Analyzing dominant oscillatory modes in power system using unidirectional and bidirectional LSTM Techniques

In the realm of power systems, the assessment of dominant oscillatory modes plays a critical role. If operators can swiftly identify and detect low-frequency electromechanical oscillatory modes and their characteristics, they can respond more effectively to specific events. With the progress in wide area monitoring systems (WAMS) and the abundance of measurements, there is a growing need for mode estimation algorithms capable of processing massive amounts of data in a fast and efficient manner. Deep learning algorithms, particularly the Long Short-Term Memory (LSTM), have outperformed other prediction techniques when it comes to forecasting oscillatory modes. This paper adopts a systematic approach to investigate the impact of deep-stacked unidirectional (uni-LSTM) and bidirectional LSTM (bi-LSTM) networks on predicting oscillatory mode parameters in power systems. The simulations are conducted using MATLAB software and the Python Tensorflow library, allowing for a comprehensive evaluation of the performance of these AI-based algorithms in power system analysis. The results are evaluated using analytical techniques based on the simulation results of the Kundur two-area system and the IEEE 39 bus system. The results confirm the superior viability and adaptability of bidirectional LSTM for the application of power oscillation analysis.