Evaluation of the Capacitance State of Autotransformer Based on TCN and LSTM Network Structures

This study introduces a method to evaluate the capacitance state of transformer windings using TCN and LSTM network structures, to solve the problem of winding deformation caused by factors such as insulation structure, ambient temperature, operating state variations, and internal defects. Considering the variation range of the actual winding capacitance parameters, a simulation calculation model of a three‐phase autotransformer was established to obtain input data of low‐voltage winding voltage imbalance through simulation tests. Two neural network models, including TCN and LSTM, were constructed and trained using the low‐voltage winding unbalanced voltage phase and winding capacitance as input and output data. Finally, the voltage imbalance data of a transformer low‐voltage winding was employed to evaluate the accuracy of winding capacitance state simulation. The results show that the TCN neural network model has higher prediction accuracy and can judge whether the winding deformation occurs by comparing the predicted capacitance value with the historical data, which has reference value for diagnostic analysis of transformer winding deformation.