Whole-Facility Option C at a Plant With a New Production Line
A food-processing plant (annual electricity 14 GWh, natural gas 90,000 therms) implemented a compressed-air leak program, VSD compressors, and heat recovery on the refrigeration rack. The owner chose IPMVP Option C (whole facility) because sub-metering every end use was deemed too costly. Utility interval data (15-minute) exist for 3 years pre- and 1 year post-install. Midway through the post-install year, the plant added a second production line, raising throughput by 18% and adding a 400-kW process load that runs 16 hours/day, 5 days/week. Production tonnage is recorded daily; heating degree days and cooling degree days are available from a nearby station. The CFO wants a single number for the first post-install year.
1. Build the adjustment logic. Which independent variables belong in the baseline model, and how do you treat the new line?
Read the worked answer
Because the boundary is the whole facility, the baseline model must explain facility energy, so the independent variables should be the facility's real drivers: production tonnage (or a normalized throughput index), HDD, CDD, and possibly a weekday/weekend or shift indicator. Fit separate electricity and gas models - gas is dominated by process steam/heat, electricity by motors and refrigeration; forcing one model on both is a classic Option C error.
The new line is a non-routine adjustment. It is not weather and not ordinary production variation - it is a physical change inside the boundary. Two acceptable treatments: (1) add the new line's metered or engineering-estimated load (400 kW × 16 h × 5 d × weeks) as an explicit adjustment to the reporting-period energy, or (2) if the line's output is captured in the tonnage variable, verify the baseline model's production coefficient still predicts the post-install period using only pre-line months, then extrapolate. Option 2 is only defensible if you can demonstrate model validity on the pre-line portion of the post-install year; otherwise use Option 1 and disclose the estimate.
2. How would you test whether the Option C model is good enough to report savings, and what would make you reject Option C?
Read the worked answer
Check, at minimum: adjusted R² and the t-statistics/sign of each coefficient (a negative production coefficient is a red flag), residual plots against time and against each independent variable, and the CV(RMSE) against a threshold appropriate to the facility (commonly ~5–15% for whole-facility electricity; tighter for gas if the model is stable). Also test for autocorrelation in interval or daily residuals and check that the model's prediction error is small relative to the expected savings.
Reject Option C if the expected savings are within the model's noise band - with an 18% throughput increase mid-year and a 400-kW load, the savings signal from leak repair and VSDs may be swamped. If the combined measures' expected savings are, say, 4% of facility energy and CV(RMSE) is 9%, Option C cannot resolve the savings. In that case fall back to Option B for the compressed-air and refrigeration measures individually, or report Option C with an explicit uncertainty statement and refuse to present a single point estimate.