A Probabilistic-based Technique using Transient RMS Currents for Earth Fault Detection in Medium Voltage Distribution Networks

This paper presents a transient probabilistic-based technique for earth-fault detection in isolated and compensated neutral medium voltage networks. In these networks, the magnitudes of the fault currents are very small and the straight comparison of magnitudes may lead to risk of malfunction, especially at high fault resistance. The proposed technique is based on Bayesian theorem for selecting the faulted feeder using the transient RMS fault current for each feeder. The transient RMS value is calculated from the residual currents of the feeders in a sliding window of 2.5 ms. The values of the currents are normalized. Then it is used to apply Bayesian theorem as probabilistic-based selectivity function to indicate the probability of the feeder being faulted (or healthy). The faulted feeder can be detected easily without voltage measurements. The proposed technique is less dependent on the fault resistance and the faulted feeder parameters. The MV network is simulated by EMTP-ATP Draw program. Extensive simulations that cover different fault conditions are performed to validate the technique. The residual value is equal to the summation of the instantaneous phase values (a, b and c). The residual current is very small in normal operation and becomes meaningful in fault condition. It is very sensitive to earth faults and, from a practical point of view, it can be measured easily by one sensor for each feeder; hence it is suggested that it can be used for fault detection. The transient RMS value of the residual current is calculated in the constant sliding fault detection window. The sliding (moving) window is updated after collecting the new samples, after a constant step of sampling time. The smaller value of the sampling time enables coverage of the high frequencies contained in the waveform, which increases the sensitivity of the detection. A sampling frequency of 25 kHz is used. A normalization method is used to limit the current values for all currents to be between 0 and 1 under any condition. This limited band technique is preferred and recommended in probabilistic-based applications. The normalized values of the residual currents of the feeders are regarded as random variables. These are assumed to be independent and identically distributed random variables obeying normal distributions. The Bayesian theorem is utilized as a probabilistic-based selectivity function to indicate the probability of the feeder being faulted. The transient fault probability selectivity function is calculated in the sliding window for all feeders. The threshold limits are investigated for different fault conditions, different inception angles, different fault locations and higher fault resistances, for both isolated and compensated neutral networks. The limits are always 1 for a faulty feeder and 0 for a healthy feeder, which presents a superior selectivity technique over the direct magnitude comparison techniques. The setting value of the critical fault probability is proposed to be at the middle of the isolating gap, which is equal to 0.5.