Transient Stability Assessment Considering Number and Location of PMUs Using CNN-LSTM

Electric power systems in the future will increasingly adopt wide area monitoring systems (WAMS) based on phasor measurement units (PMUs). Conventional methods are not able to utilize the data effectively and efficiently. This research focuses on utilizing PMU data measurement for transient stability detection using convolutional neural network and long short-term memory (CNN-LSTM) by considering the number and location of PMUs. This research aims to detect stable and unstable transient stability based on bus voltage magnitude and angle data. The CNN-LSTM architecture consists of several layers, including the time-distributed layer, two-dimensional convolution layer, batch normalization layer, dropout layer, max-pooling layer, flatten layer, LSTM layer, and dense layer. The case study used in this research is a modified IEEE 39 bus with a PV system. The proposed method produces an accuracy above 99.5% in normal and distorted data quality for all test scenarios. In addition, the results of this study show a trend that the more PMUs used, the better the detection performance, and PMU locations that pay attention to observability and dynamic stability have better detection performance.

I. Introduction

Electric power systems in the future will increasingly adopt wide area monitoring systems (WAMS) based on phasor measurement units (PMUs). It is based on real-time electric power system monitoring to increase system reliability and security. The use of WAMS in the power system will impact improving the system’s capability to detect power system disturbances. It is possible because the data generated by WAMS is real-time and synchronized between the buses installed by the PMU. Real-time monitoring of electric power systems will produce large amounts of data, so a method or technology is needed to utilize this data effectively and efficiently. Conventional methods are not able to process the data. One method that can be used is the deep learning method. Deep learning methods can be used to study historical data generated by the PMU, which is then applied to predict certain events in the power system.