Interpretable data‐driven contingency classification for real‐time corrective security‐constrained economic dispatch

High penetrations of renewable energy are crucial for low‐carbon power systems. However, the higher volatility of renewable power generation pushes real‐time operations closer to equipment limits. It is thus important to utilize flexibilities in the system through corrective security‐constrained economic dispatch (SCED) that allows generators to take corrective adjustments after contingencies. The corrective SCED problem, containing a large number of contingencies, and corresponding post‐contingency decisions and constraints, is very large in scale and difficult to solve using purely model‐based methods within the strict time limits of real‐time markets. To accelerate the solution process, this paper develops a novel interpretable data‐driven contingency classification method. Historical data and their potentially useful patterns are utilized in interpretable data‐driven decision tree classifiers. To directly consider continuous features, such as net load values, and to consider imbalanced datasets without much additional complexity, Improved Strong Optimal Classification Trees (ISOCTs) are developed with new branching threshold constraints and category weights in the objective function. ISOCTs are then embedded into a hybrid model‐based and data‐driven framework to guarantee the accuracy of the real‐time active contingency set and the resulting security of dispatch decisions. Numerical testing results demonstrate the classification accuracy, computational efficiency, and interpretability of the proposed approach. To accelerate the solution of real‐time corrective security‐constrained economic dispatch while ensuring accuracy and interpretability, this paper develops a novel data‐driven contingency classification method with Improved Strong Optimal Classification Trees. They are then embedded into a new hybrid model‐based and data‐driven framework to guarantee the security of dispatch decisions.