1. Prediction of Out-of-Step Condition for Synchronous Generators Using Decision Tree Based on the Dynamic Data by WAMS/PMU
- Author
-
Sasan Azad, Morteza Nazari-Heris, Somayeh Asadi, Mohammad Reza Aghamohammadi, and Morteza Abedi
- Subjects
Electric power system ,Generator (computer programming) ,Control theory ,Computer science ,Dynamic data ,Stability (learning theory) ,Phasor ,Permanent magnet synchronous generator ,AC power ,Fault (power engineering) - Abstract
Power system security is the power system’s capability to maintain its stability when a disturbance occurs. Transient instability is a phenomenon that threatens the security of the power system and synchronous generators. When rotor angle instability occurs in a synchronous generator, it is capable of driving the generator O.S. If the O.S operation of a generator sustains even for a brief period of time, it may result to serious mechanical and thermal damages to the machine. Therefore, faster identification of such conditions and in time isolation of the unstable generator from the power system is necessary for keeping the power system safe. Conventional methods for the identification of the generator O.S is based on the use of impedance relays. The purpose of this chapter is the prediction of the generator O.S with dynamic data of generator and power system, which can be gathered by PMUs in the fault and clearance period. Generator dynamic information includes variations of voltage phasor, active and reactive power, rotor angle, and speed of generator as well as the data of voltage and line flows in the adjacency of the generator that are collected in different time intervals. The measured information is continually updated in the moving window and with a special algorithm such as a three-step DT that performs the detection fault occurrence and clearance and prediction of O.S of generators. The proposed algorithm is working constantly as a relay, and the proposed algorithm recognizes the stability of the generator whenever changes occur in the generator parameter according to the severity of variations. Since training data perform an effective role in DT learning or other machine learning methods, to increase the learning ability of DT, training samples are prepared for having complete coverage of almost all load levels and different operation points. The proposed scheme is implemented on the generator of the IEEE 39-bus system with promising results.
- Published
- 2021