Academy/Baseline & Event Performance/Demand Response Baselines: Methods and Tradeoffs
Baseline Engineering

Demand Response Baselines: Methods and Tradeoffs

How to choose between high-of-Y, middle-of-Y, day-matching, median, and EWMA baseline approaches depending on program goals and data conditions.

Intermediate8 minBaseline & Event Performance

High-of-Y and middle-of-Y methods

High-of-Y methods remain common because they are intuitive and align well with programs that want event-day comparisons against strong historical usage. In GridMango, this includes configurations like PJM High 4-of-5 and High 5-of-10, where the selected top candidates intentionally reflect heavier event-window load.

Middle-of-Y configurations are useful when the goal is to reduce sensitivity to the most extreme candidate days while still keeping the day-selection approach familiar. That makes them a practical bridge between peak-driven methodologies and more smoothing-oriented alternatives.

Day-matching, median, and EWMA

Day-matching averages are straightforward when consistency and interpretability matter more than emphasizing peak behavior. Median candidates help when outlier suppression is important, particularly with noisy telemetry or mixed occupancy patterns. EWMA is helpful when recent behavior should dominate, but it also requires teams to be comfortable explaining recency weighting to reviewers.

No single method is universally correct. The best method is the one that matches the program’s review expectations and the stability of the underlying telemetry.

Choosing method and candidate depth together

The baseline method and the candidate pool depth should be treated as one design decision. Asking for a different X, Y, or candidate style changes not only the resulting baseline but also what you are telling reviewers about representativeness. GridMango’s baseline workflow is strongest when these choices are explained up front and validated against multiple event scenarios.

Key takeaway: Baseline selection is not just a formula choice. It is a policy decision about what historical behavior should count as representative under review.