This post summarizes our recent open-access publication in Energies.Read the full article here.
Why Focus on Long-Term Load Forecasting?
Long-term load forecasting (LTLF) is a crucial tool for planning and operating electric power systems. Accurate forecasts help system operators and planners make informed decisions about investments, operations, and policy. However, LTLF faces major challenges due to the increasing uncertainty and variability in modern power systems, particularly with the integration of new technologies and variable resources.
Traditional long-term forecasting approaches rely on data at a single time scale—hourly, daily, or annual. But as anyone involved in Alberta’s grid knows, electricity demand is shaped by everything from sharp winter cold snaps to gradual economic shifts, and these play out across many time scales at once.
The Multi-Resolution Solution: Seeing the Big Picture and the Details
In our recent research, we developed a multi-resolution probabilistic forecasting framework using what’s called temporal hierarchies. In plain terms, this means we combine data and forecasts from different time scales—hourly, daily, monthly, yearly—into a unified model.
Why does this matter?
Alberta’s electricity demand changes at different speeds—some shifts are gradual, while others happen suddenly. By combining information from multiple time scales, our approach captures both the slow trends and rapid events that drive electricity use.
Importantly, our approach is probabilistic. Rather than giving a single forecast number, we provide a range of possible futures, each with an associated likelihood. This allows planners and policymakers to understand not just what’s likely, but what’s possible—including rare but critical events.
Our method starts by generating forecasts independently at each time scale—for example, forecasting annual, monthly, daily, and hourly electricity demand. Next, we use a statistical reconciliation technique to ensure these forecasts add up consistently (e.g., that 12 months’ demand matches the annual total, and so on). The result is a single, coherent set of forecasts that respects both the big-picture trends and the fine-grained variations.
We applied this method to historical electricity load data of different regions in Alberta, covering a wide range of conditions and years. By using hierarchical reconciliation, our model borrows strength across time scales, improving accuracy—especially when it comes to predicting extreme events like record peaks or valleys.
What We Found: Results from Alberta’s Grid
- Higher accuracy across the board. The multi-resolution approach consistently outperformed standard single-resolution models, particularly when forecasting peak demand—crucial for reliability planning.
- More informative risk assessment. Because the method is probabilistic, it gives decision-makers a better sense of the range and likelihood of possible outcomes, not just the average.
- Better resilience to uncertainty. By drawing on multiple time scales, the approach handled the complexity and volatility of Alberta’s grid more robustly.
Conclusion
Accurate long-term load forecasts are foundational for everything from grid investment and market design to integrating renewables and supporting the energy transition. Our research demonstrates that multi-resolution probabilistic load forecasting using temporal hierarchies can improve both the accuracy and the coherency of long-term load forecasts. This framework could be a useful addition to the toolbox of system planners and forecasters looking for reliable, consistent, and informative demand projections at various time scales.
Want to Learn More?
For full details, including methodology and case studies, check out the open-access paper in Energies.
If you’d like to connect about applying these methods to your organization or want to discuss future research, reach out at shafie.bahman@ucalgary.ca.
Let’s work together to build a brighter, smarter energy future for Alberta.
