Thermostat DR Behavior Model
Thermostat DR Behavior Model
Module:
insight(always on) | Role: Any member | Licence: Free+
Thermostat DR treats the building's thermal mass as a battery. The output is a delta profile (change vs normal HVAC operation), not an absolute load.
Source: EPRI demand response studies, ev.energy engineering calibration against utility DR event data, Nest/Google and Ecobee published thermostat DR performance data.
Three-Phase Model
Each DR event runs three consecutive 1-hour phases (4 intervals each):
| Phase | Power Delta (default) | Duration Range | Timing | Purpose |
|---|---|---|---|---|
| Pre-charge | +0.7 kW/home | 0.5-2 hours | Before dr_start | Over-condition building (pre-heat or pre-cool) |
| Reduction | -1.0 kW/home | 1-4 hours | dr_start onwards | Curtail HVAC; thermal mass maintains comfort |
| Rebound | +0.5 kW/home | 0.5-3 hours | After reduction | HVAC restores setpoint (snapback) |
Energy roughly balances: positive energy (pre-charge + rebound) ≈ 1.2x negative energy. ~80% thermal round-trip efficiency.
Pre-conditioning detail: Setpoint offset of -2°F / -1°C (range: 1-4°F), increasing load 20-40% above baseline depending on HVAC capacity.
Curtailment detail: Setpoint offset of +4°F / +2°C (range: 2-6°F), reducing load 50-80% below baseline (0.5-1.5 kW per household).
Recovery detail: Snapback ratio of 1.3x curtailed load (range: 1.1-1.5x), exponential decay returning to baseline over the recovery period.
HVAC Types
| Type | Reduction per Home | Best Season | Region |
|---|---|---|---|
| heat_pump | 1.0 kW | Winter | UK (default) |
| ac | 0.9 kW | Summer | US |
| resistance | 1.5 kW | Winter | UK |
Seasonal HVAC Availability
| Season | Heat Pump | AC | Resistance |
|---|---|---|---|
| Winter | 100% | 0% | 100% |
| Spring | 50% | 50% | 30% |
| Summer | 20% | 100% | 0% |
| Autumn | 70% | 30% | 60% |
Effective homes = count x prob_compliance x seasonal_availability[hvac_type][season].
Default Parameters
| Parameter | Default | Range | Notes |
|---|---|---|---|
reduction_kw_per_home | 1.0 kW | 0.5-1.5 | Based on Nest/Pecan Street data |
precharge_factor | 0.7 | 0.3-1.0 | Pre-charge as fraction of reduction |
rebound_factor | 0.5 | 0.3-0.8 | Rebound as fraction of reduction |
prob_compliance | 0.3 | 0.0-1.0 | 0.3 for recurring, 0.7 for single event |
hvac_type | heat_pump | heat_pump/ac/resistance | Determines reduction kW and seasonal availability |
season | winter | winter/spring/summer/autumn | Affects HVAC availability |
dr_start | 17:00 | HH:MM | Start of reduction phase (unmanaged mode only) |
Industry Compliance Benchmarks
| Metric | Benchmark | Source |
|---|---|---|
| Event participation rate | 85-90% | Enrolled customers who remain opted-in during event |
| Customer opt-out rate | 10-15% per event | Customers who override setpoint during event |
| Average curtailment | 0.8-1.2 kW per home | AC cycling programs in cooling season |
| Snapback peak | 1.2-1.5x curtailed load | Post-event recovery demand |
| Event fatigue factor | -2% per event per month | Declining participation with frequent events |
Source: Compiled from published utility DR program results, Ecobee/Nest/Honeywell published thermostat DR performance data, EPRI and LBNL demand response program evaluation reports.
Compliance Guidance (prob_compliance)
The fraction of enrolled homes that actually deliver a load reduction. Captures: enrollment x occupancy x HVAC availability x override rate x technical success.
| Scenario | Recommended | Rationale |
|---|---|---|
| Default (recurring daily) | 0.3 | Conservative for sustained operation |
| Single well-designed event | 0.7 | Real-world best case (Nest/Ecobee data) |
| Occasional (10-15/season) | 0.5-0.6 | Standard utility program |
| Daily for 30+ days | 0.3-0.4 | Significant occupant fatigue |
| Contractual/commercial | 0.8-0.9 | Financial penalties for non-compliance |
Key insight: 0.7 is a single-event best case. Real programs run ~12 events/season, not daily. The default 0.3 is honest about what repeating=True actually delivers.
Optimized Mode
In optimized mode, each home's 3-hour unit profile is slid across all 96 daily positions. The algorithm picks the placement that minimizes variance of (target_load + unit). Large cohorts (>10,000 homes) batch to cap at 10,000 iterations. Homes naturally stagger to flatten peaks.
Limitations
- Building thermal mass varies enormously; single model cannot capture all building types
- Customer behavior (opening windows, manual override) is not modeled
- Multi-zone HVAC systems have more complex response patterns
- Fixed phase durations (each phase is exactly 1 hour); real buildings have variable thermal time constants
- Newer smart thermostat programs tend to outperform older direct load control
- Climate and season heavily influence curtailment magnitude