Reduction du risque barrage par l'utilisation des reseaux Bayesiens

No one can predict risk with certainty because two of its main characteristics are complexity and uncertainty. Even when a dam is judged to be safe and is monitored and operated in a responsible way, there still exists a residual risk that is always present if population or property are located in the potential flood zone. Dam failures are not common but they always have important human, material and environmental consequences. The establishment of risk-reduction measures is therefore an important part of the responsibilities of a dam owner.; Risk is a danger that implies a well-planned action. The main objectives of this research project are first to establish this danger, which is defined in relation with the probability of failure of the dam, and second to determine the most appropriate action to reduce it.; The existing risk analysis methodologies do not allow the adequate prioritization of dam safety interventions and do not take into account the complexity and the uncertainties that characterize risk. A literature review and the realization of risk analyses for embankment dams in Hydro Quebec have shown that these methodologies present limitations related to the calculation and representation of risk in absolute terms. Important difficulties have also been identified in relation to the modelling of the effects of interventions on a dam, the interrelations between the various failure mechanisms and the lack of flexibility of the existing tools for analysis.; The use of bayesian networks is proposed to overcome the identified weaknesses of the existing tools and to capitalize on their merits. A bayesian network is a causal graph where the variables of the considered system are represented by nodes and their dependencies by directional links. The underlying probabilistic representation allows the quantification of the strength of these dependencies and the realization of inferences to aid decision-making. These calculations are based on Bayes' theorem which can be considered as the mathematical expression of learning from experience.; The concepts put forward in this research allow the determination of an updated picture of dam risk which takes into account the numerous interrelations between failure mechanisms as well as the uncertainties and the expert judgements that are always present in the analyses.; The possibilities offered by the proposed approach allows the realization of multiple forms of queries including previsions and diagnostics used, among others, to identify the most probable causes of a dam failure. (Abstract shortened by UMI.)...