Towards A Digital Twin For Agriculture: Modeling Of Complex Processes For Monitoring, Prediction And Control In Greenhouse Farming

Ten years ago, industrial processes underwent a significant change due to the prominent application of digital technologies in various forms, which greatly helped boost productivity and safety. In recent times, digital technologies have reached the agricultural sector, where they appear to be the solution for higher yields, better quality, and resource optimization. Despite such premise, however, nowadays, digital technologies are poorly used in this field, as they are often simply relegated to a way of collecting and processing data. This research shows that benefits provided by computer science span over ones already used by integrating advanced modeling and control methodologies from systems engineering to monitor, predict, and control complex processes in a greenhouse context. First, we address and solve the problem of microclimatic variability within different zones by developing a modeling technique based on soft sensors to predict location-specific temperature starting from a centralized reading, thus removing the need for physical sensors. Secondly, throughout the usage of Hybrid Automata, we re-implement the well-known tomato model TOMGRO using the mentioned formalism alongside a pathogen model for Oidium Lycopersici, which, thanks to automata formalism, we manage to compose with the tomato model. Next, we optimize control over such automata thanks to the supervisory control theory, which allows us to automatically synthesize supervisors for a system given the requirement, or by taking an already existing parametric controller and optimize it via simulation. Finally, we put all the pieces together using soft sensors to specialize climatic readings to act as input to different pairs of tomato-pathogen automata dislocated in a field, where the pathogen can grow and spread to neighbors, propagating the infection. Meanwhile, the supervisor tries to keep the temperature value within a favorable range for crops. The presented models have been validated using actual data retrieved from fields, and simulation models’ outputs have been proved to be trustworthy by comparing results with actual ones retrieved from past experiments.