Modeling Concepts and Application
Models are used to normalize baseline energy to reporting-period conditions and to estimate savings when direct comparison is impractical. Common model types: mean-baseline (constant), linear regression (single variable such as HDD or CDD), multiple regression (weather plus production or occupancy), change-point models (temperature-dependent with a balance point), and calibrated simulation (Option D).
Key concepts: independent vs. dependent variables, model form selection, balance point, R² and adjusted R², standard error of the estimate, t-statistics for variable significance, autocorrelation, multicollinearity, extrapolation risk, and calibration criteria for simulation models (e.g., comparison of simulated vs. measured monthly energy within accepted tolerances). Models should be built on a baseline period that represents the range of conditions expected in the reporting period; extrapolation beyond the baseline range increases uncertainty.
Traps: choosing a model with high R² but nonsensical coefficients; using a model outside its valid range; ignoring autocorrelation in hourly or daily data (which understates uncertainty); and treating a change-point model as a simple linear model. Questions may ask when a change-point model is preferred (when energy use is flat below/above a balance temperature).
Worked example: Monthly gas use vs. HDD yields use = 200 + 1.5 × HDD with R² = 0.90. For a reporting month with HDD = 400, predicted baseline = 200 + 1.5 × 400 = 800 units. If actual = 720, savings = 80 units. If HDD in the reporting period exceeds the baseline maximum, flag the extrapolation and widen the uncertainty statement.