Improving microalgae growth modeling of outdoor cultivation with light history data using machine learning models: A...

Area of Science:

• Biotechnology
• Algaculture
• Computational Biology

Background:

• Accurate microalgae growth prediction is vital for optimizing cultivation and understanding environmental impacts.
• Existing mathematical models often lack validation in real-world outdoor conditions.
• Light dynamics significantly influence microalgae productivity.

Purpose of the Study:

• To evaluate machine learning algorithms for microalgae growth modeling.
• To compare Long Short-Term Memory (LSTM) and Support Vector Regression (SVR) against traditional Monod and Haldane models.
• To assess model performance in outdoor cultivation settings.

Main Methods:

• Utilized 50-day outdoor cultivation data of Phaeodactylum tricornutum.
• Employed flat-panel airlift photobioreactors for cultivation.
• Compared LSTM and SVR models with Monod and Haldane models.

Main Results:

• Machine learning models (LSTM, SVR) significantly outperformed traditional models.
• The ability to incorporate light history as input was key to ML model success.
• LSTM demonstrated a strong capacity for modeling light acclimation effects.

Conclusions:

• Machine learning offers superior microalgae growth prediction in outdoor cultivation.
• LSTM and SVR can be effectively applied as biomass soft sensors.
• These models facilitate the development of optimal harvest strategies for enhanced microalgae production.