Hybrid algorithm combining matched filtering and convolutional neural networks for searching gravitational waves from binary black hole mergers
匹配滤波与卷积神经网络混合算法搜索双黑洞并合引力波
Efficient searches for gravitational waves from compact binary coalescence are crucial for gravitational-wave observations. We present a proof-of-concept for a method that utilizes a neural network taking a signal-to-noise ratio map, a stack of signal-to-noise ratio time series calculated by the matched filter, as input and predicting the presence or absence of gravitational waves in observational data. We train the neural network with a data set of gravitational-wave signals from stellar-mass black hole mergers injected into stationary Gaussian noise. We use data set 1 of the mock data challenge MLGWSC-1 to assess the ability of the proposed algorithm. The estimated sensitivity distance is 2428.10 Mpc at the false alarm rate of 1 per month. These results indicate that our algorithm achieves reasonable sensitivity with practical computational resources.
展开 ▾创新性地将所有模板的SNR时间序列堆叠为“SNR地图”作为网络输入,利用模板间相关性提升灵敏度,并在MLGWSC-1 mock data challenge中展现了与传统流水线可比拟的检测能力。
引力波探测依赖匹配滤波方法,但实际噪声的非高斯性导致大量计算时间消耗在后处理(veto)阶段。深度学习方法被证明能快速分析数据,George & Huerta 2018 首次展示了神经网络在双黑洞并合检测中的潜力。此后,多种架构如 Nousi et al. 2023 和 Schäfer et al. 2022 不断涌现,证明了其可比甚至超越传统流水线的性能。在预处理中,已有工作将最佳匹配模板的SNR时间序列输入网络(如 Beveridge et al. 2025),本文进一步将模板库中全部模板的SNR时间序列堆叠为图像,尝试利用模板间的相关性提升灵敏度。在MLGWSC-1 mock data challenge Schäfer et al. 2023 中,该方法表现出与标准PyCBC流水线相当的灵敏度,且计算资源可行。未来可扩展到真实非高斯噪声、高质双黑洞(减少glitch混淆)及低质系统(需考虑高阶模和自旋进动),并有望集成到现有流水线中作为候选筛选器,大幅缩短搜索时间。
预印本 2025-12-13 · 刊出 2026-03-26 · 收录 2026-10-12