Determination of fault location and fault type in transmission lines using transient signals and machine learning algorithms

In this thesis, different machine learning algorithms are used for the fault location estimation and fault classification for the short circuit faults in transmission lines on the base of wave propagation theory. Using harmonic frequency-fault distance relationship on the base of this theory, the fault location is estimated from harmonic frequencies of the current and voltage transients in the spectrum obtained by applying the Fast Fourier Transform. Developed fault locating algorithms are tested by simulations both for the lines without series compensation and for the lines which have different series compensation levels. To determine the fault type, classification features are extracted using the effective values of one-period line currents sampled during the short circuit, the proportions of effective values of these quantities, effective values of modal components obtained by applying modal transformation to the line currents, and the proportions of effective values of these quantities. The fault classification is then carried out by applying various machine learning algorithms to these features. The Alternative Transients Program is used for modeling the transmission lines and to obtain transient signals during the faults, and the algorithms for fault location and fault type detection are developed in the Matlab Environment. Several machine learning algorithms are applied for fault classification using the WEKA software. With the knowledge of the physical configuration of the lines or electrical parameters and the value of source inductance only, it has been shown that the fault location and the fault type can be detected with a reasonable error. In order to prevent the method from being adversely affected by the source inductance value, the regression feature of the Extreme Learning Machine is used for the uncompensated line and the Waveform Relaxation Method is used for the transmission line with series compensation. Finally, the algorithms developed for determination of fault location and fault type using transient signals and Extreme Learning Machine were embedded in the Digital Signal Processor and tested in real time in the laboratory environment.