Accurate Fault Location of Two-Terminal Transmission Line based on One End Voltage Measurement and Smooth Support Vector Machines Eyada A J Alanzi, Mahmoud A A Younis ELECTRICAL POWER ENGINEERING COLLEGE OF ENGINEERING UNIVERSITI TENAGA NASIONAL 43009 KAJANG, SELANGOR, MALAYSIA Abstract This paper presents a new technique for accurate fault location based on voltage measurement from one end of the two-terminal transmission line and Smooth Support Vector Machine (SSVM). Due to common problems of current transformer during fault location and as a result increasing the cost and reduction of the accuracy, proposed technique is independent of current measurement and based on one terminal voltage measurement of the transmission line. Post-fault voltage at one end of the line is measured and used in calculation of the fault location. GPS (global positioning system) is not required for this technique resulting in a reduction of economic cost. Using the proposed technique, fault location can be estimated with a lower than 0.025% error without using current transformers and GPS. EMTP/ATP simulation and SSVM results show that the proposed fault location technique is independent of fault type, fault resistance and fault inception of the transmission line. Key words Fault location, Smooth Support Vector Machines, ATP-EMTP 1. Introduction Transmission line faults detection is a highly important task for network maintenance engineers where power recovery time and cost are essential. Most of previous studies use synchronized or unsynchronized voltage and current of faulted system to develop the algorithm for fault location to increase the speed and accuracy and decrease the economical cost of fault locator. These studies can be listed as follows with the method used to develop the required algorithm. Damir et al. [1] use unsynchronized voltage and current from both ends and a conventional method to develop their algorithm. Salat [2], El-Sayed [3] and Coury et al. [4] use the voltage and current (per-fault and post- fault) from one end with different approaches. Refernces [2], [3] and [4] use support vector machine and frequency characteristics of voltage and current, radial base function artificial neural network (RBFANN) and modular approach of different neural network modules, respectively, to develop the required algorithm for each method. Razi et al. [5] and Malathi and Marimuthu [6] use current (pre-fault and post-fault) from one end and the approaches of fuzzy logic system (FLS) and wavelet transform and support vector machine, respectively, for each method. Firouzjah [7] and Sukumar [8] use synchronized voltage (pre-fault and post-fault) from both end and Thevenin Model (TM) to calculate fault location of the transmission line. References [7] and [8] in their approach tried to avoid the dependency on current measurement devices (such as current transformers) by using synchronized voltage measurement (pre-fault and post fault) from both ends. This helped them to accomplish more accurate process. In this paper, the technique utilizes only voltage waveform (post-fault) from one end of the transmission line based on the approach of smooth support vector machine (SSVM) algorithm to calculate fault location is presented. This will eliminate the effect of CT behavior during fault period on fault location and classification. In fact, during the fault period, current transformers do not operate properly due to transient state and over voltage of the power system. The problem of CT saturation and inaccurate measurement of current will increase the cost of protection of transmission system. This creates the necessity of finding another approach not depending on current and with optimum accuracy. The synchronized voltage method as in reference [7] is based on GPS (global positioning system) to gather the field data. In our method we use only one terminal post-faulted voltage waveform and eliminate the use of GPS and CT which may become a source of inaccuracy of the fault location system and has an impact on economical cost. The technique presented in this paper is independent of fault type, fault resistance, fault inception angle and loading angle of the transmission line. SSVM has great advantages in formulation of its learning problem, leading to the quadratic optimization (QO) task and the reduction of number of operations in the learning mode and hence is much faster for large datasets [5]. This 2011 IEEE GCC Conference and Exhibition, February 19-22, 2011, Dubai, United Arab Emirates 978-1-61284-117-5/11/$26.00 ©2011 IEEE 485