An Improved Dro-Based Recurrent Neural Networks for Large-Scale Light Curve Time Series Prediction

In light curves, the brightness of stars is associated with time, and it is an image of the brightness with respect to time. The traditional data processing methods cannot effectively handle real-time and large-volume data of various light curves. To address this issue, this work develops a deep neural network, named Dropout Recurrent Neural Networks (DRNN). It extracts complicated characteristics of all images captured by Mini Ground-based Wide-Angle Camera array (Mini-GWAC) for point source extraction and cross-certification through Long Short-Term Memory units. Furthermore, this work optimizes the training model by combining a dropout method, which predicts changes of the star brightness in advance. Extensive experiments with Mini-GWAC dataset demonstrate that DRNN outperforms several typical baseline methods with respective to forecasting performance of star brightness in large-scale astronomical light curves.