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).

Prediction Of Period-Doubling Bifurcation Based On Dynamic Recognition And Its Application To Power Systems

System Resilience Enhancement Against Cascading Failure in Consideration of Uncertainty

Temporal Difference Learning Based Critical Component Identifying Method with Cascading Failure Data in Power Systems

Selecting Decision Trees for Power System Security Assessment

On-line control of Hopf bifurcations in power systems

WAMS-based Monitoring and Control of Hopf Bifurcations in Multi-Machine Power Systems

Predicting discrete-time bifurcations with deep learning

A Frequency Domain Approach to Predict Power System Transients

On-line Hopf Bifurcation Control in Multi-machine Power Systems

Analysis of Subsynchronous Resonance in Power System Based on Bifurcation Theory

Frequency prediction of a post-disturbance power system using a hybrid ARIMA and DBN model

Inferring Attracting Basins of Power System with Machine Learning

Analytic Interval Prediction of Power System Dynamic under Interval Uncertainty

Modeling of Long-term Dynamics for Interconnected Power Systems with Structure-preserved Network

Research on Power Systems Voltage Stability Assessment and Prediction Based on Critical Transition

TWO-PARAMETER BIFURCATION ANALYSIS ON A TYPICAL POWER SYSTEM MODEL

A Unified Approach for Learning the Dynamics of Power System Generators and Inverter-based Resources

Dynamical Behavior and Singularities of a Single-machine Infinite-bus Power System

Application of higher order dynamic mode decomposition to modal analysis and prediction of power systems with renewable sources of energy

Machine Learning Based on Bayes Networks to Predict the Cascading Failure Propagation

Data-driven bifurcation analysis using parameter-dependent trajectories