TRANSFORMER CONDITION PREDICTION BASED ON DATA AND PHYSICAL MODELS

. Deploying different kinds of driving models in the routine monitoring and maintenance of power transformers is an effective way to improve transformer safety. However, it poses a challenge to the accuracy of the models as well as the interpret ability of the outputs. To address this issue, we first simplified the internal temperature field of the transformer and constructed a physical driver model of the transformer incorporating its internal heat transfer process. Then, we constructed a data-driven model for transformer condition prediction by using the transformer daily operation data and mining the change information in the data based on the metabolic gray algorithm (MGA). Finally, we fused these two models, specifically updating the parameters in the physical model through the data-driven model to achieve the purpose of accurately predicting the transformer condition. The experimental results provide valuable design insights for the actual power transformer monitoring system.