An online monitoring method of current transformer based on the fusion of multiple deep learning models

For the accurate evaluation of the operational performance of current transformers, this paper proposes a performance assessment method based on a hybrid deep learning network. Firstly, a pre-trained Sparse Auto Encoder (SAE) model is utilized to conduct preliminary feature extraction from current transformer monitoring data, followed by smoothing the model output using an exponential moving average. Secondly, a Multi Scale Dynamic Convolution (MSDC) network with kernels of varying sizes is employed to extract multi-scale temporal features from the time series data. Then, these multi-scale temporal features are fed into a Long Short-Term Memory (LSTM) network to capture long-term relationships within the time series. Finally, the performance of the current transformer error prediction model is evaluated using the mean squared error as a metric. Experimental results demonstrate that the hybrid model outperforms single models in terms of the prediction accuracy of ratio and phase angle errors, showcasing the effectiveness of the SAE-MSDC-LSTM integrated deep learning approach for error prediction in current transformers.