Choosing a Baseline for a Heat-Recovery Measure Under IPMVP

Recommendation: Option C (whole facility) or Option B (retrofit isolation with continuous measurement), depending on metering.

Reasoning. The measure affects gas consumption at the boiler and electricity at the compressor.

• Option A (retrofit isolation, key parameters measured, others stipulated) is risky here because the recovered heat varies with compressor load; stipulating it would introduce large error.
• Option B is viable if we continuously meter the recovered heat (flow and temperature difference) and the boiler gas attributable to water heating. It isolates the measure and is robust to the production change, because savings are measured at the measure boundary.
• Option C (whole-facility) is attractive because gas and electricity interval data already exist and the meter boundary is clear. However, the January 2024 product-line change is a non-routine adjustment that must be modelled; otherwise Option C savings are confounded.

Decision. Use Option B if sub-metering of recovered heat is feasible (best accuracy, least confounding). If sub-metering is not feasible, use Option C with a baseline regression model and an explicit non-routine adjustment for the new product line. Do not use Option A with a stipulated recovery factor.

Baseline period. Use 12 months of pre-installation data (October 2022 – September 2023), avoiding the installation month as a transition period.

Model form. Regress baseline gas consumption (for water heating) against relevant independent variables: production volume (tonnes or machine hours), heating-degree days for any space-heating share, and possibly water draw. Check for autocorrelation with interval data and use appropriate time resolution (daily or weekly aggregates are usually adequate for a boiler). Report R², t-statistics, and standard error; require a reasonable model fit before accepting savings.

Non-routine adjustment. The new product line changed the energy–production relationship. Two approaches:

1. If the new line has separate metering, exclude it from the analysis boundary.
2. Otherwise, add a binary or continuous variable for the new line's output and re-estimate, or build a separate post-installation model for the new line and combine.

Savings calculation. Savings = baseline energy (adjusted to reporting-period conditions using the model) − reporting-period energy, both at the measure boundary. Apply the model's standard error to give a confidence interval (e.g., ±10% at 90% confidence). Report uncertainty explicitly and state that savings are conditional on the model's validity.

Why it fails. A year-on-year bill comparison ignores the 18% output increase and weather differences. Higher output means more process water heating, so the post-installation year would show higher gas use than the baseline even if the heat recovery worked perfectly. The comparison would therefore understate savings, possibly showing none.

Error sources:

• Production volume change (+18%): dominant error.
• Weather (degree-day) variation for any space heating.
• Tariff and unit-price changes if comparing costs rather than energy.
• The installation month's mixed operation.

Correct approach. Normalise both years to common conditions using the baseline model, then compare. The verified saving is the difference between adjusted baseline and actual reporting-period consumption, not between two raw bills. I would show the client a simple worked example: if baseline gas for water heating was 2,000 MWh at 10,000 t output and the reporting year produced 11,800 t, the adjusted baseline is ~2,360 MWh; if actual was 1,900 MWh, savings are ~460 MWh - a very different number from a naive bill comparison.