تجزیه و تحلیل حساسیت و اعتبار سنجی مدل حرارتی ساختمان ها با استفاده از روش های کد الحاقی

The objective of this study is to present a validation method for simulation codes describing thermal behavior of buildings. Beginning with the classical form of comparison of theoretical and experimental results, it is proposed to improve the procedure by calculating the degree of uncertainty associated with the theoretical results. It can be shown that uncertainties associated with code input data, that is, the simulation parameters, propagate through the calculation and generate a range of uncertainty in the results. After describing the problem and objectives of the study, the computation code analysis will be presented. This concerns use of the CA-SIS (Conditionnement d'Air-Simulation de Systèmes) code, and employs the TRNSYS calculation environment. Experimental validation studies for the evaluation of this code have been carried out using ETNA (Essais Thermiques Naturels ou Artificiels) cells, which have been constructed and are maintained at a site near Paris. These cells and experimental procedures are described. To determine ranges of uncertainty in the numerical results, a sensitivity analysis is first carried out by an “adjoint” method. This method and the relationship linking the uncertainty to calculated sensitivities is presented. Notably, it can be demonstrated that the adjoint method simplifies the calculation of uncertainties. The results presented focus on the cell air temperature. The experimental air temperature evolves during the climatic heating sequence within the range of uncertainty of the theoretical results. It can be deduced that the first step of validation being reached, the developer of CA-SIS code can discuss the complete validation for this configuration. The high sensitivity of the internal air temperature to the parameter of heating power shows the limitation of isothermal air volume hypothesis. From this it can be concluded that it is necessary to improve the numerical modeling of the injection heating power.

The purpose of this study is to describe a validation method of complex computation codes for simulation of thermal conditions in buildings. These codes use a large number of input data. These data, which one can vary at will, are also termed parameters. The study is based on the idea that the validation must be made through a comparison of experimental and theoretical results, each set including its level of uncertainty. Usually, experimental results are presented with their uncertainties, which correspond to imperfections in the measurement apparatus. On the other hand, this is rarely the case for theoretical results, which are simply those generated by the simulation. However, data obtained and used for the simulation are themselves associated with an uncertainty which propagates through to the results of the computation. We can come to more appropriate conclusions on the effective validation of a theoretical result if we compare it with the experimental result outcome along with their respective uncertainties. The determination of the uncertainty in computed results is linked to that of data entered into the code. We have the opportunity, using the adjoint-code method, to identify those parameters having the most influence on the magnitude of uncertainty. In order to improve the validation procedure, we can thus emphasize measurement precision for some experimental data, such as the building parameters or climatic perturbations (Fig. 1). Full-size image (17 K) Fig. 1. Validation process of a predictive computation code using theoretical uncertainty range. Figure options Our work focuses on the validation of the CA-SIS (Conditionnement d'Air-Simulation de Systèmes) code developed by the Studies and Research Department of France Electricity (EDF). This code is designed for use by engineers groups studying thermal behavior of buildings. Prior to its use by professionals, the code must also be able to provide them automatically with information concerning the confidence interval of results, i.e., the uncertainty associated with them. For example, an energy requirement calculated for a given project will be indicated with a confidence interval of ±5%. Similarly, a variation of the internal air temperature, used for the prediction of the thermal comfort, will be given at ±7%. The data given here provide important complementary information that we deduce in our validation procedure. We wish to propose a method which is used not only for the validation of the code but also for an automatic analysis of uncertainties. 2 and 3of this paper present the methods to be used in the theoretical (CA-SIS computation code) and experimental (ETNA — Essais Thermiques Naturels ou Artificiels — cell) approach of the problem. In Section 4we describe the adjoint-code method which will allow us to perform a sensitivity analysis concerning the building parameters and climatic perturbations. The purpose of this is to identify data having a strong influence on results. Those inputs whose influence is weak are then separated from the process. In Section 5, uncertainties associated with the theoretical results will be evaluated using the computed sensitivities. Further in Section 5, we compare theoretical results with experimental results, matching each to its level of uncertainty. From this we deduce the level of validation of the code and present some conclusions in Section 6.

Concerning our study, we observe that only a small number of parameters, from the entire ensemble of those necessary for modeling a system, exert a real influence on quantities whose evolution one would like to predict. The study carried out with CA-SIS confirms this hypothesis since we have determined that only 10% (19 out of 198) of thermo-physical parameters of the ETNA cell capsule predominate. To proceed with a sensitivity analysis, we compared and contrasted several deterministic and probabilistic methods. Given that they all provided the same qualitative and quantitative results, they cannot, therefore, be separated on the basis of accuracy of the information they produce but rather on the differences in analysis investment that they demand and on the computation costs that they involve. As for the uncertainty analysis, we are obliged to consider two cases involving the building envelope and the perturbations within the building. Although the uncertainty in the air temperature remains moderate when other parameters fluctuate, by contrast, it increases considerably when the heating power is varied. For a detailed analysis in the face of the problems concerning the influence of the injected power, it appears timely to extend the study by an analysis of discrepancies or residues between measurements and simulations. With the availability of good quality experimental results, one can proceed to the validation of the code for the configuration studied. We have given an outline of the process of validation of CA-SIS in a monozone situation and are currently investigating complex multizone buildings. Finally, one last point should be made: the process of validation — including the analysis of sensitivity — permits not only an improved knowledge of the capacities of the code, but also leads to proposals for improvements to the code. In relation to CA-SIS, we are working to improve the modeling of the behavior of glazed surfaces and to take into account the power injected into internal volumes. Concerning the adjoint-code approach, it have been demonstrated that this method is the most powerful, providing sensitivity to any code output for the entire range of parameters in just one simulation. The computed sensitivities are exact and independent of the number of parameters included in the analysis. The adjoint-code approach is nonetheless the most difficult to implement, requiring that the code have to be duplicated. Because it is a very big task, this constraint is only acceptable in case of systematic sensitivity and uncertainty analysis. Even so, the adjoint-code method remains very useful in the absence of experimental results. Indeed, it determines influential parameters. The values of these parameters must be known with great accuracy in order to limit the uncertainty in the numerical prediction. Nevertheless, this method is not recommended in case of very complex code, as computational fluid dynamics, because duplication and implementation will be a too expensive and too long task.