HVAC Energy and Cooling Degree Days: Separate Weather from Performance

Two monthly energy totals cannot tell the whole HVAC performance story. A system may use more energy because the weather was hotter, operating hours were longer or more floor area was occupied. Weather adjustment creates a more useful comparison by asking what the earlier operating pattern would have consumed under the later weather. It does not automatically identify the cause of the remaining difference.

1. Define the meter boundary and comparison question

Start by stating whether the dataset represents the whole building, HVAC input energy, a chiller plant or purchased cooling. Record the units and included equipment. A whole-building meter includes loads that may have little relationship with cooling weather. A plant meter may exclude distribution pumps. Changing that boundary between periods undermines the comparison before any regression begins.

There are two related questions. A reporting-period adjustment predicts baseline energy under the weather actually experienced during the later period. Normalization to reference weather evaluates different periods under a common weather dataset. The EPA climate and weather technical reference describes weather-normalized energy as a within-building comparison under average conditions. The simple example here is an independent teaching model, not a reproduction of the Portfolio Manager calculation.

2. Build degree days on a declared temperature basis

For this illustrative daily method, cooling degree days equal the sum of max(0, daily mean outdoor temperature minus the selected base temperature). If the assumed base is 18°C and three daily means are 22°C, 27°C and 16°C, their contributions are 4, 9 and 0°C-days. The total is 13°C-days. A day below the base contributes zero rather than cancelling a hotter day.

The 18°C base is an example, not a universal building balance point. Internal gains, schedules and envelope characteristics influence when cooling becomes necessary. Also identify how the daily mean is obtained. Hourly averaging and averaging daily maximum and minimum temperatures need not produce identical results. Use one documented method consistently and retain its source data.

Do not mix Celsius-based degree days with Fahrenheit-based coefficients. A coefficient fitted against one temperature scale cannot be carried unchanged into the other. Identify the weather station, period boundaries, time zone, missing-day treatment and any station changes alongside the final total. A nearby station should also be representative of the building's local conditions.

3. Keep the base load separate from weather sensitivity

Assume a previously fitted cooling-season model is E = 200 × D + 40 × CDD, where E is input energy in kWh, D is the number of days and CDD is cooling degree days in °C-days. The invented coefficients are 200 kWh/day and 40 kWh per °C-day. They demonstrate the arithmetic only; they were not estimated from a real building.

The first term captures modeled energy that persists independently of cooling degree days over this dataset. It is not necessarily a measured fan load or proof that a particular component consumes 200 kWh daily. The second term describes the fitted weather-sensitive contribution. Physically labeling either term requires additional evidence.

For a 30-day baseline month with 250°C-days, predicted consumption is 200 × 30 + 40 × 250 = 16,000 kWh. Assume its measured consumption is also 16,000 kWh. That exact agreement is deliberately chosen to make the example easy to reproduce, not to imply that real monthly models fit perfectly.

4. Adjust the baseline to the reporting weather

Now consider a second 30-day month with 300°C-days and measured consumption of 17,000 kWh. The unadjusted total increased by 1,000 kWh, or 6.25%. However, the baseline model predicts 200 × 30 + 40 × 300 = 18,000 kWh under this hotter weather.

Original illustrative model: 200 kWh/day plus 40 kWh per Celsius cooling degree day. Equal 30-day periods; coefficients are assumed.

The reporting-period difference is 18,000 − 17,000 = 1,000 kWh below the adjusted baseline. Relative to that adjusted baseline, the difference is 1,000/18,000 = 5.56%. The weather contribution to the predicted change is 40 × (300 − 250) = 2,000 kWh. A raw increase and an adjusted reduction can therefore coexist without contradiction.

If the sole adjustment were this model's weather term, the second month's energy translated to 250°C-days would be 17,000 − 40 × (300 − 250) = 15,000 kWh. This is another way of expressing the same 1,000 kWh difference against the original 16,000 kWh baseline. Do not apply both adjustments to the same comparison; they are alternative reference frames.

5. Match time periods before trusting the result

A bill spanning 34 days should not be compared with 30 calendar days of weather. In the example model, four additional days add 800 kWh to the base term before weather is considered. Use actual meter-read start and end dates, document whether endpoints are inclusive and calculate weather for precisely that interval.

For interval-meter data, aggregate meter and weather records onto the same time boundary. A missing meter day is not a zero-consumption day. Estimated bills, meter replacements and data corrections need explicit flags. Keep the original observations so a later reviewer can reproduce every aggregate rather than receiving only a polished monthly chart.

6. Test the model before calling the difference savings

The DOE measurement and verification supplement discusses custom degree-day bases, model validation and aligning weather with billing periods. For a practical review, plot measured versus predicted energy and residuals over time. Check whether errors cluster around weekends, humid periods or unusual occupancy. Inspect performance on withheld observations when sufficient data exist.

• Include a representative range of cooling weather and operating conditions in the baseline, rather than fitting two convenient months.

• Record occupancy, operating-hour and floor-area changes separately. A new tenant or extended shift can invalidate the earlier relationship.

• Check humidity and other influential variables where dry-bulb degree days leave systematic unexplained variation.

• Avoid extrapolating far beyond the baseline weather range. State the model's valid operating range and uncertainty.

An illustrative residual of 1,000 kWh is not automatically verified savings. If normal prediction uncertainty is of similar magnitude, the apparent improvement may be inconclusive. Even a statistically meaningful reduction needs investigation before attributing it to a specific maintenance action. Use the adjusted comparison to focus that investigation, preserve the evidence and report what the data can genuinely support.