Williamson, T. J.
Bibliographic info:
Building Simulation, Madison, USA, 1995, p. 268-275

This paper deals with the problem of empirical validation of thermal performance computer programs. It begins with a brief review of a number of techniques which have been used as a measure of the goodness-of-fit between measured and predicted data in a variety of empirical validation exercises. Several inadequacies inherent in existing techniques are identified as, a) no attempt is made to take into account the severity of the validation test. b) none give a single measure of the success (or otherwise) of the test. c) isolation of sources of error are difficult. d) tests cannot be used easily for internal validation and/or algorithm "tuning". An argument is presented that an objective technique for establishing the accuracy of simulation predictions and which addresses these inadequacies is required. To satisfy this requirement a confirmation factor Cs based on an inequality coefficient is defined. Following from Cs it is shown that a degree of confirmation D may be evaluated. Multiple input variables may be taken into account by calculating Cs and D from the estimates of the Principal Components derived from independent linear functions of the original variables. The confirmation technique is illustrated with an example comparing the predictions of the computer program TEMPAL with internal temperatures derived from monitoring a house. Possible sources of program error are investigated and the improvements made with a new program EnCom2 are shown.