Sensor errors greatly affect the performance of control, diagnosis, and optimization systems within building energy systems, negatively impacting energy efficiency. Virtual in-situ sensor calibration (VIC), a Bayesian theory based method, can improve building energy performance by calibrating erroneous sensors in working building energy systems on a large scale. Working sensors do not need to be removed nor will reference sensors need to be added, as is done in a conventional calibration. To improve the calibration accuracy, hidden factors and their negative effects on the accuracy of a VIC must be addressed properly. In this study, we define (1) prior information and (2) cancellation effects as the negative effects. The suggested VIC method is applied to a single energy system component and to a LiBr-H2O absorption refrigeration system, respectively, to discuss the two primary effects (mentioned above). In addition to adding data sets, two strategies-inclusion of local calibration and conducting repetitive prior updates-are proposed to solve the hidden factors' issue. The case study (1) shows that the proposed local calibration with the prior updates can solve the two negative effects, thus suggesting the high calibration accuracy and (2) demonstrates that the calibrated measurements improve the accuracy of energy performance analysis for a building energy system (up to 17.82%).