Multi-agent digital twin of broccoli: development and test evaluation

Abstract

Crop production is a complex multi-domain field of knowledge dealing with living objects. Ongoing global climate change has been destroying this long-established sustainable knowledge system. Precise farming needs digital integration of several domain fields: technologies, machinery and equipment, knowledge about plant growth, predicting the impact of activities on crop yield online. Digital twin of plant, mirroring and predicting plant’s state and growth in real-time should be the central element of precision farming system. In the paper, a Smart Plant Digital Twin (SPDT) is proposed as a smart software system with a knowledge base and methods of reasoning. SPDT is developed for online management and simulation of plant behaviour in sync with development of the real plant. A multi-agent implementation of multi-level plant structure is discussed in the paper, which considers crop physiology and resource demand, describing internal processes inside a plant, and a method for calculating the crop parameters and duration of plant development stages based on expert knowledge. Ontological model of SDTP for crop cultivation domain reflects the production process of each field or greenhouse crop and allows the scale up of the number of simulated cultures, specifying the differences between the varieties and cultivars, and widening the effects of agricultural measures. The plant model is illustrated in broccoli growth process. The model was validated in real experiment compared to the growth of real broccoli crops, planted in the Taiwan region, and real data from the field sensors and agronomists from the farms acquired data from their sensors and worked with digital twins of their crops.