کاربرد عملی از تجزیه و تحلیل عدم قطعیت و تجزیه و تحلیل حساسیت در یک خانه تجربی

Today, simulation tools are widely used to design buildings because their energy performance is increasing. Simulation is used at different stages to predict the building's energy performance and to improve the thermal comfort of its occupants, but also to reduce the environmental impact of the building over its whole life cycle and lower the cost of construction and operation. Simulation has become an essential decision support tool, but its reliability should not be overlooked. It is important to evaluate the reliability of simulation and measurement as well as uncertainty so as to improve building design. This work aimed to evaluate and order the uncertainty of the simulation results during the design process. A three-step methodology was developed to determine influential parameters in the building's energy performance and to identify the influence of parameter uncertainty on the building performance. This methodology was applied at the INCAS experimental platform of the French National Institute of Solar Energy (INES) in Le-Bourget-du-Lac to identify and measure the uncertainty in a simulation hypothesis. The method can be used during the entire design process of a building, from preliminary sketches to operating phase.

Today, increasingly energy-efficient buildings are being built and energy performance simulation is widely used in the design process. Simulation is used to predict the building's energy performance and to improve the thermal comfort of its occupants, but also to reduce the environmental impact of the building over its whole life cycle and to lower the cost of construction and operation. The energy performance of buildings is increasing rapidly due to thermal regulations instituted by several countries. Today an efficient building needs almost no energy for heating, whereas 10 years ago energy consumption was around 200 kWh/(m2 year). To check the building performance that was defined during the project's design phase, sensors are installed and measured data are compared with simulation results, and usually a difference between the two is found. Won-Jun et al. [1] compared simulation results with measurements for a campus library building in Suwon City, South Korea. Differences were observed that may have been the result of errors in, for example, input parameters, occupation scenarios, or weather data. To improve the simulation results and the building design process, some studies focused on assessing uncertainty in simulation output (e.g., [2], [3], [4] and [5]). Many other studies focused on checking the model validity by comparing predictions with measured data (e.g., [6], [7] and [8]). It is important to explain the difference between measured and predicted data and to try to minimize it. This paper presents a method applied to an experimental low-energy building located in an experimental platform at the French National Institute of Solar Energy near Chambéry, France. The aim of this work was to provide a comparison between predictions and the first available measurements. The studied house was recently built for a research program and much attention was focused on the construction details. However, there is always uncertainty; for example, there is a difference between the theoretical flow ventilation and the one measured. We wanted to evaluate the effects of the input parameter uncertainty on the predicted air temperature. For this reason, the most influential parameters were first determined by means of local sensitivity analysis and global sensitivity analysis. Uncertainty analysis was then performed using a Monte Carlo method to predict air temperature within uncertainty bands.

Influential parameters on air temperature of one experimental energy-efficient house were studied. Sensitivity analysis is a method that determines which of the uncertainty input factors is more important in determining the uncertainty in the output of interest. In this study, we used three sensitivity analyses: - Local sensitivity analysis - Global sensitivity analysis - Uncertainty analysis Local sensitivity analysis identifies the most important factors among a large number of parameters. This analysis uses the parameter perturbation method. A sensitivity index is determined for each parameter and the influential parameters are ordered. Global sensitivity analysis is the study of how uncertainty in the output of a model can be apportioned to different sources of uncertainty in the model input. The Sobol method was used with a small number of parameters. Uncertainty analysis focuses on quantifying uncertainty in model output. The test case used in our study is a heavily instrumented experimental house on the INCAS platform. First, a numerical model was defined and run with a user-defined weather file in EnergyPlus. A comparison between measured and simulation data was made and uncertainty was considered. Then local sensitivity analysis was run in order to define a set of the most dominant input parameters, and correlation analysis was used to group the parameters. At the beginning, we had 139 parameters from which we selected the 10 most influential parameters on the output air temperature. This set of parameters was used in an uncertainty analysis. We chose a distribution for each parameter and we performed 2000 simulations. Finally, the numerical uncertainty band was compared with the experimental data. The comparison between the experiment and simulation data gave a satisfactory result for the ground air temperature but not for the first floor temperature. To improve the result between the measured and simulation data comparison, one can either extend the uncertainty band of some parameters because the defined uncertainty band may be too much optimistic or to ascertain more precisely the value of the influential parameter identified with global sensitivity analysis. If the measured uncertainty for certain parameters is improved, for instance, the capacity of electric heating, the heat exchanger efficiency and infiltration simulation results will be reliable. The aim of this work was to define this methodology, which is useful for identifying the most important parameter and determining a simulation uncertainty band. In conclusion, it is important to know more precisely the value of the most influential parameters and to be able to reduce the uncertainty band. There are many perspectives for continuing and improving this work: extend this methodology to a full annual comparison, analyze the sources of differences between experimental and numerical results, improve the numerical model, reduce the model uncertainty band, improve the knowledge on input parameter uncertainty, and analyze the parameter interdependence. On another hand chaos polynomials formulation will be investigated to reduce simulation time and make this approach more suitable in practice. It consists in building a low order meta-model that approximates the original model in order to perform simulations much faster while getting information about sensitivity against input parameters.