Every HVAC optimisation project promises savings, but savings can only be proven against a credible baseline. Without 12 months of clean energy, weather and operating data, a retuned chiller plant or a new schedule produces a number nobody can defend to finance or auditors. This guide covers what to collect, how to clean it, how to judge a baseline model and how maintenance records explain the changes hiding in the data. It also shows how work orders and asset history in a CMMS support defensible measurement and verification.
HVAC Energy Baseline: 12-Month Data Before You Optimize
Measure the system as it runs today, across a full year of weather and occupancy, so every future saving has a reference point that stands up to scrutiny.
Why the baseline comes before the optimisation
Energy savings cannot be metered directly, because the energy you did not use never passes through a meter. They are calculated by comparing what the building used after a change with what it would have used under the old conditions. That second number is the adjusted baseline.
Why twelve months
- HVAC load swings with season, so a partial year hides entire operating modes
- A full cycle captures peak cooling, peak heating and shoulder-season behaviour
- Holiday shutdowns, events and occupancy patterns appear at least once
- ASHRAE Guideline 14 expects a baseline period of at least 12 months for whole-building models
The data you need, and where it comes from
| Data stream | Preferred interval | Typical source | Why it matters |
|---|---|---|---|
| Whole-building electricity | 15 minutes or hourly | Utility interval data or main meter | Anchors whole-facility models |
| HVAC sub-metered electricity | 15 minutes or hourly | Chiller, pump and AHU meters | Isolates HVAC from other loads |
| Gas or district heating | Daily or hourly | Utility or building meter | Captures heating energy |
| Outdoor temperature and humidity | Hourly | On-site sensor or nearby weather station | Main driver of cooling and heating load |
| Occupancy or operating hours | Daily | Badge data, schedules, shift calendars | Explains load differences between days |
| Setpoints and schedules | On change | Building automation system | Shows control intent against actual behaviour |
| Equipment runtime and status | Trend logs | Building automation system | Links energy to what was running |
| Maintenance and repair events | Event based | Work orders and asset history | Explains step changes in energy use |
A 12-month collection plan
Treat the baseline year as a project with its own schedule. Setting up instrumentation early prevents the common problem of discovering missing months at the end.
- Define the measurement boundary
- Check meter accuracy and units
- Connect weather and trend sources
- Set naming rules for points
- Review data completeness monthly
- Fill gaps through repair or estimation rules
- Log all schedule and setpoint changes
- Draft a preliminary model
- Capture the main cooling season
- Test the model against new months
- Record occupancy changes
- Document non-routine events
- Close remaining data gaps
- Finalise and validate the model
- Agree acceptance criteria
- Sign off the baseline report
Give your baseline the maintenance context it needs
Record repairs, tuning and equipment changes against each asset, so every shift in energy use has an explanation on file.
Cleaning the data before it touches a model
Raw interval data is rarely ready to use. Meters reset, communications drop, clocks shift and units get mixed. A short, documented cleaning pipeline keeps these problems from quietly distorting the model.
Weather normalisation: degree days and regression
Outdoor conditions drive most of the variation in HVAC energy. Two buildings with identical equipment can show very different bills simply because one year was hotter. Normalising for weather removes that noise so efficiency changes become visible.
Common model forms
- Simple linear regression of energy against heating or cooling degree days
- Change-point models that switch behaviour above and below a balance temperature
- Multivariable models adding humidity, occupancy, daylight or production
- Time-of-week and temperature models for hourly interval data
Practical tips
- Choose the balance temperature from the data rather than assuming a default
- Use a weather source close to the site and keep the same one for the reporting period
- Check residuals for patterns, since a pattern means a driver is missing
- Keep the model as simple as the data allows, because complexity can hide errors
Instrumentation gaps and how to close them
Many sites discover during the baseline year that the meters they assumed existed do not, or that existing meters are mislabelled. Resolve these early, because a missing sub-meter cannot be recovered retroactively.
Documenting operating conditions
A model only explains the variables it is given. Written records of how the building was meant to run turn unexplained variance into understood behaviour.
| Condition | Record to keep | Effect on the baseline |
|---|---|---|
| Occupancy hours | Scheduled and actual hours, holidays and events | Shifts ventilation and cooling load |
| Setpoints | Cooling, heating and humidity targets with change dates | Changes energy for the same weather |
| Ventilation strategy | Minimum outdoor air, demand control and economizer settings | Alters conditioning load |
| Floor area use | Tenant moves, vacant floors, new equipment loads | Changes internal gains |
| Equipment staging | Lead and lag order, rotation and lockouts | Affects plant efficiency |
What the baseline report should contain
The baseline report is the document that finance, auditors and project partners will return to long after the project ends. It should make every assumption visible.
