Gravitational-wave background detection using machine learning
基于机器学习的引力波背景探测
Extracting the faint gravitational-wave background (GWB) signal from dominant detector noise and disentangling its astrophysical and cosmological components remain significant challenges for traditional methods like cross-correlation analysis. We propose a novel hybrid approach that combines deep learning with Bayesian inference to identify and characterize the GWB more rapidly than current techniques. Our method utilizes a custom-designed multiscale multiheaded autoencoder (MSMHAutoencoder) architecture to separate GWB signals from detector noise and, subsequently, Markov chain Monte Carlo parameter estimation to disentangle the GWB components. Using simulated data representative of the LIGO-Virgo-KAGRA network at design sensitivity, we show that our MSMHAutoencoder can detect with high confidence (log noise Bayes factor of 3) a GWB from compact binary mergers with fractional energy density <inline-formula><mml:math><mml:mrow><mml:msub><mml:mrow><mml:mi>Ω</mml:mi></mml:mrow><mml:mrow><mml:mi>CBC</mml:mi></mml:mrow></mml:msub><mml:mo>≈</mml:mo><mml:msup><mml:mrow><mml:mn>10</mml:mn></mml:mrow><mml:mrow><mml:mo>-</mml:mo><mml:mn>9</mml:mn></mml:mrow></mml:msup></mml:mrow></mml:math></inline-formula> at 25 Hz. In the presence of such an astrophysical GWB, we can simultaneously measure a flat cosmological component as faint as <inline-formula><mml:math><mml:msub><mml:mi>Ω</mml:mi><mml:mtext>cosmo</mml:mtext></mml:msub><mml:mo>≈</mml:mo><mml:mn>1.3</mml:mn><mml:mo>×</mml:mo><mml:msup><mml:mn>10</mml:mn><mml:mrow><mml:mo>-</mml:mo><mml:mn>10</mml:mn></mml:mrow></mml:msup></mml:math></inline-formula> using 47.4 days of training data.
展开 ▾设计多尺度多头自编码器,结合物理启发损失与课程学习,在 47.4 天训练数据上以高置信度探测到 Ω_BBH≈10⁻⁹,并可通过 MCMC 同时区分低至 ~1.3×10⁻¹⁰ 的宇宙学成分,灵敏度增益超越传统交叉相关方法。
以往 LIGO-Virgo-KAGRA(LVK)合作组主要采用交叉相关方法对各向同性引力波背景(GWB)进行上限估计 Abbott+ 2021,并利用贝叶斯推断尝试区分天体物理与宇宙学成分,但受限于探测器灵敏度,Martinovic+ 2021 指出当前网络难以有效分离。本文提出的多尺度多头自编码器(MSMHAutoencoder)配合物理启发损失及课程学习策略,在模拟设计灵敏度数据中展示了快速检测与成分解析能力。这一混合方法为克服传统方法瓶颈开辟了新路径,未来通过融入更真实的噪声模型、多类型宇宙学源以及应用于实际 LVK 数据,有望进一步提升对微弱宇宙学背景的探测能力;同时,深度学习在引力波数据处理中的广泛应用 Cuoco+ 2025 也将促进该架构的持续优化。
预印本 2025-06-17 · 刊出 2025-09-26 · 收录 2026-10-12