Abstract:Power systems transport an increasing amount of electricity, and in the future, involve more distributed renewables and dynamic interactions of the equipment. The system response to disturbances must be secure and predictable to avoid power blackouts. The system response can be simulated in the time domain. However, this dynamic security assessment (DSA) is not computationally tractable in real-time. Particularly promising is to train decision trees (DTs) from machine learning as interpretable classifiers to predict whether the system-wide responses to disturbances are secure. In most research, selecting the best DT model focuses on predictive accuracy. However, it is insufficient to focus solely on predictive accuracy. Missed alarms and false alarms have drastically different costs, and as security assessment is a critical task, interpretability is crucial for operators. In this work, the multiple objectives of interpretability, varying costs, and accuracies are considered for DT model selection. We propose a rigorous workflow to select the best classifier. In addition, we present two graphical approaches for visual inspection to illustrate the selection sensitivity to probability and impacts of disturbances. We propose cost curves to inspect selection combining all three objectives for the first time. Case studies on the IEEE 68 bus system and the French system show that the proposed approach allows for better DT-selections, with an 80% increase in interpretability, 5% reduction in expected operating cost, while making almost zero accuracy compromises. The proposed approach scales well with larger systems and can be used for models beyond DTs. Hence, this work provides insights into criteria for model selection in a promising application for methods from artificial intelligence (AI).

Alternate Support Vector Machine Decision Trees for Power Systems Rule Extractions

Selecting Decision Trees for Power System Security Assessment

Sparse Oblique Decision Tree for Power System Security Rules Extraction and Embedding

Rule Extraction in Transient Stability Study Using Linear Decision Trees

Real-time Emergency Control Decision in Power System Based on Support Vector Machines

Data-Driven Security and Stability Rule in High Renewable Penetrated Power System Operation

Machine Learning-Based Method for Power System Operation Rules Generation

A Decision Support System Using Two-Level Classifier for Smart Grid

Clustering Based on Support Vector Regression Model and Its Application in Power System Transient Stability Prediction

Voltage Stability Constrained Operation Optimization: an Ensemble Sparse Oblique Regression Tree Method

Rule Extraction based Data Augmentation Method for Transient Instability Identification of Power Systems Using Machine Learning

Multi-support Vector Machine Power System Transient Stability Assessment Based on Relief Algorithm

Sample-Derived Disjunctive Rules for Secure Power System Operation

Ensemble Methods of Classification for Power Systems Security Assessment

Artificial Intelligence-Based Methods for Renewable Power System Operation

Data-driven approaches for power system operation and planning under high renewable energy penetration

Mapping Rule Estimation for Power Flow Analysis in Distribution Grids

Enhancing short-term wind power forecasting accuracy for reliable and safe integration into power systems: A gray relational analysis and optimized support vector regression machine approach

Optimization over decision trees: a case study for the design of stable direct-current electricity networks

Real-time Transient Stability Assessment in Power System Based on Improved SVM

A Novel Stability Classifier Based on Reformed Support Vector Machines for Online Stability Assessment