Abstract
Agriculture, and market gardening in particular, is a complex business that requires both theoretical knowledge and experience to produce successfully. Market gardeners need to be familiar with the principles of vegetable production and technical cropping itineraries, i.e. all the tasks involved in growing vegetables efficiently, as well as the mechanisms between crops and their impact on plots in order to plan both temporally and spatially. This theoretical and technical knowledge needs to be combined with practical knowledge of the field and the context, so that planning choices can be adapted both to the crops and to the associated tasks. The complexity of technical cropping itineraries and planning requires modelling using decision support systems to facilitate the planning work of market gardeners. Different types of artificial intelligence models can be used to address specific planning problems. We have decided to focus on the problem of crop rotation, i.e. which crop should follow another in order to construct crop sequences that comply both with known rules (theoretical and technical knowledge) and are also context-dependent (practical knowledge). To address this problem, we propose two approaches: (i) A semantic approach based on the construction of a domain ontology, called Crop Planning and Production Process Ontology (C3PO), aimed at representing the knowledge needed to plan and manage crops from farm to production. (ii) A learning approach consisting of implementing sequential learning models to predict the next crop following a given sequence based on historical data or experience. Our work has shown that these two approaches are complementary, because modelling using semantics makes it possible to provide part of the agronomic knowledge during the planning choices, and the learning models can learn part of the market gardeners' rotation choices so that they can propose them to others. Finally, we are seeking to combine the two approaches in order to build a decision support system for market gardeners that is both knowledge-based and experience-based. We tested our work in the context of the development of the Elzeard application, a digital companion developed by the eponymous company with which this research work was carried out.