Optimization of Power Transformer Defect Recognition Algorithm with Dga Based on Probabilistic Neural Network

The trouble-free operation of power oil-filled transformer equipment is impossible without an effective diagnostic system. It is generally recognized that the chromatographic analysis of gases (DGA) dissolved in oil is the most informative method for diagnosing the condition of oil-filled transformers. Nowadays, it is increasingly being implemented as continuous online monitoring. In this case, new methods of processing and displaying the flow of diagnostic parameters and analyzing the ratios of key gases, with the interpretation of their palette and accuracy in making a "diagnosis" to prevent emergency conditions will be in demand. A method of forming an adaptive diagnostic model for TO based on online monitoring data with a cluster database for IESD is proposed. Organization of offline and online monitoring subsystems, scaling of object models and server process based on PostgreSQL are described. The advantages of probabilistic neural network (PNN) for non-normal distribution of diagnostic results are highlighted. The Moth Flame algorithm for optimizing weights within the model is described.