Efficient Optimal Power Flow Flexibility Assessment: A Machine Learning Approach

We propose a framework based on machine learning to assess the flexibility of power systems with optimal power flow (OPF) model. The definition of flexibility is a range within which all demands are feasible. Conventional methods to evaluate the flexibility by solving a robust optimization problem are time-consuming for large-scale power systems. Machine learning provides us the opportunity to accelerate the computing process. We formulate the problem as a nonlinear binary classification problem and use a support vector machine (SVM) classifier with a Gaussian RBF kernel. To compute the flexibility, we solve a simple nonlinear equation based on the trained classification boundary. Then, we employ active learning to enhance the SVM's precision and adaptability. The simulation results for the five IEEE test cases indicate that our framework can compute the flexibility with a low error rate and significantly less execution time than the benchmark method for large-scale power systems.