Integrated Monitoring and Fault Warning of Transformer Based on Stacked Auto-Encoder

In recent years, numerous monitoring devices have been installed in power systems to track nonelectrical variables, opening new avenues for transformer safety assessment using multidimensional monitoring data. However, current transformer safety assessments predominantly rely on the single monitoring signal thresholds, resulting to serious fault warnings delay and low reliability of the outcomes. To address this issue, this paper proposes a transformer safety assessment and fault warning method employing Stacked Auto-Encoding (SAE) network. Firstly, the multidimensional monitoring data of the transformer and its associated equipment during normal operation are gathered. Subsequently, a Stacked Auto-Encoding model is constructed utilizing the Restricted Boltzmann Mechanism (RBM), which is trained on the multidimensional monitoring dataset. This model characterizes the nonlinear relationship between the multidimensional monitoring data during normal operation of the transformer. Furthermore, the established model takes real-time multidimensional monitoring data as input and outputs its reconstructed values. Early detection of transformer faults is achieved by assessing the reconstruction error between the model’s output and the input data, and by monitoring the rate of change of the reconstruction error integral using a sliding time window. Finally, The proposed methodology is validated through a case study involving a $110-35 \mathrm{kV}$ transformer at a Shanghai substation. Experimental results underscore the method’s efficacy in enhancing the reliability of transformer safety assessments and significantly improving the timeliness of fault warnings.