MACHINE LEARNING-BASED FRAMEWORK FOR REMEDIAL CONTROL ACTION PREDICTION USING WIDE-AREA MEASUREMENTS IN INTERCONNECTED POWER SYSTEMS
Growing demand for power systems, economic, and environmental issues, lead to power systems operating close to their stability margin. Power systems are always exposed to disturbances, leading to either instability or cascading outages and blackouts in the worst cases. Although numerous methods have been proposed since 1920 to prevent disturbances, instability and blackout still exist. Among all the instabilities, the fastest occurring one is rotor angle instability or transient instability. Since this instability happens in a fraction of a second, time must be considered in designing remedial control actions (RCAs). Different types of remedial control actions have been proposed in the past, but due to the lack of time consideration in their design, they are not practical for those cases quickly lead to transient instability. Additionally, pre-planned remedial control actions have been employed to overcome time limitations, but they are not able to cover most of the possible scenarios that may occur in the power system. Based on the literature done for this research, predicting remedial control actions has not been implemented yet. This study presents an innovative idea to predict remedial control action schemes that are able to include time limitations and cover possible scenarios properly. There are numerous challenges to consider in performing such a method, such as remedial control actions selection, implementation, practical aspects, and wide-area measurement systems (WAMS). In this study, the different parts of the framework are discussed in detail and implemented. Based on the above discussion, first, an optimized artificial neural network (ANN) is implemented to make a comprehensive framework that can predict a proper remedial control action to prevent cascading outages and blackouts. The different steps of the framework are predicted using this comprehensive algorithm. A micro model strategy has been employed, which builds a model for each line separately. This micro model decreases prediction complexity and increases the prediction accuracies of the modules. The common RCAs, including controlled islanding, load shedding, and generator rejection, are implemented in this research project. To address controlled islanding prediction, in the first step, using voltage data, the stability status was predicted. In the second step, a new method to identify coherent groups of generators was developed, and based on that method; the coherency patterns have been predicted. In the third step, a combination of islanding and load shedding is selected as a control action, and a mixed-integer linear programming (MILP) method is designed to compute islands, the amount of load shedding, and load buses. Since the load shedding prediction has two aspects and it is a very challenging problem, a new concept called the specific set of loads (SSLs) had been proposed to simplify this issue. Finally, the islanding and load shedding patterns are predicted. The framework was tested via the IEEE 39 bus system and 74-bus Nordic power system, and the results show the effectiveness of the framework. To implement generator rejection prediction, the bus voltage data are used to predict the stability status. Next, the critical generators are predicted. Then, using the equal area criterion, the amount of generator rejection for each critical generator is calculated, and the patterns are extracted. Finally, the number of generator rejections is predicted using the dataset and designed ANN. The performance of the generator rejection prediction framework is tested via the IEEE 9-bus system and 74-Bus Nordic power network.
Transient stability, Remedial control action, wide-area measurements, power system stability
Doctor of Philosophy (Ph.D.)
Electrical and Computer Engineering