An Online Power System Stability Monitoring System Using Convolutional Neural Networks

A continuous Online Monitoring System (OMS) for power system stability based on Phasor Measurements (PMU measurements) at all the generator buses is proposed in this paper. Unlike the state-of-the-art methods, the proposed OMS does not require information about fault clearance. This paper proposes a convolutional neural network, whose input is the heatmap representation of the measurements, for instability prediction. Through extensive simulations on standard IEEE 118-bus and IEEE 145-bus systems, the effectiveness of the proposed OMS is demonstrated under varying loading conditions, fault scenarios, topology changes, and generator parameter variations. Two different methods are also proposed to identify the set of critical generators that are most impacted in the unstable cases.

I. Introduction

Monitoring the rotor angle stability and early recognition of the potentially dangerous conditions is very crucial for reliable operation of power systems. Various techniques have been traditionally used to assess the rotor angle stability. Time domain simulations (TDS) rely on solving nonlinear differential algebraic equations (DAE) that model power systems [1]. But they are computationally intensive and require accurate system data. Transient-energy-function (TEF) methods [2], compare the potential and kinetic energy of the system against a reference value. However, there are difficulties in estimating these energies in practical scenarios due to unavailability of some state variable measurements [2], [3]. Equal area criteria (EAC) and extended equal area criterion assess the transient stability based on a single machine connected to infinite bus (SMIB) model approximations [4], [5]. But they allow only the classical generator models. The SIME (SIngle Machine Equivalent) Method [6], is a hybrid approach which combines the advantages of TDS and EEAC and allows use of detailed models. This method is computationally more efficient than TDS but at the cost of reduced accuracy.