Neural network technique for the joint time-frequency analysis of power system quantities

The joint time-frequency analysis of power system quantities is a powerful approach to monitor the distorting effects produced by nonlinear, time-variant loads. Essential in shorting the time taken by the analysis is the knowledge of the power system frequency. In this paper a technique based on neural network (NN) is proposed to estimate such a frequency in real time. The frequency is represented by a NN weight, adjusted online through a suitable learning process or the line voltage. The same is done for the magnitude of the harmonic content since this makes the frequency estimation accurate. The performance or the proposed technique is described in terms of dynamic behavior and steady-state accuracy. In particular, it is found that a change in the power system frequency is tracked in much less than a line period