Neural Network Hardware and Software Real Time Control of Electrical Signals Phase Shift

Abstract

Artificial neural networks are increasingly being used in the intelligent electric power industry. Smart grids are the key components of digital economy. The work purpose is to show the possibility of using a simple architecture neural network in the appropriate microprocessor engineering for improvement such microprocessor device characteristics as reducing the device reaction time, increasing the decision-making accuracy in the accident and the ability to more accurate localization of the accident site. This reduces the negative accident consequences, time of determination of the accident location and, accordingly, the time to eliminate the accident consequences, and to restore the normal power system operation. Neural network training is a long-lasting process. At the same time, neural network “deep learning” does not guarantee their default-free operation. So, it is proposed to introduce a pre-trained neural network into intelligent electronic devices when electrical signals can be described by analytical formulas and the possible variation ranges of such signal parameters are set in advance. The corresponding approach has been implemented and tested in a microprocessor device for signal phase shift in transient mode rapid estimation. It is shown that the phase difference estimation can be carried out in a time not exceeding 1 ms., which significantly exceeds conventional algorithms based on the Fourier filter capabilities. The practical application and joint use of the Fourier filter and the artificial neural network possibilities for the creation relay protection device hybrid measuring elements are discussed. The approach and the obtained results can potentially be applied in a wide range of signal processing tasks.