Time Series Prediction and Anomaly Detection of Light Curve Using LSTM Neural Network

Abstract

Ground-based Wide-angle Camera array (GWAC) is a short-time survey telescope which can produce images every 15 seconds for more than 30,000 stars. Light curve is generated from star image with a series of processing. Research on light curve is a new task in time domain astronomy which can detect anomaly astronomical events. We explore a series prediction model of LSTM neural network for light curve prediction. Through model training and validation we obtain the optimal structure. Then we predict one time-step ahead light luminance for test star. We evaluate the performance of model by calculating prediction error. Anomaly detection mechanism is based on prediction error. Experimental results based on real light curve data demonstrates that our model is promising in light curve prediction and anomaly detection.