Statistical methodologies for verification of building energy performance simulation
Amin Nouri, Jérôme Frisch, Christoph van Treeck
Abstract
Building energy performance simulation tools are being increasingly deployed by researchers and professionals to predict the thermal behavior of buildings. Validation methods are applied to ensure the accuracy of simulation results. Results and methods described in this paper are resulting from a research project funded by the German Federal Ministry for Economic Affairs and Energy (BMWi), which addresses the development of quality standards for building and systems energy performance simulations. The objective of this project is to develop a validation methodology, to define standards for simulation applications and to transfer them into planning practice. Another aspect of the research project is the development of a platform to provide a facility for defining individual test cases, to create individual simulation code, to perform a comparative validation, and to evaluate the accuracy of the simulation tools. The first part of this paper interprets the sources of error and uncertainty in building simulation and presents five statistical indices Mean Bias Error (MBE), Normalized Mean Bias Error (NMBE), Root Mean Squared Error (RMSE), Coefficient of Variance of Root Mean Squared Error (CVRMSE) and Coefficient of Determination (R2), which are used in the verification and validation of the building energy performance simulation tools. The second part describes a systematic set of test cases based on the ANSI/ASHRAE Standard 140 and discusses the simulation approach in Modelica/Dymola. The third part presents the development and implementation of statistical indices and evaluates these indices' ability to deduce deviations in simulation results based on their interpretation. It is shown that the R2 give a different interpretation of the discrepancies between predicted and reference values and it is recommended to apply this index only along with other statistical indices to correctly evaluate the model accuracy.