Artificial intelligence model for gravitational wave search based on the waveform envelope

Abstract

In recent years, artificial intelligence technology for gravitational wave data analysis has developed rapidly. In this paper, we put forward a new artificial intelligence model for gravitational wave search. The framework of such a model includes a detection stage and a testing stage. We first use the deep learning technology to extract the envelope information of the gravitational wave candidate and use the coalescence time obtained from the envelope to further confirm the detection results. Within the detection stage, we use wavelet denoising and a special training strategy to improve the performance of the gravitational wave detection model. The lowest false alarm rate of the detection model is about 1.7 per month without the testing stage. When the testing stage is added, the lowest false alarm rate decreases to 0.046 per year. The efficiency of our model is demonstrated by the data obtained from the first, second, and third observing runs of the LIGO-VIRGO-KAGRA collaboration. The search results of confident events on the three observing runs indicate that the efficiency of our deep learning algorithm can achieve 80% of the traditional algorithm based on matched filtering.