Investigations　of　seismic　response　of　underground　structures　often　assume　homogeneous　or　layered　homogeneous　sites.　However,　significant　spatial　variability　in　soil　parameters　may　lead　to　vastly　different　underground　structure　performance　from　that　obtained　for　homogeneous　sites.　Based　on　random　field　theory,　this　study　models　the　spatial　variability　of　the　soil　elastic　modulus,　cohesion,　and　friction　angle　using　the　Karhunen-Loe`ve　(K-L)　expansion　method.　Target　acceleration　response　spectra　are　generated　according　to　standards,　and　the　trigonometric　series　method　is　employed　to　create　artificial　seismic　waves　of　four　different　intensities.　Nonlinear　dynamic　analyses　of　underground　structures　under　deterministic　and　random　field　conditions　are　conducted　using　ABAQUS　software.　The　study　comprehensively　analyzes　the　structural　damage　state,　internal　forces,　interstory　displacement,　and　drift　ratio　to　evaluate　the　station　structure＇s　performance　under　different　seismic　intensities.　Results　show　that　the　spatial　variability　of　soil　parameters　significantly　impacts　the　dynamic　response　of　underground　structures,　especially　for　stronger　earthquakes.　The　variability　of　soil　stiffness　and　strength　parameters　leads　to　greater　fluctuations　and　uncertainties　in　displacement　and　internal　force　responses,　exacerbating　structural　damage.　It　is　recommended　that　when　the　peak　ground　acceleration　(PGA)　reaches　or　exceeds　0.5　g,　the　spatial　variability　of　soil　parameters　should　be　incorporated　into　the　analysis　to　ensure　a　reliable　assessment　of　the　structural　seismic　performance.