A novel belief rule base expert system with interval-valued references.

Area of Science:

• Decision Support Systems
• Artificial Intelligence
• Expert Systems

Background:

• Referential values are crucial for belief rule base (BRB) modeling accuracy.
• Current BRB studies primarily use single-valued references, which inadequately represent uncertain expert knowledge.
• Expert knowledge often involves inherent uncertainty, ambiguity, and vagueness, necessitating more robust reference types.

Purpose of the Study:

• To propose a novel belief rule base with interval-valued references (BRB-IR).
• To enhance the representation of qualitative expert knowledge within BRB models.
• To improve the accuracy and effectiveness of BRB models by integrating diverse data types.

Main Methods:

• Developed a BRB incorporating interval-valued referential values (BRB-IR).
• Optimized interval-valued references using a nonlinear optimization algorithm.
• Tuned other model parameters with the projection covariance matrix adaptation evolutionary strategy (P-CMA-ES).

Main Results:

• The proposed BRB-IR effectively integrates qualitative and quantitative information.
• Optimized interval-valued references improved model performance.
• The BRB-IR demonstrated superior effectiveness compared to classical BRBs with single-valued references in a pipeline leak detection case study.

Conclusions:

• The BRB-IR model offers a more effective approach to characterizing expert knowledge.
• Interval-valued references provide a better way to handle uncertainty and vagueness in expert elicitation.
• The proposed method enhances the accuracy and applicability of belief rule base systems.