Fundamental Principles of Smart Protection Device

摘要

The report outlines the general principles of building relay protection devices with their inherently developed intelligence created by deep learning neural networks. It is shown that the functioning of such a smart device in the pace of the emergency process development in the protected electrical system is not constrained by the complexity of solving the problem of training a neural network. The entire computationally complexity procedure for selecting supporting precedents for deep learning of a neural network, developing and playing out a full set of simulation modeling scenarios, and confirming the effectiveness of device training is performed, as before, at the design stage. In contrast to the characteristics of classical relays, which use a limited number of geometric figures (circles, ellipses, polygons, etc.) on a plane to form the tripping characteristics, Smart Protection Devices present their characteristics in the multidimensional space of tracked emergency mode parameters. An important feature of such devices is the absence of a branched rigid logic computing scheme, and its intelligence has access to characteristics of any complexity because the classification of the current mode of the protected electrical system and the decision making by a new generation relay protection device is based on discriminant functions with developed nonlinear cores. It is demonstrated that, despite the use of discriminant functions that are complex in terms of design principles, the functioning of Smart Protection Devices does not require an avantgarde computing environment.