Methodology for Controlling Greenhouse Microclimate Parameters and Yield Forecast Using Neural Network Technologies

Developments in the combination of technologies for data analysis, sensors, and self-driving vehicles are very popular nowadays. They use network solutions, control systems, platforms, and applications. A significant area of application of neural networks is product quality control connected with environmental factors. Artificial intelligence technologies allow real-time monitoring of microclimate parameters. This makes it possible to subsequently influence the general state of the grown products and report any problems detected in real-time. Systems using artificial intelligence technologies can operate 24/7. Therefore, the creation of a methodology using neural network technologies is intended to control the parameters of the greenhouse microclimate, which should have a positive influence on the quality of the harvest and increase yields. After analyzing the accumulated information on mathematical models of greenhouses, a new model was created that is applicable for calculating coolant consumption, steam consumption, and carbon dioxide emissions. The optimal greenhouse microclimate conditions were also analyzed and determined: temperature, humidity, and carbon dioxide concentration, according to which it is possible to predict the yield. A structural schema of the greenhouse control and monitoring system is proposed. The necessary and sufficient components of the system for maintaining the microclimate in the greenhouse for growing oyster mushrooms were identified and selected, such as temperature and humidity sensor, carbon dioxide sensor, circulation pump, ultrasonic humidifier, and ventilator. Thus, a neural network was developed to predict yields and control microclimate parameters such as temperature, humidity, and carbon dioxide content. Based on the mathematical model, a program was designed to train the neural network. For an accurate forecast, a neural network was developed that is based on a multilayer perceptron with three hidden layers.