AI cuts energy use and boosts CO2 absorption in spirulina cultivation, UNIST team says

Researchers have developed a technology that could enable the efficient mass production of microalgae, a promising alternative food source for a world grappling with food shortages.

A research team led by Im Han-kwon, a professor at the Ulsan National Institute of Science and Technology (UNIST) Graduate School of Carbon Neutrality, announced Thursday that it had developed an AI control algorithm for spirulina cultivation based on multi-agent reinforcement learning, in collaboration with a team led by Riezka Andika, a professor at the University of Indonesia.

Spirulina is a microalga with a protein content of up to 60 percent and is rich in essential amino acids, vitamins and minerals. It can grow rapidly in cultivation facilities rather than on farmland, drawing attention as a health food and alternative protein source — but making it commercially viable has proven difficult. Boosting output while cutting the energy consumed by lighting and heating and cooling systems is a persistent challenge, as temperatures and sunlight levels vary widely by region and season, and growth conditions such as light, temperature, nutrients, carbon dioxide and acidity all interact with one another. Supplying more CO2 can accelerate growth, but it can also acidify the culture medium and inhibit photosynthesis and development; stronger artificial lighting aids photosynthesis but drives up electricity consumption.

The newly developed algorithm deploys five AI agents, each responsible for one of five conditions — light, temperature, nutrients, carbon dioxide and acidity. Together they adjust lighting brightness, heating and cooling output, and the amount of each substance supplied in response to cultivation status and local climate, raising spirulina growth rates while reducing energy use.

The team first built a virtual cultivation environment that calculates how growth and culture conditions change across seven variables — light, temperature, acidity, nutrients, CO2 and oxygen concentrations, and spirulina biomass — then trained the five AI agents through reinforcement learning to produce the algorithm.

The team compared the AI-based approach against rule-based control, which operates equipment according to preset criteria. Performance was evaluated using seasonal weather data from eight cities representing distinct climates: Depok, Indonesia; Ulsan; Riyadh, Saudi Arabia; La Paz, Bolivia; Oslo, Norway; Beijing; Salt Lake City; and Cape Town, South Africa.

The results showed that under AI control, spirulina concentration reached the harvest threshold assumed in the research 59 hours sooner than with the conventional method. Energy required to produce 1 kilogram of dried spirulina fell by an average of 14 percent — with reductions of up to 20 percent in Ulsan, where seasonal temperature swings are large, and 21 percent in Beijing. CO2 absorption increased by an average of 78 percent, and the share of supplied CO2 actually absorbed by the spirulina rose from 44 percent to 49 percent.

"Since we have confirmed performance in a virtual environment, we plan to validate the technology in actual cultivation facilities equipped with sensors and control devices, and develop it into something that can be used in agricultural and fisheries settings," Im said.

The findings were published in Bioresource Technology, an international journal in the field of agricultural engineering.