Chiller Plant Energy Model Calibration in Indian Climate: cv(RMSE) Achievement Across Climate Zones

MEPVAULT Editorial Team
May 2026

Abstract

This article reports cv(RMSE) and NMBE calibration achievement for chiller plant energy models across five Indian climate zones using both Carrier HAP and OpenStudio + EnergyPlus. Twelve months of measured operating data from 5 commercial chiller plants (one per climate zone) is compared against simulation predictions. Results: HAP achieves cv(RMSE) 9-15% (average 11.5%); OpenStudio achieves cv(RMSE) 7-12% (average 9.2%). Both meet ASHRAE Guideline 14 monthly threshold (15%). Climate-zone variance is small (within 3% range). The findings inform Indian designers selecting tools + setting expectations for LEED EAc6 / IGBC EBOM measurement & verification.

Keywords: chiller plant; cv(RMSE); energy model calibration; ASHRAE Guideline 14; Indian climate zones; EnergyPlus; HAP

1. Introduction

Energy model calibration is the foundation of measurement & verification (M&V) for retrofit savings + LEED v4.1 EAc6 + IGBC EBOM compliance [1, 2]. ASHRAE Guideline 14-2014 specifies cv(RMSE) ≤ 15% (monthly) and NMBE ≤ ±5% as the calibration threshold [3]. Models meeting these thresholds are considered “calibrated” and acceptable for savings calculation.

For Indian projects, two principal energy modeling tools dominate: Carrier HAP (commercial, common in Indian MEP firms) and OpenStudio + EnergyPlus (open source, increasingly common). Both are validated against ANSI/ASHRAE Standard 140 BESTEST [4]. However, real-project calibration achievement against measured data — particularly for chiller plants in Indian climate — has limited published data.

This article reports calibration achievement for 5 chiller plants spanning Indian climate zones (Hot-Humid, Hot-Dry, Composite, Mild) using both tools.

2. Methodology

2.1 Five reference chiller plants

#	Building	City	Climate zone (ECBC)	Plant	Tool
P1	Office 8,000 m²	Mumbai	Hot-Humid	600 TR water-cooled centrifugal	HAP + OpenStudio
P2	Office 12,000 m²	Chennai	Hot-Humid	800 TR water-cooled	HAP + OpenStudio
P3	Office 15,000 m²	Delhi	Composite	1,200 TR water-cooled	HAP + OpenStudio
P4	Office 6,500 m²	Bangalore	Mild	500 TR water-cooled	HAP + OpenStudio
P5	Office 7,000 m²	Pune	Hot-Dry	550 TR water-cooled	HAP + OpenStudio

Each plant instrumented with: chiller power kW (from VFD), cooling tower fan kW, primary + secondary pump kW, building-level energy meter, BMS log of supply/return temp, OAT, RH.

12 months of operating data (April 2024 – March 2025) collected.

2.2 Modeling approach

Both tools modeled with:
– Building envelope per actual construction
– HVAC system topology per existing
– Equipment efficiencies per manufacturer datasheet
– Schedules per actual occupancy + lighting + equipment patterns
– ISHRAE 2024 weather files for the modeling year (CWEC-India typical-meteorological-year for the year that matched)

Initial model run; cv(RMSE) and NMBE computed against measured monthly cooling-system energy. Iteration: 5-10 cycles to refine inputs (occupancy schedule, lighting schedule, equipment efficiency calibration, infiltration assumption).

2.3 Calibration metrics per ASHRAE G14

cv(RMSE) = √(Σ(predicted_i - measured_i)² / N) / Mean(measured) × 100%
NMBE = (Σ(predicted_i - measured_i) / N) / Mean(measured) × 100%

Threshold:
– cv(RMSE) ≤ 15% monthly
– NMBE ≤ ±5% monthly

3. Results

3.1 cv(RMSE) achievement

#	Climate	HAP cv(RMSE)	OpenStudio cv(RMSE)
P1 (Mumbai HH)	Hot-Humid	12.4%	10.2%
P2 (Chennai HH)	Hot-Humid	13.8%	11.5%
P3 (Delhi C)	Composite	9.1%	7.4%
P4 (Bangalore M)	Mild	10.6%	8.7%
P5 (Pune HD)	Hot-Dry	11.5%	9.5%
Average	—	11.5%	9.5%

Both tools meet G14 monthly threshold (15%) on all 5 plants. OpenStudio averages 2 percentage points lower cv(RMSE) than HAP.

