Calibration of the Energy Simulation Model of a Library with a Meta-Model-Based Optimization Approach

Building simulations play a fundamental role both in applications like the design of new constructions and the optimization of building operation and control. This is quite relevant in the current energy framework, in which the energy consumption of buildings has increased over past decades. The reliability of the results’ model does not depend only on the model itself, like the mathematical expression or the resolution process, but it is also related to the uncertainty that those parameters involve. This can cause discrepancies between the simulated and the real behavior of the building, causing a deviation from the expected one of the performance of a building. Hence, the calibration procedure of the model is a necessary process which allows more accurate results to be obtained and predictions that are closer to the real behavior of the building to be made, minimizing the discrepancy between predicted and actual performance by changing the values of the simulation parameters. When it comes to calibration of simulation models, many approaches are available in the literature, comprising manual and iterative ones, graphic comparative procedures, techniques based on specific tests, and many others. Among all possible approaches, optimization-based calibration is the most widely adopted in model calibration. However, this approach, which is usually based on evolutionary algorithms, has the disadvantage that it requires many expensive simulations to be run, especially when the number of parameters to be calibrated is high. This issue can be overcome by a preliminary sensitivity analysis that reduces the number of parameters to be calibrated and by an efficient optimization algorithm. For this reason, this work proposes a framework based on a sensitivity analysis designed to identify the most significant parameters separately on the energy budgets and other monitored environmental variables. The proposed calibration procedure is based on functional approximation models, which greatly increases the efficiency of the optimization algorithm. The case study is a university library placed in the municipality of Trento, Italy. The building was monitored in terms of indoor carbon dioxide, indoor temperature, and relative humidity. Results show how successful the proposed approach is in reducing the computational time required for calibration, especially when considering models with a high degree of complexity.