Exploration On Two Kinds Of Uncertainty Reasoning Of Causality Diagrams

The research on uncertain problems is the core task of artificial intelligence.At present,the solutions to the problem of uncertainty are roughly classified into two categories: one is the probability-based method and the other is based on the non-probabilistic method.The causality diagram reasoning is a kind of uncertainty reasoning model based on probability theory knowledge.The dynamic causality diagrams directly express the uncertain relationship of the system in graphics,and we establish a qualitative model to study a complex system.This paper deals with the probability of connection events in multivalued causal graphs in an appropriate way surrounding the knowledge expression and reasoning of causal graphs.The basic event probability in causality diagram is replaced by interval number which can show the uncertainty of the system more vividly.The main contents are as follows:(1)When we exhibit a complicated system in causality diagrams to diagnose the system failures,we use the node event to express the fault source and the causal relationship is expressed by the directed edge.Due to the differences of the assigned numbers of sub-variables,the causality diagrams are divided into single-valued causal graphs and multivariate causal graphs.When the traditional causal reasoning is applied to multivariate causal graphs,the phenomenon of probability nonuniformity arises.Thus,an inexact reasoning of multivalued causality is proposed.This reasoning method is based on the degree of causality to find the probability value of the connection events.And the probability value is obtained on the condition that the default events and connection events are added and the mutual exclusion between the events is assumed.We transform the probability matrix into a definite probability value according to a kind of possibility allocation,which transforms the multivariate causal graph into a single value causal graph,thus solving the reasoning problem of multivalued causal graph.(2)As a knowledge expression method based on probability,causality diagram is to perform derivation and calculation when the exact probability value of the event is known.In the actual situation,it is difficult to obtain the accurate probability value due to the errors of data,the subjective bias of the experts and so on.In this case,we propose to extend the exact value into the interval number.According to the Dempster-Shafer evidence theory(DS theory),we merge the expert knowledge or system data together,and obtain the Pls function(Plausibility Function)and the Bel function(Belief Function)by means of calculation.And then we use them as the upper and lower bounds of the probability interval,thus effectively solving the ambiguity and uncertainty problems of the system and reducing the difficulty of obtaining accurate probability value at the same time.