Detection of Cause-Effect Relations Based on Information Granulation and Transfer Entropy.

Area of Science:

• Complex Systems Analysis
• Industrial Process Monitoring
• Data Science

Background:

• Causality inference identifies cause-effect relationships in complex systems, crucial for root cause analysis in industries.
• Transfer entropy (TE) is a non-parametric method effective for detecting causality in linear and nonlinear systems.
• High computational complexity of TE limits its application in real-time systems.

Purpose of the Study:

• To develop an improved causality inference method addressing the computational limitations of transfer entropy.
• To integrate information granulation and delay estimation for efficient real-time causality detection.
• To enhance the applicability of transfer entropy in large-scale industrial processes.

Main Methods:

• A novel framework integrating information granulation as a preprocessing step for transfer entropy calculation.
• A window-length determination method based on delay estimation for optimized data compression.
• Validation using a numerical example and an industrial case study with a two-tank simulation model.

Main Results:

• Significant reduction in computational complexity compared to traditional transfer entropy methods.
• Maintained high accuracy in detecting cause-effect relationships.
• Demonstrated effectiveness in a simulated industrial process (two-tank model).

Conclusions:

• The proposed method enhances transfer entropy by incorporating information granulation and delay estimation.
• This approach offers a computationally efficient and accurate solution for real-time causality inference.
• The method shows strong potential for root cause analysis in complex industrial systems.