La gestion des centrales hydroelectriques en serie par balancement des reservoirs

This thesis presents our approach for determining a weekly operating policy of a hydropower system consisting of several reservoirs in series using an aggregation-disaggregation method and a balancing rule. The objective function consists in maximizing the difference between the revenues of the energy surplus sold on the spot market and the cost of the energy bought to meet demand. The first step of the method consists in aggregating all installations on the river and then solving the aggregate model with stochastic dynamic programming. The feedback operating policy obtained by solving the aggregate model gives the amount of energy that should be stored at the end of each period. The second step consists of finding a rule and a relation which optimally allocates the entire storage between the reservoirs using a balancing rule. These rules were obtained by applying both, classical linear regression and the Multivariate Adaptive Regression Method (MARS) to results from a deterministic model that maximize hydropower production. Balancing reservoirs means managing the hydroelectric installations so that the total amount of potential energy stored in the system is distributed among the reservoirs of the system. This distribution must take into account the randomness of the reservoir inflows, the efficiency of the installations, and the variations of the water heads and the risk of spillage. This report will show how to determine and improve balancing rules for several reservoirs in series and will compare the results obtained for two and three reservoirs to those obtained with stochastic dynamic programming. The proposed method was also extended to a large scale system and more precisely to seven reservoirs in series.