Distributed PV Power Anomaly Detection Based on LSTM-Informer

Distributed photovoltaic power plants cover a wide range of areas, and due to the unpredictability of meteorological factors and the diverse and complex installation environments, interference factors are difficult to eliminate completely. Under such a background, it is particularly difficult to accurately determine the power abnormality of distributed PV equipment. Therefore, how to accurately identify the power abnormality of distributed PV systems with the help of limited metering data has become an urgent problem for the industry. In this paper, a distributed PV power anomaly detection model based on LSTM-Informer is proposed, which is able to identify the three weather conditions of sunny, cloudy and rainy days by calculating the kurtosis of the power curve according to the characteristics of the PV power curve of different weather types, and use the Monte Carlo method based on LSTM-Informer to train the normal output interval, and determine whether the distributed PV system is operating abnormally by calculating the coverage of the time-series power data to be detected in the normal output interval. In this paper, simulation experiments are carried out using PV data from a power station in Australia, and the results show that the model can effectively detect power anomalies in distributed PV systems, with high detection rate, accuracy rate and low false alarm rate.