Neural network for zero-crossing detection of distorted line voltages in weak AC-systems

In thyristor power converters, zero-crossings of the line voltage signal are used for the synchronization of thyristor gating pulses. In weak ac-systems, however, the line voltage can be distorted, and faulty zero-crossings can occur. Besides, in isolated power transmission networks, the line frequency can alter. For the detection of the true zero-crossings in such cases, we describe a neural network which is capable of tracking the true zero-crossing instants by utilizing the measurements of the three line voltage components in a three-phase power delivery system. The line voltages are measured with comparators, thus enabling low cost implementation. The network structure is extended by using a logic circuit which produces the time elapsed from the previous detected zero-crossing instant as a feedback signal for the network. The logic circuit is thus placed inside the network structure. Therefore, we can utilize the knowledge that the true zero-crossings occur at regular intervals in practical power delivery systems. The simulation results show that the proposed neural network provides competitive performance