Data Analytics of PMU Measurement Features for Real-time Short-term Voltage Stability Prediction

Stability problems have been taking place in recent years in power systems due to different factors without the possibility of network operators to anticipate its occurrence and impacting the continuity of the electric power service. However, the current paradigm is different as a result of new technologies that allow monitoring the dynamics based on PMU equipment and the prediction of problems in very short times with data mining. This paper presents a novel method for predicting short-term voltage stability problems in real-time through data mining and analytics techniques. These techniques are used in the proposed method to i) select the measurement features that are required to predict the post-contingency operation status by solving a multiobjective optimization problem, ii) perform pattern extraction based on symbols and iii) train an intelligent classifier to predict the state of post-contingency operation. Case studies are presented in the New England 39-bus test system in which it was obtained that the installation of only 5 PMU equipment is required to predict the post-contingency operation status with an error less than 4% and using a post-disturbance data window equal to 180 ms, this time is enough to activate control actions that allow mitigating the problem.