deep learning–driven adaptive framework for precision greenhouse management based on species-specific needs

Conventional greenhouse automation systems often rely on static, setpoint-based controls that overlook species-specific physiological needs, limiting efficiency across diverse environments. to address this gap, a deep learning–driven adaptive framework is proposed for dynamic optimization of key environmental parameters, including air temperature, relative humidity, soil moisture, and nutrient concentration, tailored to individual plant species. the model employs a hybrid convolutional neural networks and long short-term memory (cnn-lstm) architecture, integrating real-time multispectral imagery with temporal sensor data to enable continuous monitoring and species-aware control. simulation experiments with four species, tomato, lettuce, basil, and orchid, under normal, heatwave, and cold-humid scenarios demonstrate superior performance compared to conventional model predictive control and proportional-integral-derivative (pid) frameworks. the proposed method improved growth performance, reduced plant stress, and enhanced control stability, highlighting its robustness under abiotic stress. this study establishes a new paradigm for cognitive greenhouse management, enabling real-time, species-specific optimization that enhances crop productivity and resource-use efficiency in sustainable agriculture.