Weather Normalized Metered Demand Response

Foreword

This specification has been prepared to standardize how demand response (DR) savings and associated load reduction impacts are measured using whole-building meter data when a short-term counterfactual is required. For devices with direct metering and explicit dispatch signals (e.g., batteries), a device-level methodology SHALL be used instead.

About the OpenEAC Alliance

The OpenEAC Alliance develops open, auditable methodologies for distributed energy resources (DERs) to enable trustworthy Energy Attribute Certificates (EACs). This document aligns with the Alliance’s emphasis on clarity, reproducibility, and market interoperability.

Normative References

• OpenDSM, CalTRACK 2.0 (EEMeter) — daily/interval M&V framework

  • CAISO Study

• LBNL Time-of-Week and Temperature (TOWT) model

• ASHRAE Guideline 14 — Measurement of Energy and Demand Savings

Introduction

This document specifies a statistically rigorous method to quantify DR energy savings and hourly emissions impacts at the building level. The approach draws on OpenDSM, CalTRACK 2.0 methods, the Time-of-Week and Temperature (TOWT) model developed by LBNL, and guidance consistent with ASHRAE Guideline 14. Only the following inputs are required: (a) interval electric meter data at hourly (or finer) resolution and (b) hourly outdoor temperature. All other building conditions (e.g., occupancy patterns) are inferred by the method.

Terms and Definitions

• Event Window: The set of hours where DR has been called, as defined by the program or operator.

• Baseline: A period leading up to the event window to be used to fit a model for prediction.

• Counterfactual: The predicted electricity use had no DR occurred.

Scope

This methodology supports event-based demand response delivered as a reduction in metered electricity use relative to a short-term counterfactual. The following are IN SCOPE:

• Temporary load reductions (e.g., curtailing lighting/plug loads).

The following are OUT OF SCOPE for this method:

• Permanent load reductions (energy efficiency baselines).

• Load shifting around an event (e.g., pre-cooling / post-event recovery).

• Ongoing daily habitual shifts absent event triggers.

• Frequency regulation or services requiring sub-minute telemetry.

• Device-metered dispatch resources (e.g., batteries).

Data Requirement Definitions

• Electricity Interval Data: Hourly (or sub-hourly aggregated to hourly) whole-building Wh.

  • Baseline Data Sufficiency: ≥90% hourly coverage of the event window day for all days in baseline period.

  • Event Day Coverage: 100% hourly coverage across the event window.

    • Hours that are missing observed data in the event window will not have measurable savings.

• Weather Data: Hourly dry-bulb temperature representative of the location.

• Solar Irradiation Data (Optional): Hourly irradiation representative of the location.

• Event Metadata: Start/stop times and any program identifiers for each event window.

Methodology

This process is to be done once per DR event window.

1 Construct a baseline period

1.1 The counterfactual SHALL be trained on a rolling baseline ending prior to the Event day (excluding the Event and its Shoulders).

1.1.1 This baseline period SHOULD be 28 days.If material schedule/season changes are present, the baseline MAY be shortened/extended (14–56 days) with justification, but SHOULD avoid spanning distinct seasonal regimes.

1.1.2 Event days MAY be excluded, but the baseline period should include at least 4 instances of the event’s day-of-the-week, and must include at least 2 instances.

1.1.3 Holidays and known atypical schedules MAY be excluded from training if they degrade model stability.

2 Fit a baseline model (Time-Of-Week and Temperature)

2.1 Fit an energy prediction model that incorporates

2.1.1 Temporal patterns (hour-of-day, day-of-week, monthly)

2.1.2 Temperature via segmented splines or bins (cooling and, if present, heating sensitivities).

2.1.3 Solar Irradiation (for locations with behind-the-meter PV).

2.2 Compute CVRMSE of the baseline model evaluated on the baseline data.

3 Calculate savings within the event window

3.1 Use the model to predict the hourly counterfactual consumption for all hours in the event window.

3.2 For each hour in the event period, subtract the observed Wh from the counterfactual Wh. The sum of these savings over the course of the event window represent the impact of this DR event.

Model Disqualification

• Single-site eligibility: CVRMSE ≤ 1.0. Sites exceeding the threshold SHALL be excluded from Event savings claims.

• Portfolio target: Aggregated portfolio SHOULD have a CVRMSE ≤ 0.25.