Data-Driven Greenhouse Climate Regulation in Lettuce Cultivation Using BiLSTM and GRU Predictive Control
Abstract
Efficient greenhouse management is essential for sustainable food production in response to a growing global population. However, maintaining optimal indoor climates requires significant energy and resources, making advanced control systems critical for economic viability and environmental sustainability. Traditional greenhouse models are often complex and imprecise, limiting the effectiveness of conventional control strategies. To address these challenges, this study investigates data-driven predictive control methods using Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM) neural networks. Our experiments showed that GRU-based predictive control reduced temperature and humidity violations by up to 5\% and required 40\% less computation time than the LSTM approach, all while maintaining equivalent economic performance and crop yield. These findings demonstrate that GRU-based predictive control offers a more efficient and practical solution for real-time greenhouse climate regulation in precision agriculture.