Purpose. Application of a specially constructed neural network with the task of identifying the parameter substitution scheme in non-invasive monitoring conditions. Methodology. Use of electrical measurement methods by non-invasive monitoring, methods of identification and training of neural networks based on anterior and back propagation error, NARX networks. Findings. The power transformer is an important object of the power system of the electric shop substation. At the same time, frequent transitions from underload to partial overload mode are possible, which creates preconditions for the development of undesirable phenomena in transformers. Monitoring of the power transformer based on control of its substitution scheme, allows to pass to control of its basic parameters irrespective of an operating mode. Non-invasive monitoring works well in the context of the task, because research shows that the parameters of voltage and current, which it operates well reflect the dependence on changes in the parameters of the substitution scheme. Based on T equivalent the substitution scheme can detect and predict changes in parameters substitution schemes according to the parameters of the current and voltage regime flowing in this system. Many neural networks will work out the calculation of electrical and electrotechnical equivalent circuits as a task of identifying the parameters of electrical circuits in static conditions. In the process of identification, the ability of neural networks of different designs to identify one of the substitution scheme parameters, to resist damage was tested, which corrected the value of an unknown parameter. The experiment made it possible to obtain data for comparing the effectiveness of various architectures of neural networks in relation to the real parameters of the equivalent circuit. Concretizing the obtained results, we say that the NARX architecture is able to identify parameters in standard modes for all elements of the substitution circuit, which further opens up opportunities for its improvement in the calculation of nonlinear elements of the transformer when operating in a saturated state. Originality. It is established that the neural boundaries studied in the work are able to determine the parameters of the replacement circuit of a transformer or electric machine in static operating modes, which allows in the future to monitor the state of windings and magnetic circuit according to their values. Practical value. Using a neural network in the monitoring system allows you to get clear values of the equivalent circuit parameters, regardless of the mode, the proposed method significantly reduces the amount of time spent on monitoring the transformer parameters, allows you to control the power level, and, if necessary, reduce the amount of information required for the transformer monitoring.