Recently, unicellular microalgae have attracted considerable interest due to their ability to utilize CO2 in the photosynthetic process and produce various valuable compounds such as biofuels, bioplastics, pigments, and vitamins. Microalgae are a promising source for a sustainable replacement for fossil fuels in transportation and power generation. Their efficient use solar energy utilization and capacity to generate energy-rich biofuels through advanced thermochemical techniques contribute to this. An overview of the microalgae cultivation process for the production of different products is shown in Figure 1.
Figure 1: Microalgae cultivation principle, downstreaming with different bioproducts obtained
Microalgae can be cultivated under phototrophic conditions, utilizing either artificial lighting indoors or natural sunlight outdoors. Closed photobioreactors (PBRs) for outdoor cultivation have been regarded as a feasible choice for large-scale microalgae cultivation. This method utilizes natural sunlight within a controlled, enclosed system, reducing the possibility of contamination. Nevertheless, the biological processes involved in outdoor cultivation become complex due to variations in light and temperature, which are fundamental factors for microalgae cultivation. In this work, we investigate the influence of sunlight dynamics on the growth of Phaeodactylum tricornutum. Growth models play a crucial role in understanding the mechanisms of microalgal growth and improving production efficiency. Based on first principles, several conventional models have been proposed to describe the growth dynamics of microalgae concerning light intensity. However, most of these models need further validation in the context of outdoor cultivation. In addition, the existing models omit the light acclimation phenomenon, which describes how microalgae cells adjust their biochemical structures in response to varying light intensities to improve their light utilization.
The light acclimation effect is important for the growth of microalgae, affecting factors such as changes in the chlorophyll a to carbon ratio, the composition of accessory pigments, the abundance of photosynthetic proteins, the parameters of the photosynthesis-irradiance curve, and the coupling between light absorption and electron transfer. In higher light conditions, microalgae decrease their pigment content to maintain a balance between light absorption for energy and the requirements for growth and carbon fixation. Consequently, the incorporation of the light acclimation effect into growth models is crucial for accurate predictions of microalgal growth dynamics. Numerous researchers have attempted to incorporate the light acclimation effect into growth models, aiming to improve the accuracy and applicability of these models. Nevertheless, the majority of these models have been developed using controlled indoor laboratory experiments and have not been verified in outdoor cultivation settings utilizing natural sunlight. Therefore, validated growth models that consider the light acclimation effect in outdoor cultivation of microalgae are required.
Machine learning models have attracted considerable attention from various fields in recent decades. In contrast to traditional models that rely on explicit mathematical equations, machine learning algorithms are designed to obtain knowledge from data, thus classifying them as data-centric models. Although machine learning has shown success in areas related to microalgae, such as genetic engineering, species identification, and medium screening, its use for growth modeling in microalgae cultivation remains limited. In addition, existing studies have not adequately considered the effect of light acclimation. In order to consider the influence of light acclimation, we integrated a time sequence of light-related information as an additional input into the machine learning models. This allowed the prediction of microalgae growth to account for the history of light exposure. We employed support vector regression (SVR) and long short-term memory (LSTM) networks for addressing the regression task, given their widespread use in time series analysis in machine learning.
In the SVR and LSTM models, we used a light history for growth prediction within the machine learning frameworks. The LSTM and SVR models were trained using the current biomass (X) and the light history time series data (I) as inputs. The results demonstrate that the LSTM model performs exceptionally well in predicting microalgae growth within the test dataset compared to SVR. However, the SVR model shows signs of overfitting as its performance on the training dataset significantly outperforms its performance on the testing dataset.