AutoFarming: Autonomous production control of greenhouse farming system using Reinforcement Learning
Greenhouse is an important protected horticulture system in providing fresh food for the growing global population. However, it is also a big resource consumer, such as the large amount of energy usage from LED lighting and heating to maintain ideal growing climate for the crops. The main objective of greenhouse production control is realizing resource-effective crop growth and development through operating indoor climate actuators (e.g. lighting, heating, CO2 dosing, ventilation, screening, watering). In high-tech greenhouses, the modern sensing, control and computing systems are equipped but the operating set-points for the climate actuators are still rely on growers. However, the number of experienced growers is rare while the scale of greenhouse production system is expanding worldwide. Therefore, using advanced techniques to control the greenhouse production system with more optimality and autonomous in a reliable and robust way is in need nowadays challenge.
Reinforcement learning is the kind of learning system, which uses continuous feedback to adjust its own actions to obtain the best. It is also a dynamic control strategy which can update automatically the current control algorithm (policy) through incorporating newly developed knowledge learning from historical and real time data. The motivation of this project is developing autonomous production control of greenhouse farming systems using reinforcement learning, aiming to optimize the production system with desired growing climate, efficient resource usage, and at the same time adaptability to variable conditions (amongst individual plants, species and growing stages).
In this project, you will develop control-oriented crop-climate model based on crop/climate sensor measurements; develop autonomous control scheme of greenhouse production systems using reinforcement learning to achieve efficient crop growth/yield and energy usage. All the proposed approaches will be validated in real greenhouse for lettuce production. The performance will also be explored and analysed with state-of-the-practice.