Prediction of Dissolved Gas Content in Transformer Oil Based on BWO-BiLSTM-Attention Model

This paper proposes a BWO-BiLSTM-Attention method for predicting the dissolved gas content in transformer oil, which can detect potential problems of the transformer in time and improve its reliability and safety. The method consists of three steps: (1) Inputting the historical data of dissolved gas in transformer oil into a bidirectional long short-term memory (BiLSTM) network to capture the temporal features of the data and obtain the hidden state vector of each time step. (2) Applying an attention mechanism to perform weighted summation on the hidden state vectors and obtain a global representation vector that reflects the important features and relationships of the data. (3) Using a beluga whale optimization (BWO) algorithm to optimize six hyperparameters of the BiLSTM-Attention network, including learning rate, epoch, batch size, hidden layer 1, hidden layer 2, and fully connected layer, to enhance the generalization ability of the model. The experimental results demonstrate that the proposed model achieves a fitting rate of 99.77% and a prediction accuracy of 99.41%, which are superior to those of five other models and indicate that it can predict the dissolved gas content in transformer oil accurately.