Identification of Abnormal Incremental Data on Power Demand Side Based on Long Short-Term Memory-Transformer Hybrid Model

When the abnormal alarm threshold of the power demand side is high, it is difficult to distinguish the incremental data only by relying on the abnormal characteristics of the historical data, resulting in low recognition accuracy of abnormal incremental data on the power demand side. To this end, a method for identifying abnormal incremental data on the power demand side based on the LSTM-Transformer hybrid model is designed. The Long Short-Term Memory (LSTM) model is used to capture the long-term dependency of incremental data and extract time series features. The LSTM-Transformer hybrid model is used to output the abnormal probability of abnormal incremental data on the power demand side through the fully connected layer. The experimental results show that the LSTM-Transformer hybrid model achieves the best performance in the abnormal recognition redundancy (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$&lt;10 \%$</tex>), Receiver Operating Characteristic (ROC) curve performance (Area Under the Curve (AUC) <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\approx 0.96$</tex>) and multi-category abnormal recognition confusion matrix in the task of identifying abnormal incremental data on the power demand side, verifying its practical value in the scenario of high precision and low false alarm demand.