2011 International Conference on Electrical Engineering and Informatics 17-19 July 2011, Bandung, Indonesia A Novel Power Swing Detection Algorithm Using Adaptive Neuro Fuzzy Technique Ahad Esmaeilian, Sajjad Astinfeshan School of Electrical and Computer Engineering, University of Tehran, Iran a.esmaeilian@ece.ut.ac.ir Abstract— Any sudden change in the configuration or the loading of an electrical network causes power swing between the load concentrations of the network. Power swing can affect the distance relay performance by entering the impedance locus into protection zone of the relay causing unnecessary tripping. In this paper a novel technique to prevent the distance relay from tripping during power swing is presented. The presented method is based on adaptive neuro-fuzzy inference system (ANFIS) which has three inputs, include the rate of change of positive sequence current, active and reactive powers. Adaptive neuro fuzzy method is implemented utilizing MATLAB/ANFIS toolbox. For the purpose of testing a typical power network which is used in previous works is modelled in PSCAD software, a data set consisting of more than 2500 different power swing scenarios is also considered to guarantee the results are accurate in different conditions. Keywords— Power swing detection, Distance relay, Adaptive Neuro Fuzzy Inference System (ANFIS). I. INTRODUCTION Power systems under steady state conditions operate typically close to their nominal frequency. A balance between generated and consumed active power exists during steady state operating conditions. One of the most important reasons that lead to the power grid blackout is the power swing. Power system faults, line switching, generator disconnection, and the loss or application of large blocks of load result in sudden changes to electrical power, whereas the mechanical power input to generators remains relatively constant. These system disturbances cause oscillations in machine rotor angles and can result in severe power flow swings [1]. During power swings because of variation of rotor angles, the angle between two areas of the power system fluctuates. If the swing is stable, the fluctuations die down [2]. However, unstable swings result in progressive separation of angle between two areas of the power system, causing large swings of power, large fluctuations of voltages and currents and eventual loss of synchronism between such areas [2]. If two areas are in phase, the voltages will become the maximum value and the currents will inversely become minimum, while in the out of phase areas (by 180°), currents and voltages will be at their peak value and close to zero respectively. Since the system frequency is a function of rotor speed, the frequency of voltages and currents during power swings is not constant. The frequency of occurrence of voltage/current maximum depends on the rate of change of the power angle between the two areas and is characterized by a ‘slip’ frequency. The slip frequency can be as low as 1–3 Hz (slow swing) and as high as 4–7 Hz (fast swing) [3]. Many techniques have been introduced to block the trip signal during power swing. In [2], Brahma utilized wavelet transform to identify power swing. However the proposed method is able to detect the symmetrical faults during power swing, its sampling rate is more than 40 kHz which cause the method to be expensive for the purpose of implementation. In [4] author compares some reported methods and concludes that tracking of apparent resistance gives better performance, since it changes during power swings but does not change during the period when a symmetrical fault exists. Ref. [5] applies a simple approach which is based on DC component of current achieved from FFT analysis to detect a fault quickly and reliably during a power swing. However, further testing is needed in larger power systems before the existing technique can be deployed to a distance relay. In [6], a method is based on the extraction of the current waveform components using the Prony method is introduced. While the method is accurate for detection of symmetrical faults from power swing, it is neglected to report the PSB output in the case of high resistance single phase faults. In [7] the authors present a new blocking method during fast swings based on load angle differences identification. While the method is accurate for both slow and fast swings, the authors did not consider the symmetrical fault in their case studies. In this paper, an adaptive neuro fuzzy approach is used to detect power swings and a very accurate PSB scheme is designed. The proposed ANFIS scheme has three inputs, include the rate of change of positive sequence current, active power and reactive power. Since power swings are characterized by slow variation currents, it is very typical to use the rate of change of currents as an input to ANFIS network. On the other hand, the maximum between normalized |dP/dt| and |dQ/dt| is always equal or greater than 0.7 during power swing. However, they will level off to 0 when a three phase fault occurs during power swings. Use of these two inputs is also seemed to be E15 - 5 978-1-4577-0752-0/11/$26.00 ©2011 IEEE