IPMVP change-point & ASHRAE G14 (pandas oracle)

PyPI helpers for Option C–style weather-normalized baselines and Guideline 14 calibration scores. Live in open_fdd.ecm_engineering (also re-exported from open_fdd.analytics). Not product DataFusion FDD — notebooks, twin honesty, and Wave S4 SQL twin characterization only.

Install: pip install "open-fdd[oracle]" (needs numpy for change-point fits).

External algorithm reference: Camber (Apache-2.0). Open-FDD ports are clean-room IMT/IPMVP shapes — never vendor Camber into GHCR central/web or treat camber serve as product UI.

G14 monthly / hourly gates

Granularity	\|NMBE\| max	CVRMSE max
Monthly (calibrated simulation)	5%	15%
Hourly	10%	30%

from open_fdd.ecm_engineering import score_g14_monthly, score_g14_fuels

measured = [1200, 1100, 900, 850, 700, 650, 800, 950, 1000, 1050, 1150, 1250]
modeled = [1180, 1120, 910, 840, 710, 640, 790, 960, 1010, 1040, 1160, 1240]

s = score_g14_monthly(measured, modeled, n_params=4, fuel="elec")
assert s.pass_  # |NMBE| ≤ 5% and CVRMSE ≤ 15%

both = score_g14_fuels(
    elec_measured=measured,
    elec_predicted=modeled,
    gas_measured=measured,
    gas_predicted=modeled,
)
assert both["pass"]

Formulas (ASHRAE Guideline 14):

\[\mathrm{NMBE} = 100 \cdot \frac{\sum (m_i - \hat{m}_i)}{(n-p)\,\bar{m}}\] \[\mathrm{CV(RMSE)} = 100 \cdot \frac{\sqrt{\sum (m_i - \hat{m}_i)^2 / (n-p)}}{\bar{m}}\]

where (m_i) is measured, (\hat{m}_i) predicted, (n) samples, (p) model parameters, and (\bar{m}) the mean of measured.

Change-point models (2P / 3P / 4P / 5P)

Independent variable is usually outdoor dry-bulb (°F); dependent is period energy (kWh, therms, …).

Kind	Shape
`2P`	linear: Y = a + b·X
`3PH`	heating: Y = a + b·(β − X)₊
`3PC`	cooling: Y = a + b·(X − β)₊
`4P`	dual slope, one change-point β
`5P`	dual slope, heating + cooling change-points

import numpy as np
from open_fdd.ecm_engineering import fit_changepoint, select_changepoint, option_c_savings

oat_f = np.linspace(30, 90, 48)
kwh = 200 + 8 * np.maximum(oat_f - 55, 0)

model = fit_changepoint(oat_f, kwh, kind="3PC")
# or: model = select_changepoint(oat_f, kwh)

# Reporting period: same weather, lower use → positive savings
savings = option_c_savings(model, reporting_x=oat_f, reporting_y=kwh * 0.9)
print(savings["savings_total"], savings["baseline_kind"])

Same entry points via open_fdd.analytics for oracle notebooks that already import analytics.

Dual catalog — Camber ↔ Open-FDD role aliases (doc only)

Until Wave S4 SQL M&V twins land, this table is documentation only — not a runtime adapter.

Concept	Camber-ish / IMT	Open-FDD package / SQL role
Outdoor temperature (X)	`temp`, OAT	`outside-air-temp`, `oa_t`, weather `dry_bulb_f` / `web-outside-air-temp`
Energy (Y) (electric)	`energy`, kWh	`elec-power` → monthly kWh, `kwh`, UTIL monthly bills
Energy (Y) (gas)	gas / therms	`gas-flow` / `gas-rate`, UTIL gas bills
Period grain	daily / monthly	package utilities monthly; historian resample in oracle
Fit quality	CVRMSE / NMBE	`score_g14*` / model `.cvrmse_pct`
Product charts	Camber UI / serve	Metering radios + future DataFusion `/api/analytics/*` (S4)

Boundary

Layer	Owns
This module (PyPI)	Clean-room math + twin honesty helpers
Product GHCR	DataFusion SQL FDD / analytics — no pandas in request path
Soft-OPEN	Full Camber family ports, PyPI publish tip, SQL twins (Wave S4)