Hydro-Québec has certainly set out on an ambitious path. As one example, they are advancing their load forecasting capabilities by further incorporating artificial intelligence (AI). With this initiative, we are addressing the growing divide between projected energy needs and real-world energy usage. For one, it only focuses on peak periods, when major miscalculations can lead to operational chaos. The utility company has developed AI capabilities to produce short-term forecasts, working out only 36 hours in advance. In tandem, they’re pushing the envelope with hourly forecasts up to 42 days out by integrating meteorological predictions into their models.
The mega heatwave of May 22, 2024 made it clear that we need to use forecasting in new and improved ways. Hydro-Québec’s legacy models failed, missing the dramatic spike in demand. This incident showcased the limitations in the existing non-linear models built off the current ENLSIP algorithm. Even with modifications along the way, these models were unable to adapt fast enough to changing conditions that were changing quickly. As we know all too well, extreme weather events are increasing in intensity and frequency. Hydro-Québec understood that they couldn’t keep on only counting on traditional models.
Read Hydro-Québec’s strategy for a gradual introduction of AI, with ongoing steps and enhancements going into 2026 and 2027. This 5-year vision includes a lot of work with smart meter data and creating a renewable energy forecasting prototype. By 2028, the utility intends to deploy a bottom-up regional forecasting model. This program will upgrade over 350 substations and help to make the energy grid more resilient.
The Role of AI in Short-Term Forecasting
Hydro-Québec uses AI to deliver short-term load forecasts of up to 36 hours ahead. Similarly, this short-term focus allows operators to take advantage of real-time data to better respond with rapid changes in demand. Furthermore, the incorporation of meteorological forecasts makes it possible to project hourly out to 10 or 12 days. The utility’s goal is to improve accuracy at peak demand times. Even small mistakes during this brief window can pose significant operational headaches.
The company’s AI system has proven its value on days with inclement weather. For example, it was able to flag deviations from typical demand patterns in the above-mentioned June heatwave, encouraging action by operators. FERC rebutted this claim, noting that severe corrections were still required, 1,500 megawatts worth. This seems to point to an urgent need for additional fine-tuning of these AI systems.
Sylvain Clermont, Hydro-Québec’s lead author for the UNECE Task Force on Digitalisation in Energy, emphasized the importance of developing confidence in AI systems.
“When they get confident, Hydro-Québec will stop using the old model,” – Sylvain Clermont
His insights are a reflection of a growing maturity within the organization. With AI models continuing to change, ongoing training and validation of these models will be critical.
The Challenge of Legacy Models
The challenge Hydro-Québec’s current forecasting models are still very much tied to forecasting methodologies that have successfully guided the utility for decades. These black box models use algorithms with hundreds of thousands of parameters that are continuously tuned and modified through machine learning techniques on past data. As climate patterns grow more unpredictable, this has put long-standing systems to the test. Failing to do so, they fail to accurately model the load change for extreme conditions.
Clermont noticed that the models truly excel in finding patterns that are expected. The reality is that they have great difficulty in handling unforeseen, web-breaking events.
“Then comes something totally out of the box – whether extreme weather or something else that you have never experienced in the past – and your model will be off.” – Sylvain Clermont
While traditional forecasting certainly has its place, it falls short in several ways. That’s why Hydro-Québec is innovating with AI as a positive complement, rather than total replacement. Its goal is a hybrid approach, which combines the best of both worlds — historical data and real-time inputs from public, private and crowdsourced sources.
Aroon Vijaykar, Hydro-Québec’s AI commercial business lead, highlighted the industry’s cautious approach towards adopting new technologies.
“Utilities are understandably conservative about the system. Unlike the tech industry, which can afford to move fast and break things, for utilities one mistake can be catastrophic.” – Aroon Vijaykar
Hydro-Québec has a philosophy that emphasizes an iterative approach, where AI solutions are tested in small scales before being fully implemented.
Future Outlook and Goals
Looking ahead, Hydro-Québec’s AI-related objectives focus on realizing ongoing incremental gains in operational efficiency. By harnessing advanced analytics, the utility seeks not only to enhance forecasting accuracy but to proactively manage grid operations. This starts with decreasing operational expenditures and becoming more agile to changing energy ecosystems marked by a greater dependence on renewables.
Vijaykar was clear to highlight the two-sided effects of AI on load forecasting.
“Load forecasting has always been an important function. AI is both a challenge and an opportunity on both sides of that equation.” – Aroon Vijaykar
The utility acknowledges that while AI introduces complexities to load forecasting due to heightened demand patterns, it presents an opportunity to leverage modern technology for improved outcomes.