4 Presentations

• 
  01:20 PM - 01:45 PM

  Data Science for Real-Time Management of Electricity Supply and Demand

  • Olivier Milon, presenter,

  Energy movement management as an essential and economic service at Hydro-Québec requires highly precise electricity demand forecasting to support decision-making. This level of service involves calculations of increasing quantity and quality that benefit from recent innovations in data science and computing, such as deep neural networks, massive data storage tools, optimization methods, user interfaces, and modern programming languages like Python and Julia. This work takes place in a business context that is changing with the pace of energy transition, where data modeling plays a decisive role, as does the presentation of forecasts and their derivatives to control operators of Hydro-Québec's electrical transmission network.
  This presentation will provide an overview of the business context for demand forecasting before surveying the methods and tools implemented from 2018 to the present as part of the Demand Forecasting Improvement (APD) project and the associated research and development work. Demonstrations of calculations and graphical representations will be complemented by numerical results developed as performance criteria for these services.

• 
  01:45 PM - 02:10 PM

  Modernization of an algorithm for electricity demand forecasting in Quebec

  • Pierre Borie, presenter, Université de Montréal / DIRO
  • Fabian Bastin, Université de Montréal
  • Alain Marcotte, Hydro-Québec
  • Stéphane Dellacherie, Hydro-Québec

  An accurate forecast of electricity consumption is at the heart of Hydro-Québec’s operations, the main electricity supplier in Quebec. Currently, short-term demand is estimated using a parametric model whose parameters are calibrated through nonlinear least-squares optimization, relying on a Fortran 77 library. While this model has been successfully used for over thirty years, its maintenance has become increasingly complex, highlighting the need for modernization.
  We present a reimplementation of the least-squares optimization method in Julia, which provides greater reliability and readability compared to the original library. We discuss the mathematical foundations of the algorithm and compare the results and performance of our implementation with those of the Fortran 77 version. To conduct this comparison, we performed several numerical experiments on test problems using models currently operated by Hydro-Québec. The preliminary results show strong agreement in both intermediate computations and final outputs. Additionally, we introduce algorithmic extensions of a new method aimed at improving the efficiency of the calibration process.

• 
  02:10 PM - 02:35 PM

  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.

• 
  02:35 PM - 03:00 PM

  Dynamic optimal combination of forecasting models

  • Clémence Granade, Université de Montréal
  • Fabian Bastin, presenter, Université de Montréal
  • Stéphane Dellacherie, Hydro-Québec
  • Alexandre Deslauriers, Hydro-Québec

  In many real-world forecasting problems, we have access to several error-minimization models whose performance may depend on the current state of the studied system. It is often possible to improve prediction performance by merging these models together in an appropriate way, which depends on the system conditions. Therefore, the optimal model combination must be dynamically updated according to the available data and target.

  We review various approaches that allow us to reduce prediction error, measured by a given performance indicator such
  as the root mean-squared error, and formulate a general mathematical programming framework to dynamically find the optimal combination of forecasting models with respect to the chosen indicator. We then analyze the mathematical properties of the resulting optimization problem and the solution techniques required to obtain solutions. Since the data distribution is expected to evolve over time, we further discuss how to eciently construct the training sample along with the training and validation samples.

  We illustrate the proposed methodology using short-term electricity demand forecasting, relying on real data and
  machine-learning models provided by Hydro-Québec.