LIP6 2002/020
- Thesis
Modélisation de raisonnements tenus en contexte. Application à la gestion d.incidents sur une ligne de métro - L. Pasquier
- 281 pages - 07/19/2002- document en - http://www.lip6.fr/lip6/reports/2002/lip6.2002.020.pdf - 11,518 Ko
- Contact : Laurent.Pasquier (at) nulllip6.fr
- Ancien Thème : SYSDEF
- Keywords : Context, Contextual Graphs, Knowledge representation, Cognitive activity modelling, Incremental knowledge acquisition, Context-based intelligent support system, Application to subway traffic regulation, Incident management
- Publisher : Ghislaine.Mary (at) nulllip6.fr
The work presented in this thesis was realised within the SART project, which aims at the realisation of a decision support system for traffic regulation. We worked especially on the decision support system for incident management in subway systems.
Incident management is a complex activity combining diagnosis and action. This activity is framed by procedures which must be adapted to the real situation by the operators. We observe a large diversity of solutions depending on who proposed them and on the context of the incident. The different practices obtained this way enrich the knowledge of each member of the operators' community of practice. The reasoning followed by the operators is highly linked to the knowledge they have about the situation and its evolution during the incident management. Classical representations such as rule-bases or decision trees are limited and can neither handle the quantity of practices and contextual elements useful to make the choices, nor their rapid evolution due to new experiences. Moreover the links between diagnosis and action and the dynamical aspect of the choices are not taken into account. Thus we developed a compact and explicit representation based on the context and its evolution, called contextual graph. Contextual graphs take the richness of the situation and its evolution into account. They represent a structure in which diagnosis and action are deeply linked. We also developed an algorithm to integrate new practices and to capitalise operators' experiences. Those graphs, associated with a classical task decomposition, permit modelling operators' activities and their evolution, giving a cognitive ergonomics semantics to our model. The results obtained with the prototype confirmed the efficiency of this model and its acceptance by the operators.
Incident management is a complex activity combining diagnosis and action. This activity is framed by procedures which must be adapted to the real situation by the operators. We observe a large diversity of solutions depending on who proposed them and on the context of the incident. The different practices obtained this way enrich the knowledge of each member of the operators' community of practice. The reasoning followed by the operators is highly linked to the knowledge they have about the situation and its evolution during the incident management. Classical representations such as rule-bases or decision trees are limited and can neither handle the quantity of practices and contextual elements useful to make the choices, nor their rapid evolution due to new experiences. Moreover the links between diagnosis and action and the dynamical aspect of the choices are not taken into account. Thus we developed a compact and explicit representation based on the context and its evolution, called contextual graph. Contextual graphs take the richness of the situation and its evolution into account. They represent a structure in which diagnosis and action are deeply linked. We also developed an algorithm to integrate new practices and to capitalise operators' experiences. Those graphs, associated with a classical task decomposition, permit modelling operators' activities and their evolution, giving a cognitive ergonomics semantics to our model. The results obtained with the prototype confirmed the efficiency of this model and its acceptance by the operators.