Developing an artificial neural network model for predicting the growth of Chlorella sorokiniana in a photobioreactor

• Main Authors: Magdaong, J. B., Culaba, Alvin B., Ubando, Aristotle T., Chang, J. S., Chen, W. H.
• Format: text
• Published: Animo Repository 2020
• Subjects: Chlorella sorokiniana; Biomass energy; Microalgae; Photobioreactors; Mechanical Engineering
• Online Access: https://animorepository.dlsu.edu.ph/faculty_research/1885; https://animorepository.dlsu.edu.ph/context/faculty_research/article/2884/type/native/viewcontent
• Institution: De La Salle University

Microalgae have been long considered as a potential source of biofuel. Species such as Chlorella sorokiniana can store large amounts of carbohydrates and lipids which can be used to produce biofuels. This paper demonstrates a method for developing an artificial neural network model which can predict C. sorokiniana growth in a photobioreactor. The data used for training the model came from cultivation experiments conducted at the National Cheng Kung University in Taiwan. A feedforward backpropagation ANN model with three inputs (i.e. aeration rate, biomass concentration, and nitrate concentration) and two targets (i.e. biomass concentration and nitrate concentration after 24 hours) was used for this study. Using MATLAB, multiple configurations of this ANN model were created and tested by varying the number of neurons and hidden layers and the training algorithm. Models were initially assessed in terms of their mean square error (MSE) and training performance plots. The models were then further assessed based on their simulation capabilities. After setting the initial biomass and nitrate concentration and aeration profile, the model can already predict the daily biomass and nitrate concentration of C. sorokiniana for the whole cultivation period. The final model selected has one (1) hidden layer and four (4) hidden neurons and it was trained using the Bayesian regularization backpropagation algorithm. For the final selected model, the calculated mean absolute percentage error (MAPE) for the predicted daily biomass and nitrate concentration were all below 7.59% and 3.68% respectively. Thus, the simulation results showed that the final model can accurately predict C. sorokiniana growth at varying aeration profiles. For future studies, this model can be used to determine the aeration profile that can maximize C. sorokiniana growth in a photobioreactor while minimizing aeration costs. © 2020 Institute of Physics Publishing. All rights reserved.