Deep learning for multi-data fusion in gravitational wave search

Deep learning for gravitational wave detection is a research hotspot in recent years. Compared with the traditional matched filter method, the gravitational wave detection method based on deep learning has the advantage of efficiency. However, deep learning-based methods face the problem of low robustness and physical explanability. The future parallel operation of new ground detectors will prompt the research of data analysis methods for multi-detector fusion. Fusing multi-detector data can improve the robustness of in-depth learning for gravitational wave detection. The Hanford and Livingston models are trained using published data from the Advanced Laser Gravitational Wave Observatory and synthetic data from the SEOBNRv4 method, and four fusion methods from Hanford and Livingston are studied. The test results of the test set show that the arithmetic average fusion can achieve the best detection results at low false alarm probability. The detection results of one month data in August 2017 show that the mean method has the strongest false alarm suppression ability compared with other data fusion methods with the same detection probability.