Forecasting for fermentation operational decision making

Area of Science:

• Biotechnology and Biochemical Engineering
• Industrial Fermentation Processes
• Data Analytics in Bioprocessing

Background:

• Predicting future behavior in batch and fed-batch fermentation is crucial for optimizing product consistency and profitability.
• Operational policy adjustments and scheduling of downstream processing rely on accurate forecasting of batch productivity and end times.
• Existing forecasting methods require evaluation for their efficacy in complex bioprocessing environments.

Purpose of the Study:

• To contrast forecasting performance between multivariate batch statistical data analysis and case-based reasoning (CBR).
• To investigate the impact of statistical data prescreening and case selection metrics on forecasting accuracy.
• To evaluate these methods in industrial settings, including pharmaceutical and beer fermentation.

Main Methods:

• Application of multivariate batch statistical data analysis, specifically linear projection to latent structures (PLS).
• Implementation and evaluation of case-based reasoning (CBR) for process forecasting.
• Statistical data prescreening and analysis of case selection metrics were performed.

Main Results:

• Case-based reasoning (CBR) demonstrated comparable forecasting performance to PLS in a simpler batch beer fermentation.
• For the more complex fed-batch pharmaceutical fermentation, CBR exhibited superior forecasting performance over PLS.
• Appropriate statistical prescreening of data was found to be important for achieving optimal forecasting results.

Conclusions:

• Case-based reasoning (CBR) is a viable and effective method for forecasting fermentation processes, particularly complex ones.
• The choice of forecasting method (CBR vs. PLS) and data preprocessing significantly impacts performance.
• Accurate forecasting enables better process control, scheduling, and resource management in industrial fermentation.