Abstract
In this study, a different signal recognition approximation is presented to determine applied voltage value using sound records of the electrical discharges (coronas) by a probabilistic neural network. Sound records are obtained experimentally from the electrical discharges at different 50 Hz AC high-voltage levels. Parts of the recording time on the recorded sound has been used to training and test sets of the probabilistic neural network. One of the goals of this work is to determine voltage value from the sound data, and other is optimization of data and diagnostic for less data used and to find correct voltage value. In the algorithmical method, linear prediction coefficients of the different degrees are used. It is shown that the results can be accepted for the work goals