Optimization of Spirulina sp. cultivation using reinforcement learning with state prediction based on LSTM neural network

Spirulina/Arthrospira is well known as a microscopic filamentous cyanobacterium, which is rich in nutrition and minerals. Therefore, Spirulina/Arthrospira cultivation industry has been adopting many optimization methods for enhancing biomass productivity. Recently, machine learning has been an emerging approach thanks to their efficacy. In this paper, a novel time-dependent reinforcement learning (RL) method with state prediction by long short-term memory (LSTM) networks is proposed to optimize the dry-weight yield of Spirulina sp. HH cultivation. The simulation results show that under the same light condition, with the proposed algorithm, the Spirulina sp. HH cultivation system produces a yield 17% higher than that of the traditional cultivation method and 10% higher than that of the threshold-based method. The results of the RL method from this study promise a significant benefit in Spirulina farm production when applying it for proactively planning biomass production and enhancing profit. Graphical abstract