1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
|
.. _sensitivity_fast:
Sensitivity analysis by Fourier decomposition
---------------------------------------------
FAST is a sensitivity analysis method which is based upon the ANOVA
decomposition of the variance of the model response
:math:`y = \model(\inputRV)`, the latter being represented by its Fourier
expansion. :math:`\inputRV=\{X^1,\dots,X^\inputDim\}` is an input random
vector of :math:`\inputDim` independent components.
| The extended FAST method consists in computing
alternately the first order and the total-effect indices of each
input. This approach relies upon a Fourier decomposition of the model
response. Its key idea is to recast this representation as a function
of a :math:`\mathit{scalar}` parameter :math:`s`, by defining
parametric curves :math:`s \mapsto x_i(s)`, :math:`i=1, \dots, \inputDim`
exploring the support of the input random vector :math:`\inputRV`.
| For each input, the same procedure is realized in three steps:
- Sampling:
Deterministic space-filling paths with random starting points are
defined, i.e. each input :math:`X^i` is transformed as follows:
.. math::
x^i_j = \frac{1}{2} + \frac{1}{\pi} \arcsin(\sin(\omega_i s_j + \phi_i)),
\quad i=1, \dots, \inputDim, \, j=1, \dots, \sampleSize
where :math:`\inputDim` is the number of input variables. :math:`\sampleSize` is the
length of the discretization of the s-space, with :math:`s` varying
in :math:`(-\pi, \pi)` by step of :math:`2\pi/\sampleSize`. :math:`\phi_i` is a
random phase-shift chosen uniformly in :math:`[0, 2\pi]` which
enables to make the curves start anywhere within the unit hypercube
:math:`K^{\inputDim}=(\vect{x}|0\leq x_i\leq1; i=1, \dots, \inputDim)`. The
selection of the set :math:`\{\phi_1, \dots, \phi_{\inputDim}\}` induces a
part of randomness in the procedure. So it can be asked to realize
the procedure :math:`\sampleSize r` times and then to calculate the arithmetic
means of the results over the :math:`\sampleSize r` estimates. This operation is
called :math:`\mathit{resampling}`.
The set :math:`\{\omega_i\}, \forall i=1, \dots, \inputDim` is a set of integer
frequencies assigned to each input :math:`X^i`. The frequency
associated with the input of interest is set to the maximum
admissible frequency satisfying the Nyquist criterion (which ensures
to avoid aliasing effects):
.. math:: \omega_i = \frac{\sampleSize - 1}{2M}
with :math:`M` the interference factor usually equal to 4 or higher.
It corresponds to the truncation level of the Fourier series, i.e.
the number of harmonics that are retained in the decomposition
realized in the third step of the procedure.
In the paper [saltelli1999]_, for high sample size, it is
suggested that :math:`16 \leq \omega_i/\sampleSize_r \leq 64`.
And the maximum frequency of the complementary set of frequencies is:
.. math:: \max(\omega_{-i}) = \frac{\omega_i}{2M} = \frac{\sampleSize - 1}{4M^2}
with the index ’\ :math:`-i`\ ’ which meaning ’all but :math:`i`\ ’.
The other frequencies are distributed uniformly between :math:`1` and
:math:`\max(\omega_{-i})`. The set of frequencies is the same
whatever the number of resamplings is.
Let us make an example with eight input factors, :math:`\sampleSize=513` and
:math:`M=4` i.e. :math:`\omega_i = \frac{\sampleSize - 1}{2M} = 64` and
:math:`\max(\omega_{-i}) = \frac{\sampleSize - 1}{4M^2} = 8` with :math:`i` the
index of the input of interest.
| When computing the sensitivity indices for the first input, the
considered set of frequencies is :
:math:`\{64, 1, 2, 3, 4, 5, 6, 8\}`.
| When computing the sensitivity indices for the second input, the
considered set of frequencies is :
:math:`\{1, 64, 2, 3, 4, 5, 6, 8\}`.
| etc.
The transformation defined above provides a uniformly distributed
sample for the :math:`x_i, \forall i=1, \dots, \inputDim` oscillating
between :math:`0` and :math:`1`. In order to take into account the
real distributions of the inputs, we apply an isoprobabilistic
transformation on each :math:`x_i` before the next step of the
procedure.
- Simulations:
Output is computed such as:
:math:`y = f(s) = f(x_1(s), \dots, x_\inputDim(s))`
Then :math:`f(s)` is expanded onto a Fourier series:
.. math:: f(s) = \sum_{k \in \Zset^\sampleSize} A_k \cos(ks) + B_k \sin(ks)
where :math:`A_k` and :math:`B_k` are Fourier coefficients defined as
follows:
.. math::
\begin{aligned}
A_k &=& \frac{1}{2\pi}\int_{-\pi}^{\pi}f(s) \cos(ks) \, ds \\
B_k &=& \frac{1}{2\pi}\int_{-\pi}^{\pi}f(s) \sin(ks) \, ds
\end{aligned}
These coefficients are estimated thanks to the following discrete
formulations:
.. math::
\begin{aligned}
\hat{A}_k &=& \frac{1}{\sampleSize} \sum_{j=1}^\sampleSize f(x_j^1,\dots,x_j^\inputDim) cos \left( \frac{2k\pi (j-1)}{\sampleSize} \right) \quad , \quad -\frac{\sampleSize}{2} \leq k \leq \frac{\sampleSize}{2}\\
\hat{B}_k &=& \frac{1}{\sampleSize} \sum_{j=1}^\sampleSize f(x_j^1,\dots,x_j^\inputDim) sin \left( \frac{2k\pi (j-1)}{\sampleSize} \right) \quad , \quad -\frac{\sampleSize}{2} \leq k \leq \frac{\sampleSize}{2}
\end{aligned}
- Estimations by frequency analysis:
The first order indices are estimated as follows:
.. math::
\hat{S}_i = \frac{\hat{D}_i}{\hat{D}} \\
= \frac{\sum_{p=1}^M(\hat{A}_{p\omega_i}^2 + \hat{B}_{p\omega_i}^2)^2}
{\sum_{n=1}^{(\sampleSize-1)/2}(\hat{A}_n^2 + \hat{B}_n^2)^2}
where :math:`\hat{D}` is the total variance and :math:`\hat{D}_i` the
portion of :math:`D` arising from the uncertainty of the
:math:`i^{th}` input. :math:`\sampleSize` the size of the sample using to
compute the Fourier series and :math:`M` is the interference factor.
Saltelli et al. (1999) recommended to set :math:`M` to a value in the
range :math:`[4, 6]`.
The total order indices are estimated as follows:
.. math::
\hat{T}_i = 1 - \frac{\hat{D}_{-i}}{\hat{D}} \\
= 1 - \frac{\sum_{k=1}^{\omega_i/2}(\hat{A}_k^2 + \hat{B}_k^2)^2}
{\sum_{n=1}^{(\sampleSize-1)/2}(\hat{A}_n^2 + \hat{B}_n^2)^2}
where :math:`\hat{D}_{-i}` is the part of the variance due to all the
inputs except the :math:`i^{th}` input.
.. topic:: API:
- See :class:`~openturns.FAST`
.. topic:: Examples:
- See :doc:`/auto_reliability_sensitivity/sensitivity_analysis/plot_sensitivity_fast`
.. topic:: References:
- [saltelli1999]_
|
