Definition
Measures the rate at which actual MW response per DR event declines as cumulative event count increases within a program year, isolating the fatigue effect from weather and other confounders.
The Measurement Gap
Traditional DR program tracking reports total enrolled MW, seasonal event count, and average response ratio. None of these metrics reveal whether the program's effectiveness is declining within a season as a function of dispatch frequency. A program that dispatches 12 times per summer and sees its response ratio fall from 92% in June to 61% in August is still reported as delivering roughly 75% average, which obscures the fact that the last four events produced deeply suboptimal results.
The specific shortfall of legacy tracking is the absence of a sequence variable. Legacy metrics don't connect event number 10 to lower response than event number 2. DREFDR adds that variable, computing the slope of response degradation as a function of cumulative event count, controlling for weather and day-of-week effects so the fatigue signal is isolated from confounders.
The operational decision this KPI enables is the optimal dispatch frequency policy: how many events per season maximizes total MW-hours delivered before fatigue erodes marginal event value below its cost? Without DREFDR, program managers set dispatch frequency based on historical norms or grid need alone: neither accounts for fatigue dynamics.
What goes wrong without it: the AI dispatches event 12, gets 55% of predicted MW, the reliability margin it was called to cover is not met, and the control room is left scrambling. The fatigue decay was predictable from the trend. AI and AMI together make this computable now: every event produces a response ratio data point that can be regressed against cumulative event count within the season, with weather covariates controlled by standard regression methods.
Formula
DREFDR = β₁ in the regression: Response_ratio_i = β₀ + β₁ × CumulativeEvents_i + ε_i
Variables:
- Response_ratio_i: Actual_MW_curtailed_i / Predicted_MW_curtailed_i for event i (dimensionless, 0–1+). Source: AMI M&V reads and DR platform dispatch log
- CumulativeEvents_i: Number of DR events dispatched in the program year prior to event i (count). Source: DR management platform
- β₁: The fatigue decay slope (negative values indicate fatigue is occurring; magnitude indicates severity (percentage points of response lost per additional event dispatched)) Source: Derived via OLS regression
- Covariates: Temperature at dispatch time (°F or °C), time of day, day of week, event duration (hours). Source: Weather data service, DR platform
Calculation Frequency: Computed after each program season (minimum 8 events); can be updated mid-season after 5+ events to provide early warning.
Output:
β₁ slope coefficient. Values between 0 and −0.02 per event represent normal degradation. Values more negative than −0.05 per event indicate each additional dispatch erodes response by 5+ percentage points, a program approaching its effective dispatch limit.
Proposed formula, pending validation pilots.
At a Glance
Decision Supported
How many DR events can we call before response effectiveness degrades below the threshold where the event is worth dispatching?