Building energy performance simulation benchmarking by metamodel-based sensitivity assessment

  • Building energy performance simulation benchmarking by metamodel based sensitivity assessment

Nouri, Amin; van Treeck, Christoph Alban (Thesis advisor); Grunewald, John (Thesis advisor)

Aachen : RWTH Aachen University (2023)
Dissertation / PhD Thesis

Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2023


Large discrepancies can occur between Building Energy Performance Simulation (BEPS) outputs and the actual building energy performance. Uncertainty and sensitivity analyses are performed to discover significant contributions of each input parameter to these discrepancies. Variance-based sensitivity analyses such as the Sobol’ method typically require a large number of stochastic simulations which is computationally demanding and time-consuming. In order to overcome these impediments, the presented study proposes a fast, reliable four-stage metamodel-based sensitivity analysis framework, including validation (benchmarking), Morris’ method, meta modeling and Sobol’ method, to identify the most influential input parameters on BEPS prediction output (i.e., annual energy consumption).Benchmarking and evaluation of the model quality are conducted using eight systematic test cases to assess for the accuracy of the building models. The screening-based sensitivity analysis using Morris’ method is performed prior to applying Sobol’ method to reduce the computational cost by selecting key input parameters influencing model output (annual energy consumption). A hypothetical building with one single thermal zone is used to analyze the proposed methodology. Four popular meta-models, i.e., Multivariate Adaptive Regression Splines (MARS), Polynomial Regression (PR), Random Forest (RF) and Support Vector Regression (SVR) based on a radial basis function (RBF) kernel, are evaluated to compare their performance using statistical performance metrics. The comparison shows that the MARS meta-model achieves the best performance. The variance-based global sensitivity analysis using Sobol’ method in combination with the developed MARS meta-model is implemented to perform the final, robust stage of the sensitivity analysis and to rank the most influential input parameters for the analyzed case.


  • Institute of Energy Efficiency and Sustainable Building [312410]