著者

菊本 英紀 崔 元 準 大岡 龍三

出版者

日本建築学会

雑誌

日本建築学会環境系論文集 (ISSN:13480685)

巻号頁・発行日

vol.82, no.734, pp.357-365, 2017 (Released:2017-04-30)

参考文献数

23

---

Natural ventilation is a good measure to reduce the thermal load of a building and create a good thermal environment for the occupants. To predict the ventilation performance of a building, related ventilation parameters should be estimated. However, obtaining an accurate prediction or estimation before the construction is completed is difficult because of the fluctuating characteristics of turbulent, complex local flow field affected by surrounding buildings and city-scale topology. These problems make the application of natural ventilation difficult, and therefore, an in situ experimental method adopted after the construction of a building is completed would be a more reliable way of obtaining accurate ventilation parameters. The tracer gas method is a frequently used in situ method to examine ventilation performance. However, deterministic evaluation using this method has its drawbacks in terms of accuracy, and the uncertainty of results could be large because of fluctuating environmental conditions and intrinsic error of measurement. Therefore, when parameters related to natural ventilation are estimated, the estimates and their uncertainty should be concurrently evaluated. In this study, using a Bayesian approach, we propose a new estimation method that yields not only the estimates, but also the probability distribution of the estimates. To validate the proposed method, we reproduced a tracer gas experiment using computational fluid dynamics (CFD) simulation because the experimental absolute true values that are unknown in most situations were required for evaluating the proposed method. Using the data from CFD simulation, estimations of the ventilation rate and effective room volume were obtained using two different methods: probabilistic estimation based on Bayes' theorem, and deterministic estimation using the quasi-Newton method. In the probabilistic estimation, a method of simultaneously estimating the degree of the difference between the measurement value and the prediction value required for estimation was proposed. The estimation results showed that the estimated values of the deterministic method and probabilistic method (mode of probability distribution function) were very close to the true value set in the CFD simulation. However, the probabilistic method has some advantages over the deterministic method in that the former method can provide not only the estimates, but also the probability distribution of each estimate and its confidence interval. Additionally, in the case of a simultaneous estimation of more than two unknown parameters, it is important to analyze the correlation among parameters. By creating a joint probability density function, the correlation between the estimation parameters is determined, and this is another important advantage of the probabilistic estimation method.