Nuclear power plants (NPP) are one of the main sources of electricity. The distribution of electricity generated by nuclear power plants, or backup power supply to individual consumers of the power plant's own needs, can be carried out via a 6-35 kV network. These networks operate with an isolated or compensated neutral. This causes the fact that the currents of single-phase ground faults (SPGF) in case of accidents on overhead power lines (OL) are quite small and SPGF are difficult to identify. The article deals with the issues of identifying the SPGF and the corresponding selective response of the overhead line protection system. The limitation of the functioning of relay protection against SPGF, built on the classical principle of operation "if-then", is proved. The error of the traditional protection system is explained by the lack of a mechanism for updating the setpoints in accordance with changes in environmental conditions that affect the capacitive component of the conductivity of overhead lines. As an example of similar an influence, the effect of icing of wires, as well as the triboelectric effect in the form of accumulation of a space charge in the air gap around the line, is described. In this regard, a method is described for correcting the protection setpoints due to periodic measurements of the capacitance of the lines by means of their location probing. Thus, it is proposed to improve the traditional current protection against SPGF by giving the system the ability to adapt, but within the framework of its response as an agent with a simple behavior. The construction of a more advanced system of protection against SPGF in the form of an intelligent agent of the electrical network is also considered. The essence of this system is to use an artificial neural network as a subagent processing information. The advantage of a neurocomputer system for protection against SPGF is proved, which forms an integral assessment of the state of power transmission lines and learns to detect SPGF on a digital shadow following the electrical network. It indicates the possibility of classifying the proposed system of protection against SPGF as an intelligent agent due to the ability to adapt, learn and develop.