Prediction of dissolved gas volume fraction in transformer oil based on CVMD and LSTM

In the analysis of dissolved gas monitoring data in transformer oil, a significant challenge arises: the presence of noise often interferes with the data, posing difficulties for conventional deep learning models like Long Short-Term Memory networks (LSTM) to accurately extract profound features, consequently affecting prediction accuracy. To address this issue, we propose an innovative integrated prediction model. Initially, our approach employs Continuous Variational Mode Decomposition (CVMD) technology to eliminate noise signals, thereby facilitating the decomposition of the gas sequence into relatively stationary subsequences, consequently enhancing data quality and stability. Subsequently, we construct a prediction model by integrating a Convolutional Neural Network (CNN) with LSTM to forecast these subsequences. This model is adept at capturing pertinent features within the sequence, thereby enhancing prediction accuracy. Furthermore, we employ SENet to extract deep features, formulate feature vectors, and utilize them in conjunction with the LSTM model for further predictions. This fusion of SENet and LSTM allows the model to gain a more comprehensive understanding of intricate relationships and patterns within the data, consequently augmenting prediction accuracy and stability.