4 Presentations

• 
  10:30 AM - 10:55 AM

  Modeling the Short-Term Hydropower Production Problem Using Graph Theory

  • Léo Monteiro, presenter, Université du Québec à Chicoutimi
  • Hugo Tremblay, Université du Québec à Chicoutimi
  • Sara Séguin, Université du Québec à Chicoutimi

  In the context of hydropower production, the objective is to model and optimize short-term power production using graph theory. The model is represented as a directed weighted multigraph, where each node corresponds to a possible net head for each period, considering the water inflows, turbine discharge, and spillage. The edges represent discretized turbine discharge values, with weights corresponding to the power generated over each period. The final step involves determining the optimal path through the graph that maximizes total energy production over a given number of periods.

• 
  10:55 AM - 11:20 AM

  Residential Microgrids and Energy Grid Clustering Optimal Power Flow Problem Using Graphon Mean Field Game Theory

  • Hussein Supreme, IREQ
  • Hanane Dagdougui, Ecole Polytechnique de Montréal
  • Antoine Lesage-Landry, Polytechnique Montréal
  • Mohamad Aziz, presenter, Polytechnique de Montréal

  In this work, we investigate the power flow problem in distribution grids with a large number of prosumer households integrating distributed energy resources (DERs). First, we develop a clustering architecture to partition the grid into clusters of residential microgrids (RCMs), each equipped with an aggregator that mediates the energy exchange within the RCM and the communication with other clusters and the main grid. Second, we propose a novel graphon mean field game (GMFG) theory-based approach to address the cluster of clusters power flow optimization problem. This approach formulates the power flow problem as a decentralized dynamic game and derives optimal control strategies that minimize prosumer household energy bills, considering both localized cluster and broader interconnected cluster grid effects.

• 
  11:20 AM - 11:45 AM

  User Experience: A Challenge in Big Data for Demand Forecasting

  • William Roberge, presenter, Hydro-québec / UdS

  The number of forecast points exploited by the demand forecasting team at Hydro-Québec is increasing significantly, necessitating the implementation of advanced analytical tools to monitor model performance. These new tools offer a deep understanding of customer needs, both in terms of the logical relationships between points as seen by the transporter and the geographical dimension of distribution equipment.
  Discover how this new paradigm revolutionizes our approach and allows advisors to have operational awareness at the heart of significant volumetry. The goal is to monitor hundreds of points while tracking the time series of models and observing the relationship with the time series of inputs to analyze model discrepancies. A demonstration of the screens will be made during this presentation.

• 
  11:45 AM - 12:10 PM

  Development of a reinforcement learning model for the short-term hydropower scheduling problem

  • Yoan Villeneuve, presenter, Université du Québec à Chicoutimi
  • Sara Séguin, Université du Québec à Chicoutimi
  • Miguel F. Anjos, University of Edinburgh
  • Kenjy Demeester, Rio Tinto
  • Abdellah Chehri, Royal Military College of Canada

  Hydropower generation plays a pivotal role in the global energy landscape, offering a renewable and sustainable source of electricity. In the province of Québec, Canada, hydropower meets most of the province’s electricity needs and is also utilized by various industries. The aluminum producer Rio Tinto owns and operates a hydropower system in the Saguenay-Lac-Saint-Jean region, supplying electricity to its local aluminium smelters. Therefore, effective scheduling of hydropower operations, particularly at dam facilities, is essential to optimize the water resource usage while meeting energy demands. The Short-Term Hydropower Scheduling (STHS) problem focuses on making operational decisions for daily production, aiming to maximize the efficient use of plant components and available resources. This talk focuses on the development of machine learning models for operational decisions for the STHS problem. The study focuses on the interconnected Chute-du-Diable and Chute-à-la-Savane hydropower plants. This project utilizes a comprehensive dataset spanning over 12 years of hourly data to develop and train a reinforcement learning model to predict hourly water discharge values over a span of 96 hours. To achieve this, a simulation environment is developed to model hourly state transitions, incorporating a reward function tailored to the hydropower plant constraints and industry requirements. Novel concepts, such as efficient points, are implemented to determine the action space in the environment and the energy produced. The decision policy is formulated using conventional reinforcement learning algorithms. The results are compared against historical decision-making and state-of-the-art optimization models using real test instances. This work contributes to the growing interest in literature for machine learning applications in renewable energy, emphasizing the potential for improved decision-making in hydropower production through advanced predictive modelling.