Neural Network for Real-Time Signal Processing: the Nonlinear Distortions Filtering

Artificial neural networks, after their training and testing, allow, in the "if then" mode, to process signals in real time. This is relevant for signal processing tasks in electrical engineering and the electric power industry, especially for the analysis of nonlinear signal distortions, transients in electrical networks, during switching, emergency modes, and so on. The paper shows the neuro algorithm possibilities based on an elementary perceptron for filtering nonlinear distortions of industrial frequency signals of 50 Hz over a time interval of units of milliseconds. At the same time, the accuracy for the signals’ amplitude, phase, and frequency determining is units of percent. Examples of the fundamental frequency signal selection in the presence of harmonics, the determination of the signal parameters during transients, and the correction of the transformer saturation current are given. It is shown that real-time neural network processing can be carried out in a "sliding window" with the duration of units of milliseconds. The estimates made during the implementation of the neuro algorithm in a microprocessor device showed its application possibility in the electric power industry secondary equipment.