ABSTRACT In this paper, a semisupervised machining learning technique had been utilized to analyse low-resolution stellar spectra from Large sky Area Multi-Object fiber Spectroscopic Telescope (LAMOST) Data Release 9 (DR9). We identified approximately 111 000 potential very metal-poor stars. Estimation of their stellar parameters ($T_{\rm eff}$, ${\rm log}\, \rm {g}$, [Fe/H]) indicate that over 99 per cent are metal-poor ([Fe/H] < −1.0), comprising 32 631 very metal-poor ([Fe/H] < −2.0), 702 extremely metal-poor ([Fe/H] < −3.0) and 30 ultra metal-poor ([Fe/H] < −4.0) stars. Based on kinematic characteristics, stars were categorized into thick disc-like, thin disc-like, and halo-like groups. We analysed their metallicity distributions (MDs) with respect to vertical height ($|Z|$) and orbital eccentricity (e). Thick disc-like stars in current sample show a clear trend of decreasing metallicty with increasing $|Z|$ or e for the ranges −3.0 $\lt $ [Fe/H] $\lt $ −1.2 and $|Z|$$\lt $ 3 kpc. Conversely, thin disc-like stars in current sample exhibit a slight increase in the fraction of more metal-poor stars with $|Z|$ for the ranges −3.0 $\lt $ [Fe/H] $\lt $ −1.2 and $|Z|$$\lt$ 1 kpc, but no obvious correlation with e. Additionally, we confirmed the presence of two prominent substructures among halo-like stars. One exhibits a high eccentricity ($e\ \gt\ 0.8$) orbit and higher metallicity, while the other follows a retrograde orbit with moderate eccentricity ($e\sim 0.6$) and lower metallicity. We believe they are related to the merger events known as Gaia Sausage and Sequoia, respectively. Furthermore, our observations indicate that the Sequoia has lower eccentricity and metallicity compared to the Gaia Sausage.