- Project scope, measurement boundary and chosen IPMVP option
- Baseline period dates and the reasons for choosing them
- Data sources, meter identifiers, intervals and calibration notes
- Cleaning rules, gap treatments and excluded periods
- Model form, variables and fitted coefficients
- Statistical results, including CV(RMSE), NMBE and any hold-out test
- Non-routine events and the adjustments applied
- Reporting period conditions and how savings will be calculated
Trends shaping HVAC baselining
- Interval data from smart meters is now widely available, making hourly and sub-hourly models practical
- Building automation trend data and analytics platforms allow continuous comparison against the baseline
- Fault detection and diagnostics tools use baseline behaviour to flag drift between audits
- Energy performance reporting and decarbonisation targets increase scrutiny of claimed savings
- Maintenance and energy teams are being asked to work from shared data rather than separate spreadsheets
Why this matters to maintenance teams
Maintenance actions are often the cheapest energy measures available. Proving their value requires a baseline, and protecting that value requires the preventive routines that keep performance from drifting back.
Choosing an IPMVP approach
The International Performance Measurement and Verification Protocol defines four options. The right one depends on the size of the project, the interaction with other loads and the available meters.
| Option | What is measured | Good fit for | Main caution |
|---|---|---|---|
| A: Retrofit isolation, key parameter | One key parameter measured, others estimated | Pump or fan motor upgrades | Estimates need a clear justification |
| B: Retrofit isolation, all parameters | All energy parameters of the isolated system | Chiller plant or AHU retrofits with sub-meters | Requires reliable sub-metering |
| C: Whole facility | Utility meter for the entire building | Multi-measure programmes with large expected savings | Savings can be hidden by other changes |
| D: Calibrated simulation | Simulation model tuned to measured data | New construction or missing baseline data | Calibration effort and expertise |
How good is good enough: ASHRAE Guideline 14 criteria
Guideline 14 sets statistical limits for whole-building baseline models so that a model is not accepted on appearance alone. Two measures are used, the coefficient of variation of the root mean square error and the normalised mean bias error.
Reading the numbers
- CV(RMSE) describes scatter: how far individual predictions sit from actual values
- NMBE describes bias: whether the model systematically over- or under-predicts
- Both should be met, and passing one does not excuse failing the other
- Check the current edition of the guideline and your project specification for exact requirements
Common baseline mistakes: before and after
Maintenance events: the hidden variable in every baseline
A baseline is meant to describe normal operation, but real buildings are repaired throughout the year. A failed economizer damper, a stuck valve or a refrigerant leak changes energy use, and the model will treat it as normal unless someone records it.
Events worth logging against the asset
- Component failures that forced equipment into fallback or manual operation
- Repairs and parts replacements that changed capacity or efficiency
- Control changes such as setpoint resets, schedule edits and sequence updates
- Filter changes, coil cleaning and calibration of sensors and valves
- Temporary conditions such as portable cooling, construction or tenant moves
Fix first, or measure first?
Safety, comfort and compliance faults should be repaired promptly. The key is to document the fix as a non-routine adjustment, with date and reasoning, so the baseline can be adjusted rather than silently contaminated.
How Oxmaint supports the baseline year
Oxmaint is a maintenance management platform, not a replacement for your metering or analytics tools. Its role is to hold the maintenance context that explains your data and keep upkeep consistent during measurement.
From baseline to fault detection
A strong baseline does more than prove savings. It reveals how equipment should behave, which lets analytics flag deviations that point to real maintenance work.
| Baseline signal | Deviation to watch | Typical maintenance response |
|---|---|---|
| Cooling energy at given outdoor temperature | Rising consumption for the same conditions | Inspect coils, refrigerant charge and condenser performance |
| Fan energy against airflow demand | Fans running at high speed with little demand | Check filters, dampers and static pressure control |
| Night and weekend load | Higher than the scheduled shutdown profile | Review schedules and overrides, then correct controls |
| Simultaneous heating and cooling | Both active in the same zone or system | Inspect valves, actuators and sensor calibration |
| Economizer operation in mild weather | Little or no free cooling when available | Test dampers, linkages and enthalpy or temperature sensors |
Checklist before optimisation begins
- Measurement boundary and options agreed in writing
- Twelve consecutive months of energy data collected
- Weather data aligned to the same intervals
- Occupancy and operating schedules documented
- Meter accuracy and units verified
- Data gaps and fills documented with method
- Non-routine events logged with dates
- Model tested against statistical criteria
- Raw data and cleaned data archived
- Baseline report approved by all stakeholders
Frequently asked questions
Why must an HVAC baseline cover 12 months?
A full year captures every season and occupancy pattern. Shorter periods miss operating modes and weaken the model.
What if I do not have 12 months of data yet?
Start collecting now and use any historical utility data to bridge gaps. Sign up to log events from day one.
Do maintenance activities change the baseline?
Yes. Repairs, tuning and control changes shift energy use, so they should be logged as non-routine events and adjusted for.
Which IPMVP option should I use?
Option B suits sub-metered retrofits, and Option C suits whole-building programmes. Confirm your choice in the M&V plan.
Can a CMMS calculate energy savings?
It does not replace M&V analytics, but it supplies the event history. Book a demo to see how it fits your workflow.
Make your next HVAC saving impossible to dispute
Pair clean baseline data with a complete maintenance record and give finance a number they can trust.
