Sensing Attribute Weights: A Novel Basic Belief Assignment Method

Area of Science:

• Data fusion
• Artificial intelligence
• Uncertainty quantification

Background:

• Dempster-Shafer evidence theory is valuable for soft sensor data fusion due to its handling of uncertainty.
• Determining basic belief assignment (BBA) remains a challenge, particularly concerning attribute reliability and open-world scenarios.
• Existing BBA methods often neglect attribute reliability and struggle with open-world applications.

Purpose of the Study:

• To propose a novel method for determining BBA in Dempster-Shafer evidence theory.
• To address the limitations of existing methods by incorporating attribute weights and handling both closed and open worlds.
• To enhance the reliability and applicability of soft sensor data fusion systems.

Main Methods:

• Constructing Gaussian models for each attribute using training samples.
• Measuring sample-attribute similarity via Gaussian membership functions.
• Generating attribute weights based on inter-class overlap.
• Determining BBA using the derived attribute weights.

Main Results:

• A novel method for BBA determination is proposed, effective in both closed and open worlds.
• The method explicitly considers attribute reliability through calculated weights.
• Validation on small datasets demonstrates the proposed method's effectiveness.

Conclusions:

• The proposed method offers a robust approach to BBA determination in Dempster-Shafer theory.
• Attribute weighting enhances data fusion accuracy and reliability for soft sensors.
• This work provides a significant advancement for handling uncertainty in complex systems.