A General Method for Power Equipment State Classification Using Time Series Data Based on Attention Mechanism

With refinement of power equipment, the identification and classification of its operation states is quite important for the stable operation of the system. As the power equipment states are rather complicated, it is difficult to realize general classification due to the limitation of data types and model characteristics when using deep learning. In the light of this, based on the time series data of various power equipment, an equipment state classification method is proposed by combining Recurrent Neural Network (RNN), Long short-term memory (LSTM) and attention mechanism. Firstly, RNN is used as the bottom frame network, and the LSTM core network integrating attention mechanism is added to construct the complex Feature Extraction Module (FEM). Secondly, main feature extraction is implemented in the LSTM network integrated with attention mechanism. Then, the extracted features are input into the complex FEM model and trained to match the evaluation state and obtain the evaluation results. In this paper, confusion matrix, F
1micro
and accuracy are used as evaluation indexes. Three cases were selected to verify the method and five groups of comparison models were set for each case. Results show that the proposed method has good accuracy in power equipment state classification and also has strong generalization ability in other data classification.