Abstract: Situation awareness based on power grid operation indicators is the trend of future dispatching mode. Identification of anomalies is an important content of situation awareness. A comprehensive indicator system that comprehensively reflects the power grid operation is constructed. An auto-encoder composed of LSTM is used to construct the index abnormality identification model. In the absence of abnormal data on power grid operation indicators， an unsupervised approach is adopted to learn the internal model of the indicators from the historical data of the indicators under normal operating conditions of the power grid. On the basis of model reconstruction error distribution， this paper proposes an abnormal score reflecting the deviation of the index from the normal state. The real-time data of the grid operation indicators are sent to the trained model for reconstruction， and a large abnormal score will be generated when there is an abnormality. The experimental results show that the model can effectively identify the abnormal scores when the grid operation indicators are abnormal. This helps grid dispatchers perceive grid operation risks in a timely manner and take timely control measures to ensure grid security.

Key words: operating indicators, anomaly identification, sequence to sequence（Seq2Seq）, long short term memory（LSTM）, power grid operation