22.05.2017 Views

2007_6_Nr6_EEMJ

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Robu et al. /Environmental Engineering and Management Journal 6 (<strong>2007</strong>), 6, 573-592<br />

It is very important to identify the risk and to<br />

estimate it in order to be analyzed. The risk analyzing<br />

process tries to identify all the results of an action.<br />

The risk estimation is done using the analytic<br />

methods or simulation. There are estimated thus the<br />

occurrence probability of every disaster, as well as<br />

the associated magnitude (dimension). The risk<br />

analysis process uses technical information related to<br />

estimations and other additional available<br />

information, for assessing diverse variants of possible<br />

actions. An original model based on multiple criteria<br />

optimization to appropriate financial funds for<br />

atmospheric pollution reduction is presented. For this<br />

model, the solving manner is specified by reduction<br />

to an optimization problem with a single objective<br />

function (Radulescu, 2002).<br />

5.1. Measures for calculus of the risk value<br />

The probability theory offers many adequate<br />

tools for modeling the risk phenomenon. Any activity<br />

exhibits an incertitude element. From mathematical<br />

point of view, the incertitude will be modeled using<br />

random variables or, more generally, using the<br />

stochastic processes. The risk that appears within an<br />

activity may be described with adequate measures.<br />

One of the most used measures is the dispersion of<br />

the random variable, which describes the incertitude<br />

of the respective activity. Another measure of the risk<br />

is given by the repartition function of the random<br />

variable. More precisely, if X is a random variable<br />

that describes the risk associated to a decision, F x is<br />

the repartition function associated to variable X, and f x<br />

is the probability density of the random variable X,<br />

then using Eq. (2):<br />

+∞<br />

∫<br />

−∞<br />

µ = E(X ) = xdF ( x)<br />

, (2)<br />

the risk measures result from Eqs. (3, 4):<br />

+∞<br />

∫ (<br />

−∞<br />

2<br />

2<br />

σ = var( X ) = x − µ ) dF ( x)<br />

(3)<br />

+∞<br />

1/ 2<br />

⎤<br />

2<br />

( x − µ ) dFx<br />

( x)<br />

⎥<br />

−∞<br />

⎦<br />

⎡<br />

σ = ⎢ ∫<br />

(4)<br />

⎣<br />

A measure of the risk may be considered also:<br />

+∞<br />

∫<br />

−∞<br />

| x − µ | dF ( x)<br />

(first order central moment) (5)<br />

x<br />

Stone has shown that all the measures of the<br />

risks above presented are special cases of some<br />

families of risks measures. The first measure of the<br />

risk within the Stone risk measures family that has<br />

three parameters is defined as (Eqs.2-6):<br />

x<br />

q<br />

( F x )<br />

| (k>0 (6)<br />

∫<br />

−∞<br />

k<br />

R S1 (X) = x − p(<br />

F ) | dF ( x)<br />

x<br />

where p\F x ) defines a level of the profit (success) that<br />

is used for measuring the abatement.<br />

The positive number k is a measure of the<br />

relative impact of the small and big abatements. q(F x )<br />

is a level a parameter that specifies the abatements<br />

will be included in the risk measurement. The second<br />

measure of the risk within Stone family with three<br />

parameters is defined as k order root from R S1 (X)<br />

(Eq. 7):<br />

⎡<br />

R S2 (X) = ⎢<br />

⎢⎣<br />

q(<br />

Fx<br />

)<br />

∫<br />

−∞<br />

| x − p(<br />

F<br />

x<br />

) |<br />

k<br />

x<br />

1<br />

k<br />

⎤<br />

⋅dFx<br />

( x)<br />

⎥<br />

⎥⎦<br />

(7)<br />

One may observe that through proper<br />

selection of the parameters p(F X ), q{F x ) and k, the<br />

above presented risk measurement are special cases<br />

of those from the Stone family of risk measurements.<br />

For example, if in (l) k = 2 and p(F x ) = (F x )<br />

= µ are inserted, the semi-dispersion is obtained as a<br />

measure of the risk. A more interesting case related to<br />

risk measures Stone family is the generalized risk<br />

measure Eq. (8):<br />

t<br />

R F1 (X) =<br />

∫<br />

−∞<br />

(t - x) α dF x (x) (α > 0), (8)<br />

proposed by Fishburn, where t is a superior scopelevel<br />

that is fixed. This measure results from (l) if one<br />

choose p(F x )=q(F X ) = t. The parameter “a” of<br />

Fishburn risk measurement R F1 may be interpreted as<br />

“k” parameter from the measures Stone family (l) in<br />

this way: it is a risk parameter, which characterizes<br />

the attitude toward risk. The values α > l describe a<br />

sensitive risk, while the values α ∈(0. l) describe an<br />

insensitive risk.<br />

Another known measure of the risk is<br />

Shannon entropy (Eq. 9):<br />

+∞<br />

∫<br />

−∞<br />

fx( x)ln(<br />

f<br />

x<br />

( x))<br />

dx<br />

(9)<br />

5.2. Probabilistic modeling of pollution<br />

Mathematic modeling of air pollution is<br />

done using the theory of stochastic processes. Thus,<br />

over long periods of time, the pollution degree may<br />

be described by a multidimensional stochastic<br />

process: X = (X t ) t ≥ 0 . For t ≥ 0 attached to the<br />

components of the random vector: X t = (X l,t ;<br />

X 2,t ...X m,t ) represents the concentrations of the<br />

atmospheric pollutant factors. Repartition function of<br />

the stochastic multidimensional process X, F(t,x) =<br />

584

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!