Anomaly Detection Using Long Short-Term Memory

Abstract

In this digital era, data is a valuable asset, but there can also be anomalies or errors in the collected data. Therefore identification of abnormal or error-prone information is always a tedious work to improve the efficiency & reliability of smart systems. The main target of anomaly detection is to find a change in the behaviour of a system or predict future failures, which may occur due to error-prone data before they happen. Anomaly detection using Long Short Term Memory can effectively reduce the forecasting and prediction errors. A novel anomaly detection & power consumption prediction approach using LSTM neural network is proposed to enhance the performance of a smart electric grid. When the proposed LSTM model is compared with Autoregressive Integrated Moving Average algorithm (ARIMA), a decrement of 20 % is obtained in the forecasting error. Data of consumer from Indore Zone, MP (India) for power consumption is used to perform simulation experiments. Electricity theft identification via the smart grid has improved when comparing with the existing models.