Contribution à l'amélioration des politiques d'opération des turbines hydrauliques

Abstract

In the domain of hydroelectric production, operation and maintenance are two aspects on which an active research is carried out in order to reduce costs. The two aspects are modelled on very different horizons: the operation is planned over hours while the maintenance is planned over years. This difference in scale led to the separation of the research on operation and on maintenance. However, several recent studies pointed out that the precision of the operation model as a great influence on the optimality of the maintenance plan. Due to the uncertainty in the demand, it is hard to plan the operation with precision for a long horizon. Thus, we propose in this thesis to handle this problem from the other side and to adapt the operation policy to the predefined maintenance plan. The actual representation of the operation relies on a complex system for which the optimal resolution relies on combinatorial optimization. Therefore, the optimal resolution on a long horizon is inaccessible. In order to still have the realism of the representation of the operation, we propose a simulator for the operation of a hydro-electric grid, with an interval of the hour, over an arbitrary length of time. As it is in reality, we can change our operating policy on each day, and the resolution of the day is done without prior knowledge over the following days. We also introduce the definitions for new measures to i) validate the suitability of our simulation relatively to reality and ii) quantify the influence of our operational decisions over the progression of the degradations of the turbines. In a first step, we investigate the long-term sensitivity of our system concerning the production capabilities. Once the absence of noticeable detrimental consequences has been ascertained, we propose to examine the margin for the progression of the degradations, specifically the ability to accelerate or delay the degradation of selected units independently and cohesively. Finally, we expose an extension of this approach to adapt operation on-line to obtain given degradation levels at specified instants.

Date	20 Dec 2019
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Antoine Tahan (Supervisor) & Michel Gamache (Co-supervisor)