In Uncertainty world the complex phenomena are described with many difficulties especially if their parameters belong to different application fields or their correlations are not well known. As concern the quantification of Uncertainty it’s possible to find many methodologies belonged to the group of Theory of Evidence [1], Fuzzy Logic [3,5], Hybrid Systems (with Neural Networks or Genetic Algorithm), Dempster-Shafer Theory [4], Probability Boxes and so on [2]. The development of the Uncertainty calculation techniques by means of mathematical Belief structures is useful to a best comprehension of a model results and often it’s useful to join these ones with those given by classical methods. The more conventional uses of this kind of techniques are the process monitoring, its the control and relevant factors behaviour diagnostic in real time or its use to make a prevision about the phenomenon evolution. As regard the first and second purposes it’s necessary that the tool has characteristic of promptness and flexibility cause of the rapidity with which the process often evolves and the system used cannot have a calculus structure too difficult to be elaborated. It’s can be said that the main property of this kind of methods is the promptitude of implementation and execution. The third aim with which it’s possible to use these innovative methods doesn’t require the tool promptness, for example this can be utilized for a classification algorithm involving big data sets especially using hybrid techniques (for example Fuzzy-Neural systems). In particular Belief (Bel) functions are useful to measure the inferior limit of the probability of an event occurrence and it’s possible to calculate them with the help Monte Carlo technique [6]. As regard Plausibility (Pl) functions they represent the upper probability limit and they can built as membership functions (Fuzzy Logic), or by means of the Basic Probability Assignment (Theory of Evidence) collecting data by experts in the field or by experiments. We can start by experimental measurements to build both (Bel , Pl) also because the more natural manner to translate the data is by means of numerical intervals of values in terms of frequencies of occurrence. The expected results are additional information as regard their precision and the quantification of Uncertainty on their numerical value, both purposes are very important in engineering and physics applications. As concerns hydrogen based systems we are able to melt the statistical evidence with experts notions reaching a more precise results in this kind of risk analysis which gets until now, few practical applications all over the world from a technical point of view (Hydrogen transportation, distribution and refuelling stations).

### Applications of AI Methods Evaluating Uncertainty in Risk Assessment Analysis and Decision Making approach

#####
*CARCASSI, MARCO NICOLA MARIO;*

##### 2006-01-01

#### Abstract

In Uncertainty world the complex phenomena are described with many difficulties especially if their parameters belong to different application fields or their correlations are not well known. As concern the quantification of Uncertainty it’s possible to find many methodologies belonged to the group of Theory of Evidence [1], Fuzzy Logic [3,5], Hybrid Systems (with Neural Networks or Genetic Algorithm), Dempster-Shafer Theory [4], Probability Boxes and so on [2]. The development of the Uncertainty calculation techniques by means of mathematical Belief structures is useful to a best comprehension of a model results and often it’s useful to join these ones with those given by classical methods. The more conventional uses of this kind of techniques are the process monitoring, its the control and relevant factors behaviour diagnostic in real time or its use to make a prevision about the phenomenon evolution. As regard the first and second purposes it’s necessary that the tool has characteristic of promptness and flexibility cause of the rapidity with which the process often evolves and the system used cannot have a calculus structure too difficult to be elaborated. It’s can be said that the main property of this kind of methods is the promptitude of implementation and execution. The third aim with which it’s possible to use these innovative methods doesn’t require the tool promptness, for example this can be utilized for a classification algorithm involving big data sets especially using hybrid techniques (for example Fuzzy-Neural systems). In particular Belief (Bel) functions are useful to measure the inferior limit of the probability of an event occurrence and it’s possible to calculate them with the help Monte Carlo technique [6]. As regard Plausibility (Pl) functions they represent the upper probability limit and they can built as membership functions (Fuzzy Logic), or by means of the Basic Probability Assignment (Theory of Evidence) collecting data by experts in the field or by experiments. We can start by experimental measurements to build both (Bel , Pl) also because the more natural manner to translate the data is by means of numerical intervals of values in terms of frequencies of occurrence. The expected results are additional information as regard their precision and the quantification of Uncertainty on their numerical value, both purposes are very important in engineering and physics applications. As concerns hydrogen based systems we are able to melt the statistical evidence with experts notions reaching a more precise results in this kind of risk analysis which gets until now, few practical applications all over the world from a technical point of view (Hydrogen transportation, distribution and refuelling stations).I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.