|
Short-term production planning of a hydropower cascade combined with variable energy sources is an optimization problem. The hydropower production plan is constructed within a time horizon of a few days to comply with operational requirements and to maximize revenues obtained by selling the total production from all energy sources on the day-ahead market and by penalizing imbalances.This PhD thesis focuses on the optimization problem considering data uncertainty on water inflows, electricity prices and variable generation, modeled by a finite set of multivariate scenarios. The proposed optimization problem in the stochastic framework is then written with a two-stage stochastic dynamic mixed-integer linear programming formulation.Since computation time is a crucial issue for an operational use, the optimization problem is solved by replacing the value function of the second stage with a surrogate model fitted by supervised learning during a pre-processing step. The surrogate model is fitted on input data that are simplified by specific approaches. The domain of the input data related to the decision variables of the first stage is first estimated by analog-based sampling, and the functional inputs are reduced by principal components analysis. The learning data set can then be obtained from a design of experiments using latin hypercube sampling. Several linear models are finally proposed for the surrogate model.The proposed methodology is tested on a real but simplified case study. The surrogate model obtained by linear regression on the reduced inputs provides acceptable performance, and it can be used to get a solution to the optimization problem in the stochastic framework. Since variable generation is not considered in the case study, the stochastic framework provides, however, minor profit compared to the deterministic framework. Besides, computation time is not compatible with an operational use without distributed computing. Several research tracks are then proposed to improve the methodology. A validation ...
|
