Full text
| |
| Author(s): |
Orlem Lima dos Santos
Total Authors: 1
|
| Document type: | Master's Dissertation |
| Press: | Campinas, SP. |
| Institution: | Universidade Estadual de Campinas (UNICAMP). Faculdade de Engenharia Elétrica e de Computação |
| Defense date: | 0000-00-00 |
| Examining board members: |
Daniel Dotta;
Fernando José Von Zuben;
Ildemar Cassana Decker
|
| Advisor: | Daniel Dotta |
| Abstract | |
Nowadays, vast amounts of data are collected by Wide Area Measurement Systems (WAMS). Therefore, there is an obvious necessity for Machine Learning (ML) methods, as useful knowledge to extract relevant and reliable information from this synchrophasor data. Among the ML approaches, the Deep Neural Network (DNN) models provide an important opportunity to advance direct learning from the data, making these approaches independent from feature extraction techniques. However, these deep models produce black-box classifiers that can be matter of concern when applying to high-risk environment (critical infrastructure) such as the EPS (Electric Power Systems). In this work, the application of an explainable data-driven method is carried out in order to inspect the performance of DNN classifier for event identification using synchrophasor measurements. The DNN classifier is a Long-Short Term Memory (LSTM) with positive performance in the extraction of dynamic features. The principal benefit of this approach is the use of an interpretability inspection named SHAP (SHapley Additive exPlanation) values, which are based on cooperative game theory (Shapley values). These SHAP values provide the means to evaluate the predictions of the LSTM, highlight the parts of the input time-series with the most contribution to the identification of the events, and detect possible bias. Moreover, by employing the SHAP inspection along with domain knowledge of the problem, the performance and coherence of the LSTM classifier will be improved by choosing the classifier that not only has highest Identification Accuracy Rate (IAR) but is also coherent with domain knowledge of the problem, minimizing detected bias. The application of this interpretable approach is desirable because: i) it explains how the LSTM classifier is making its decisions; ii) it helps the designer to improve the training of the classifier; iii) it certifies that the resulting classifier has a consistent and coherent performance according to domain knowledge of the problem; iv) it clearly reduces the concerns of the application of DNN methods in a critical infrastructure, in the cases that the user understands that the classifier is taking coherent decisions. The proposed method has been evaluated using real synchrophasor event records from the Brazilian Interconnected Power System (BIPS) (AU) | |
| FAPESP's process: | 17/25425-5 - Analysis of artificial neural networks methodologies for event classification using Synchrophasors |
| Grantee: | Orlem Lima dos Santos |
| Support Opportunities: | Scholarships in Brazil - Master |