Master's thesis and Internship : Stochastic Optimization of Hydroelectric Production in the Massif Central: Application to Day-ahead and Intraday auctions.

Abstract

The urgent need to reduce carbon dioxide emissions in order to mitigate global warming has led, and will continue to lead, to a rapid increase in renewable energy sources. Among these, hydropower occupies a unique position: unlike wind and solar generation, it is both flexible and storable, which gives it a distinctive role in short-term electricity markets. However, hydropower bidding in short-term electricity markets requires decisions under uncertainty in both prices and inflows. In current operational practice, deterministic optimization methods are used, while the management of uncertainty is handled manually by human experience and market intuition. This thesis investigates whether stochastic optimization provides a more systematic and robust optimization for bidding hydropower plants in the Day-Ahead (DA) market and he Intra-Day Auction 3 (IDA3). The study compares deterministic optimization with two stochastic methods using price and inflow scenarios: a a Single-Stage Stochastic Programming (SSSP) and a Stochastic Dual Dy- namic Programming (SDDP) approach. Two bidding strategies are analyzed: must-sell bids derived directly from the production plan, and exclusive blocks constructed either from deterministic schedules or from SDDP. The methods are evaluated on an existing french hydropower plant, using a rolling horizon and historical market data from 2025. Results show that when bidding in the day-ahead market, stochastic methods outperforms deterministic optimization in most cases, confirming the value of explicitly modeling uncertainty. When Intra-Day (ID) readjustment through IDA3 is allowed, all methods benefit from increased flexibility, and in this setting SDDP continues to outperform the alternatives. The systematic use of exclusive blocks yields mixed results: it is advantageous when additional water can be sold profitably, but less effective when water should be conserved. Overall, SDDP generates more cash while respecting reservoir boundaries more reliably than the other methods, and it demonstrates clear added value compared to current operational practice, increasing revenues by 21% in a six-month backtest.