With the advent of third-generation(3G) ground-based gravitational wave(GW) detectors,such as the Einstein Telescope(ET),we anticipate a substantial enhancement in sensitivity across a wide frequency range.The machine learning approach for GW search necessitates an update to address the challenges posed by data features that deviate from those of 2G detectors.In this paper,we introduce a novel GW search pipeline specifically designed for 3G ground-based detectors like ET.Our pipeline leverages three types of deep learning models:an envelope extraction model,a denoising model,and an astrophysical origin discrimination model.Additionally,we propose a signal consistency test across multiple detectors.Given that denoising results vary among different detectors,we present a new method for selecting the optimal waveform.This selected waveform serves as a template for estimating the signal-to-noise ratio(SNR) of strain data from all detectors.Furthermore,if 3G detectors operate alongside 2G detectors,the templates derived from 3G detector data can be utilized to predict the SNR for 2G detectors,significantly reducing the computational burden of GW searches for the latter.We also assess the robustness of our method when applied to data containing binary neutron star(BNS) foreground noise.We believe that the proposed method holds promise for detecting BBH events in future 3G detectors.