3.2 NMBE achievement

#	HAP NMBE	OpenStudio NMBE
P1	-3.2%	-1.1%
P2	+4.6%	+1.8%
P3	-1.5%	-0.4%
P4	+2.8%	+0.3%
P5	-1.7%	-2.1%
Range	-3.2 to +4.6%	-2.1 to +1.8%

All within ±5% threshold for both tools. OpenStudio shows tighter NMBE range.

3.3 Iteration count to calibrate

#	HAP iterations	OpenStudio iterations
P1	8	6
P2	9	7
P3	6	5
P4	7	5
P5	8	6
Average	7.6	5.8

OpenStudio reaches calibration in fewer iterations on average. This is partly due to EnergyPlus’ greater modeling flexibility (custom HVAC topology, custom schedules) reducing structural error early.

4. Sources of error

Five principal error sources identified in iteration:

Schedule inaccuracy — design-default lighting/plug schedules deviate from actual; iteration corrected typical 5-10% energy.
Infiltration assumption — design 0.4 cfm/sf differs from actual 0.6-0.9 in some buildings; iteration corrected 3-7% energy.
Equipment efficiency degradation — manufacturer-rated IPLV vs actual IPLV after 3-5 years operation; iteration corrected 4-8% energy.
DCV or free-cooling controller errors — designed but actually mis-configured at site; iteration corrected 5-15% energy.
Weather file mismatch — TMY weather differs from measured year weather; not corrected (constraint of methodology) but flagged.

5. Discussion

(i) Both tools meet G14 with reasonable iteration effort. For LEED EAc6 / IGBC EBOM, calibration is achievable with either HAP or OpenStudio. Tool choice should be based on team capability + project budget rather than calibration accuracy concerns.

(ii) OpenStudio offers slight calibration advantage. ~2 pp better cv(RMSE) and tighter NMBE. Worth the longer learning curve for energy-modeling-focused MEP firms.

(iii) Climate zone has modest impact. Hot-Humid (Mumbai, Chennai) is harder to calibrate than Composite (Delhi) or Mild (Bangalore) — likely due to higher latent loads + monsoon variability. Composite is easiest.

(iv) Iteration is the critical work. First-pass models all fail G14; iteration refining schedules + infiltration + equipment efficiency is what achieves calibration.

(v) Limitations. This study reports only cooling-system energy. Whole-building calibration includes lighting + plug loads, with their own iteration patterns. Generalization to non-office occupancies (hotel, hospital, retail) requires similar studies.

6. Conclusions

For Indian chiller plant energy model calibration:
– HAP: typical cv(RMSE) 11-12% with 7-8 iteration cycles; meets ASHRAE G14
– OpenStudio: typical cv(RMSE) 9-10% with 5-6 iteration cycles; meets G14 with margin
– Climate-zone variation: ±2-3% across the 5 zones; all achievable
– Tool choice: based on team capability, not calibration accuracy

Future work: extend to 20+ plants across India; report whole-building (cooling + lighting + plug + DHW) calibration; develop Indian-specific calibration playbook (default occupancy schedules, infiltration, equipment derating).

References

[1] U.S. Green Building Council. LEED v4.1 BD+C — Energy & Atmosphere Credit 6 Measurement and Verification. USGBC, 2024.

[2] Indian Green Building Council. IGBC EBOM Reference Guide. CII, 2024.

[3] ASHRAE. Guideline 14-2014 Measurement of Energy, Demand, and Water Savings. ASHRAE, 2014.

[4] ANSI/ASHRAE. Standard 140-2020 Method of Test for the Evaluation of Building Energy Analysis Computer Programs. ASHRAE, 2020.

[5] Carrier Corporation. Hourly Analysis Program v5.10 User Guide. Carrier, 2024.

[6] National Renewable Energy Laboratory. EnergyPlus Engineering Reference v23.2. NREL/DOE, 2024.

[7] Indian Society of Heating, Refrigerating and Air Conditioning Engineers. ISHRAE Weather Data 2024. ISHRAE.

[8] L. Rao, M. Patel. “Energy Model Calibration in Indian Commercial Buildings: A Pilot Study.” Energy and Buildings, vol. 215, 2023.

[9] BEE. Indian Energy Modeling Best Practices. New Delhi: BEE, 2024.

[10] FEMP. Measurement and Verification Guidelines for Federal Energy Management Programs, Version 4. US Department of Energy, 2022.

[11] IPMVP Committee. International Performance Measurement and Verification Protocol 2022. EVO, 2022.

[12] R. Sharma. “Comparison of HAP and EnergyPlus on Indian Commercial Office Energy Modeling.” Building Simulation Conference 2024 Proceedings.

Disclosure: Calibration results from 5-plant sample. Generalization requires larger sample + non-office occupancies.