<p>The study of the possibility of using artificial intelligence in the diagnostic of old equipment opens up new opportunities for preventing emergency situations and optimizing the maintenance process. One of the key obstacles to the successful application of artificial intelligence algorithms is a problem of little available data, namely, data characterizing emergency or pre-emergency conditions. Moreover, even if one collects large data using sensors, the problem of class imbalance is still present. In this study, adapted the method of One-Shot training of graph neural networks to the problem of finding defects in electrical equipment. The ontological knowledge base of defective states of electrical equipment will serve as a source of information for training and forecasting. The formation of an ontological knowledge base about defective states of electrical equipment plays a key role in the development of new scientific topics. It allows you to systematize and disseminate knowledge about defects in electrical equipment. The study considers link prediction tasks (recovering missing triple links) and expects that many missing pieces of information about possible equipment defects are in the graph encoded through the neighborhood structure. The work was carried out with a limited amount of initial labelled data, namely, 27 samples describing the technological state of transformers using 86 parameters were used for model training.</p>