Plant-scale biogas production prediction based on multiple hybrid machine learning technique.

Area of Science:

• Biotechnology and Bioengineering
• Environmental Science
• Machine Learning Applications

Background:

• Biogas production parameters are often nonlinear and imbalanced, hindering accurate prediction with traditional methods.
• Existing machine learning algorithms struggle with data imbalance, leading to suboptimal biogas plant performance prediction.

Purpose of the Study:

• To develop a hybrid extreme learning machine (ELM) model for enhanced biogas production prediction.
• To address data imbalance issues in biogas plant operational data.
• To identify key parameters influencing biogas production.

Main Methods:

• Implementation of a hybrid extreme learning machine (ELM) model incorporating data-balancing techniques.
• Utilizing optimization algorithms to refine model performance.
• Validation of the model using full-scale biogas plant data.

Main Results:

• The optimized ELM model achieved high prediction accuracy (R² = 0.972) on validation data.
• The developed model demonstrated a low prediction error of 2.15% when implemented as software.
• Feed volume (FV) and total volatile fatty acids of anaerobic digestion (TVFAAD) were identified as critical positive influencers of biogas production.

Conclusions:

• The hybrid ELM model effectively improves biogas production prediction accuracy, even with imbalanced datasets.
• The study successfully identified crucial operational parameters for optimizing biogas yield.
• This approach offers a robust solution for accurate biogas plant performance prediction across various operating conditions.