Abstract:
Energy meters need to be calibrated for use in Measurement and Verification (M&V) projects. However, calibration can be prohibitively
expensive and a ect project feasibility negatively. This study presents a novel low-cost in-situ meter data calibration
technique using a relatively low accuracy commercial energy meter as a calibrator. Calibration is achieved by combining two machine
learning tools: the SIMulation EXtrapolation (SIMEX) Measurement Error Model and Bayesian regression. The model is
trained or calibrated on half-hourly building energy data for 24 hours. Measurements are then compared to the true values over
the following months to verify the method. Results show that the hybrid method significantly improves parameter estimates and
goodness of fit when compared to Ordinary Least Squares regression or standard SIMEX. This study also addresses the e ect
of mismeasurement in energy monitoring, and implements a powerful technique for mitigating the bias that arises because of it.
Meters calibrated by the technique presented have adequate accuracy for most M&V applications, at a significantly lower cost.
