MONITORING OF NON-STATIONARY SIGNALS WITH MINIMAL DELAY: NEURAL NETWORK IMPLEMENTATION

Abstract

In electrical and power engineering, Fourier transform algorithms are widely used to analyze current and voltage signals. This leads to a time delay in determining the parameters, which is at least the period of the industrial frequency signal. For a number of tasks, it is relevant to determine the parameters of non-stationary signals with minimal delay. The purpose of the study is to show the possibility of the signals parameters monitoring over a time interval in a fraction of the period of industrial frequency based on the microprocessor implementation of neural network algorithms. Materials and methods. The software and hardware are implemented in standard microprocessor equipment based on the simplest neural networks of direct propagation. The experimental verification of the algorithms was carried out in laboratory conditions using the example of monitoring current signals in an asynchronous motor when power is off and on during one period of industrial frequency. Results. It is shown that the proposed approach makes it possible to record the onset of transients and the rate of change in the frequency of signals during a time window of about a millisecond. At the same time, neural networks of different structures can be used simultaneously. The calculation time of a trained neural network corresponds to real-time signal processing. Conclusions. Neural networks are trained using simple analytical formulas and can be implemented in a wide variation range of signal parameters. Since several simple neural networks can be used simultaneously to solve the tasks, the results obtained can complement and refine each other.