Deep Learning Based Fast Security Correction of Power System with Unified Power Flow Controller

With the gradual popularization and application of new generation of flexible AC transmission devices such as unified power flow controller (UPFC) in modern power system, higher requirements are requested for the formulation of power grid security correction strategy. The traditional power grid security correction method based on physical model has some limitations in the real-time field. The data-driven method moves a large number of complex calculations forward to the offline stage, so it has fast online computing performance. Therefore, a fast security correction method based on deep learning is proposed for power system with UPFC. Firstly, based on deep learning, a recognition model of node adjustment state is established. The ability of classification and learning of deep neural network(DNN) is used. It gives priority to determining the nodes with adjustment possibility, which avoids the iterative non-solution problem of physical model optimization methods. Then, aiming at the reduced optimization space, the optimization method is further used to realize the system security correction calculation, and the adjustment amount of each node in the system is quickly determined. Based on the application results of the UPFC demonstration project of west ring network in Nanjing, China, the proposed fast security correction strategy can facilitate the learning ability of DNN and improve the efficiency and practicability of the system security correction. © 2020 Automation of Electric Power Systems Press.