Civil Engineering Reference
In-Depth Information
13.5.2.2
Latin Hypercube Sampling Method (Stratifi ed Sampling)
In this method members of the population are divided into homogeneous segments
before sampling. Every sample shall be assigned to only one group, and all the
groups together shall create a whole body of possible samples. Then random sam-
pling applies within each segment (Khazaii
2012
).
Therefore the main difference between this method and Monte Carlo method is
in their input data selection distribution. Experience shows this method usually
requires less number of iterations than the Monte Carlo method.
13.5.2.3
Random Sampling
In this method samples are collected from a random collection such as random num-
bers from a database and will be scattered without any specifi c rules inside the sam-
pling fi eld. This method can create clusters (when samples are very close to each
other) or gaps (when very little number of samples are taken from some regions.)
13.5.2.4
Screening Method
Screening method is a particular case of sampling based methods which deals with
each input individually. It is also known as once at a time (OAT) method. After all
the required data sequentially modifi ed the designer evaluates the results and makes
decision based on the comparison among the outputs.
13.5.2.5
Variance-Based Method
Variance-based methods are sampling-based methods that also depend on the
computation of conditional variances. They allow a global, quantitative, and model-
independent sensitivity measure. This method is known to be more complex than
the other methods.
13.6
Internal Methods
These methods deal with conditions that the uncertainty is considered in the arith-
metical equations of the model. Since these methods are shown to have not particu-
lar use in uncertainty analysis in building energy modeling, we will stop our
description here.
