Workflow modelling for experimenting a decision system on protection of grapevine against mildews
Résumé
Works initiated in 2001 by our partners phytopathologists at INRA Bordeaux led to the design of a decision system against both powdery and downy mildews of grapevine, which has been named “GrapeMilDeWS”, in French “POD Mildium”. GrapeMilDeWS is decomposed in 7 stages distributed all along the vegetative season. Each leads basically to two decisions made at the plot scale: whether or not to spray against each of the two diseases. Our contribution to system design has been on knowledge elicitation and formal modelling, with Statecharts. Statecharts describe processes with states and events, and are a formal construct built upon finite state automata. Details and complex time behaviour can be described thanks to hierarchy and concurrency mechanisms. The modelling was undertaken in order to produce a fully explicit and logically consistent decision protocol, so that many users would test it on many different plots, in different regions. The protocol now exists in a written form and has been tested on a network that has grown up to 35 plots in 2009. Experimenting on a such a large scale, and getting back data and knowledge from this, causes numerous new challenges: (i) the decision system has to be robust to various conditions, while avoiding two loopholes: too much flexibility in the specification with risk of inappropriate implementation, or too many details (ii) the data to collect are numerous and include weather forecasts that are evolving by nature and need to collected in a timely manner; (iii) people involved are numerous: regional disease risk analysts, decision-makers for each plot, advisers who provide assistance to decision-makers and check conformance of decision to the original protocol. We have thus undertaken to develop an experimentation support system that is based on workflow nets by van Der Aalst (YAWL framework). Workflow nets are derived from Petri nets, which like Statecharts are based on processes with events and states. The emphasis of Petri nets is on management of resources, represented by tokens, and control of concurrent process. With workflow nets, tokens represent cases to handle. The workflow system allows to manage processes and associated data in a consistent manner. The paper is about how GrapeMilDeWS, which we call a Decision Worklow, is being implemented in YAWL. The advantages of workflows nets for this application are discussed in regards to alternatives like Statecharts and Colour Petri Nets.
Domaines
Sciences de l'environnement
Origine : Fichiers produits par l'(les) auteur(s)
Loading...