

• Oil refineries rely on periodic sampling for quality control, leading to delays.
• Complex distillation processes require efficient monitoring for optimal performance.
• Accurate estimation of component concentrations is crucial for product specifications.

Purpose of the Study:

• To integrate soft sensors and deep learning for improved monitoring in oil refining.
• To develop models for estimating C2/C5 content in LPG and distillation column energy consumption.
• To enhance predictive accuracy and efficiency in de-ethanization and debutanization processes.

Main Methods:

• Development of soft sensor models using deep learning techniques.
• Implementation of Artificial Neural Network (ANN) and Random Forest Regressor (RFR) models.
• Testing models with real refinery operational data, addressing scalability and data quality issues.

Main Results:

• Successfully estimated critical variables like C2 and C5 content in LPG.
• Accurately predicted energy consumption of distillation columns.
• Demonstrated the effectiveness of deep learning models in refinery applications.

Conclusions:

• Soft sensors and deep learning offer a powerful solution for real-time monitoring in oil refineries.
• The developed models show high applicability and potential for replication in similar industrial settings.
• Emphasis on model interpretability and online learning capabilities for continuous improvement.