Selecting decision trees for power system security assessment

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