This chapter discusses data-mining-based catastrophe predictors using PMU-based WASI features. The study validates the performance of the black-box models such as SVM and RF through semitransparent data-mining models such as DT and transparent models such as DT_Fuzzy. It is observed from the extensive studies of the developed data-mining-based catastrophe predictors that while switching from the black box solutions to transparent and interpretable solutions, there is an unavoidable trade-off between accuracy, reliability, and security measures. The more transparent the predictor, the easier the implementation and maintenance by human actors. Overall, the fuzzy logic-based transparent solutions are preferred over black-box solutions to ease the implementation with improved robustness and enhance their suitability for auditing process, even sacrificing the predictive performance indices.

Chapter Contents:

• 11.1 Introduction
• 11.2 System studied using wide area monitoring
• 11.3 Wide-area severity indices
• 11.3.1 WASI features
• 11.3.2 Wide-area severity indices and stability condition
• 11.4 Data-mining models
• 11.4.1 Decision tree
• 11.4.2 DT-induced fuzzy approach
• 11.4.3 Ensemble decision trees
• 11.5 Data-mining model-based catastrophe predictors
• 11.5.1 Scenarios and data count generation
• 11.5.2 Data-mining models for catastrophe predictor
• 11.5.3 Performance assessment
• 11.5.3.1 Decision tree
• 11.5.3.2 Ensemble decision trees (random forests)
• 11.5.3.3 DT_Fuzzy
• 11.5.4 Accuracy vs transparency trade off
• 11.6 Conclusion
• References

Inspec keywords: phasor measurement; power system security; data mining; fuzzy logic; power engineering computing

Other keywords: semitransparent data-mining model; SVM; black-box model; fuzzy logic-based transparent model; PMU-based wide-area security assessment; RF; PMU-based WASI feature; phasor measurement unit; data-mining-based catastrophe predictor

Subjects: Power engineering computing; Power system protection; Power system measurement and metering; Knowledge engineering techniques