Swarm intelligence based deep learning model via improved whale optimization algorithm and Bi-directional long...

Area of Science:

• Chemical Engineering
• Process Control
• Artificial Intelligence

Background:

• Chemical production processes are inherently complex and high-risk.
• Effective fault diagnosis is crucial for ensuring operational reliability and safety.
• Existing methods may struggle with complex, noisy data from chemical processes.

Purpose of the Study:

• To propose a comprehensive and robust fault diagnosis method for chemical production.
• To enhance the accuracy and efficiency of fault detection and classification.
• To improve the overall safety and reliability of chemical manufacturing operations.

Main Methods:

• Data preprocessing using time-varying filtering empirical mode decomposition (TVF-EMD) for noise reduction.
• Dimensionality reduction via kernel principal component analysis (KPCA) on fault data.
• Fault classification using a bi-directional long short-term memory (BiLSTM) network optimized by an improved whale optimization algorithm (WOA) for hyperparameter tuning.

Main Results:

• The proposed method effectively preprocesses raw data, reducing noise and dimensions.
• The optimized BiLSTM network demonstrates high accuracy in fault data classification.
• Validation through two case studies confirms the approach's efficacy and superiority over existing methods.

Conclusions:

• The integrated approach of TVF-EMD, KPCA, and WOA-optimized BiLSTM offers a powerful tool for chemical process fault diagnosis.
• This method significantly enhances the reliability and safety of chemical production.
• The study provides a valuable contribution to intelligent manufacturing and process safety.