Document Type : Research Paper
Authors
Department of Agricultural Machinery Engineering, Faculty of Agriculture, University of Tehran, Karaj, Iran.
Abstract
Keywords
Main Subjects
EXTENDED ABSTRACT
Population growth, increased demand for food, and climate change have placed significant pressure on water resources. Agricultural production accounts for 85% of global water consumption. One effective solution to this challenge, which has garnered substantial attention over the past few decades, is greenhouse cultivation. Various models, including dynamic models, black-box models, and machine learning models, have been developed to predict and control greenhouse environmental conditions and support plant growth. Dynamic models, based on mass and energy equations, govern the internal environment of greenhouses. Recent advancements in artificial intelligence have enabled smart control of greenhouse microclimates, focusing on maximizing production while reducing costs. Deep learning-based models represent cutting-edge methods capable of analyzing large volumes of data with complex patterns. This research aims to predict environmental conditions inside greenhouses using artificial intelligence algorithms and to compare these predictions with traditional dynamic models. The outputs will be assessed against actual conditions, environmental changes, and disturbances.
The deep neural network designed to simulate the environmental conditions inside the greenhouse incorporates various inputs, including wind speed, virtual sky temperature, cumulative outside global radiation, outside photosynthetically active radiation, outside temperature, outside relative humidity, and outside carbon dioxide concentration. The outputs of the network are temperature, humidity, and carbon dioxide concentration within the greenhouse. Python serves as the foundational programming language for developing deep learning models, enabling the design of complex network architectures with improved precision and efficiency. This research utilized a dataset focused on the environmental and control parameters of tomato plants. The tomato seeds were cultivated in stone wool beds using hydroponic techniques. An innovative deep neural network structure was proposed in this study to achieve automatic microclimate control in greenhouses.
The developed deep learning-based model predicted the temperature, relative humidity, and carbon dioxide concentration inside the greenhouse with coefficients of determination of 0.81, 0.61, and 0.85, respectively. The temperature in the greenhouse was predicted with good accuracy by the deep neural network model, demonstrating better results with less variation. Additionally, the relative humidity was maintained within a lower range, effectively preventing fungal growth. A similar trend was observed between deep learning models and dynamical mathematical models in predicting CO2 concentration. Artificial intelligence (AI) models offer several advantages, including the ability to make decisions over an unlimited forecasting horizon without relying on prior models, high inherent stability, adaptability, and low complexity in real-time training. AI models can analyze vast amounts of input data and identify complex patterns that may not be easily discernible using traditional system dynamics approach. This capability enables accurate predictions and optimized control of greenhouse environmental conditions, facilitating improved plant growth. AI models can adapt to and learn from new data, continuously enhancing their predictive capabilities over time. Their dynamic and responsive predictions support better decision-making processes. Ultimately, deep learning models can conduct a comprehensive analysis of multiple variables simultaneously, considering numerous influential factors such as temperature, humidity, light levels, and carbon dioxide concentration. This approach not only reduces costs but also enhances overall efficiency.
This research proposed a deep learning-based approach to predict greenhouse environmental conditions for tomato production. The exceptional performance of the deep neural network model was attributed to its structural design for pattern recognition and the precise data collected from a greenhouse managed by expert operators. Dynamic models play a crucial role in illustrating the effects of parameters on the greenhouse microclimate and in modeling product growth. They are especially valuable when setting up a greenhouse in a new geographical location without the availability of expert knowledge. In the future, other deep learning models and optimization algorithms could be employed to reduce production costs and optimize energy consumption by utilizing data from various geographical locations.
Conceptualization, H.G., S.S.M. and R.A.; methodology, H.G.; software, H.G.; validation, H.G. and R.A.; formal analysis, H.G.; investigation, H.G. and R.A.; resources, H.G. and R.A.; writing—original draft preparation, H.G., S.S.M., R.A. and M.H.; writing—review and editing, R.A. and M.H.; supervision, R.A. and S.S.M.; project administration, R.A.; funding acquisition, R.A..
Data sets generated during the current study are available from https://doi.org/10.4121/uuid:88d22c60-21b3-4ea8-90db-20249a5be2a7.
The authors express their gratitude to the University of Tehran for providing essential resources to facilitate this research.
The study was approved by the Ethics Committee of the University of ABCD (Ethical code: IR.UT.RES.2024.500). The authors avoided data fabrication, falsification, plagiarism, and misconduct.
The author declares no conflict of interest.
)