A New Evidential Reasoning Rule With Continuous Probability Distribution of Reliability

Area of Science:

• Artificial Intelligence
• Decision Science
• Probability Theory

Background:

• Evidential Reasoning (ER) rules are crucial for managing uncertainty in decision-making.
• Current ER rules often use single quantitative values for evidence reliability, which can oversimplify and lead to inaccurate results.
• The statistical properties of reliability are essential for robust evidential reasoning.

Purpose of the Study:

• To propose a novel Evidential Reasoning rule with continuous probability distribution of reliability (ERr-CR).
• To address the limitations of single-value reliability representation in existing ER rules.
• To enhance the accuracy and applicability of evidential reasoning in complex scenarios.

Main Methods:

• Introduced ERr-CR, where evidence reliability is modeled as random variables with probability distributions.
• Developed a method to characterize ERr-CR outputs using the expectation of expected utility.
• Extended the ERr-CR framework to handle multiple pieces of evidence.

Main Results:

• The proposed ERr-CR effectively handles the statistical nature of reliability, overcoming limitations of single-value approaches.
• The expectation of expected utility provides a robust measure for the output of the new ER rule.
• Demonstrated the universality and rationality of ERr-CR through theoretical exploration and performance analysis.

Conclusions:

• ERr-CR offers a more sophisticated and accurate approach to evidential reasoning under uncertainty.
• The method shows practical applicability, as evidenced by a case study on natural gas storage tank safety assessment.
• This work advances the field of uncertainty management and evidential reasoning.