Hydro-Québec is taking the lead in using artificial intelligence (AI) in order to transform their load forecasting and grid management practices. The box utility company has adopted multiple AI technologies to develop its forecasting capabilities. Now, though, it can do short-term prediction within a 36-hour window as well as hourly predictions up to 42 days out—all entirely powered by historical weather data. This initiative aims to bridge the gap between forecasting and actual energy consumption, particularly during critical periods when errors can prove costly.
Thus, the incorporation of AI into Hydro-Québec’s operations came after an extensive evaluation of its outdated forecasting models. During a historic heatwave on May 22, 2024, the oldest models found it difficult to forecast load patterns with precision. This failure has made it clear that there’s an immediate need to modernize. The utility plans to use AI to improve their operational efficiency. Beyond introducing this innovation, one of the main aims is to prevent future energy supply disruptions.
Hydro-Québec is implementing a robust strategy to deploy AI solutions across more than 350 substations. Even with this deployment, they will have to wait until 2028 as the first piece in their long-term vision. Their track record is somewhat complex. Their company has been very aggressive in research investments. Their self-proclaimed mission is to make load forecasting smarter and help manage the grid better.
Advancements in Forecasting Techniques
Hydro-Québec employs AI for multiple forecasting horizons. For day-to-day operations, it uses AI to generate short-term predictions up to 36 hours in advance. Further, the utility uses hourly forecasts for up to 12 days in advance guided by meteorological data. For medium to long-range predictions, Hydro-Québec applies “historical normal” weather values. These values assist in developing models that forecast future energy requirements for long-term planning.
The shift to AI-based forecasting comes as utilities are realizing the cost and operational challenges posed by VRE are rapidly increasing. David Adkins, an expert in grid operations, noted that “the increasing penetration of renewables introduces variability and uncertainty into grid operations, making traditional forecasting and control methods less effective.” By adopting AI, Hydro-Québec aims to address these obstacles and improve its forecasting precision.
Aroon Vijaykar, Hydro-Québec’s AI commercial business lead, emphasized the importance of this shift: “Load forecasting has always been an important function. AI is both a challenge and an opportunity on both sides of that equation.” This lens highlights why AI integration is actually a double-edged sword. It adds complexity to a process that’s already complex, but it provides an opportunity for transformational improvements.
Lessons from Legacy Models
Hydro-Québec originally developed their legacy forecasting models using opaque algorithms, with a high number of complex non-linear constraints and parameters. These models employ the Easy Nonlinear Least-Squares Inequality Programme (ENLSIP). They’ve continually been upgraded, but they continue to fail under unusual load patterns. Clermont, a specialist at Hydro-Québec, explained the limitations: “With experience, you have a lot of historic curves, so our mathematical models are quite good for regular patterns. Then counterintuitively, it turns out. It might be raging storms, or something you’ve never seen coming, and boom, now your model is broken.”
As the recent heatwave illustrated, these legacy systems are inadequate. Even worse, they misjudged the drop in expected grid load. This event led Hydro-Québec to look for more advanced AI-based solutions that were more responsive to atypical contexts.
Vijaykar stated that “utilities are understandably conservative about the system,” highlighting the cautious approach that many utilities take when adopting new technologies. He further explained, “Unlike the tech industry, which can afford to move fast and break things, for utilities one mistake can be catastrophic.” Hydro-Québec’s started its AI journey with a proof of concept. Initially, they deployed a neural network at just one substation, but they soon were able to broaden its use case applications across the entire grid.
Future Goals and Continuous Improvement
Hydro-Québec intends to make the most of its AI expertise. It’s planning to do this by using data from more than four million smart meters. The utility has released groundbreaking proposals for research and development projects scheduled to take place in 2026 and 2027. They will work primarily on providing more effective access to smart meter data and developing a renewable energy forecasting prototype.
The company’s ultimate goal is to get to autonomous grid operations. This will provide the visibility needed to optimize energy flows, balance the grid, and easily integrate distributed generation and storage solutions. Vijaykar expressed confidence in this direction: “At a minimum, AI can help to run systems more efficiently. At a maximum, it will avoid catastrophic blackouts.” These policy declarations speak directly to Hydro-Québec’s intention to improve grid reliability through cutting-edge technological development.
Even more, the approach promotes teamwork between AI-driven orchestration and state-of-the-art grid planning and forecasting. Adkins clarified that “AI complements but does not replace grid planning and forecasting.” He reiterated that effective integration requires market signals and human operational oversight, positioning AI as part of a broader framework for resilience and flexibility in grid management.