Fast prediction and avoidance of cascading line failures based on ANN with feedbacks

Predictions and avoidance of cascading branch failures are of great importance in power system operations and have long been research topics drawing much attention. Power flows will redistribute each time when the grid topology changes in the cascading branch failure process. In order to obtain the exact power flows on all branches, power flow equations with different grid topologies must be solved frequently and heavy computational burden is thus introduced. Therefore, an efficient prediction and avoidance method can only be established when power flows can be quickly obtained under all possible topologies in the cascading process without solving the complicated ac power flow equations. To this end, an artificial neural network (ANN) based method is proposed in this paper to approximate the branch flows. A salient feature of the proposed method is that apart from the nodal power injections, the branch states (0/1 variable) are also the inputs of the ANN to consider the topology changes. The power flows are the outputs of the ANN and the corresponding overload states of all branches are then fed back to the ANN as inputs (the branch states). An iterative method is thus established. Case studies are performed and the results are encouraging.