When the slurry shield passes through the composite formation, the parameters of the shield tunneling and slurry separation system often fluctuate greatly, influencing construction safety and tunnelling efficiency. It is difficult to extract and deal with the parameters of slurry separation system, there are few studies on the prediction of shield slurry separation system. Base on the Wangjing Tunnel, the coordinated control technology solid-liquid separation of screening, two-stage cyclone, and centrifugal / pressure filtration are adopted according to the stratum conditions, solving the contradiction between shield tunneling and slurry treatment. Collect shield tunneling parameters (tunnelling speed, cutter head speed, total propulsion force, penetration) and slurryr separation parameters (feed and discharge volume, specific gravity, viscosity). Improve data quality through Cook distance outlier detection and wavelet threshold denoising, . The 12 parameters are input specific gravity ratio and viscosity ratio of two-stage cyclone separation, and the output parameters are discharge volume, discharge specific gravity and discharge viscosity，BP neural network was established to predict parameters of the slurry separation system, three different formation rings were selected for prediction and analysis. The prediction results show that the average prediction errors are all within 5%, they still have high accuracy under the composite formation, which provides reference in risk prediction and parameter selection.