Vehicle　driving　conditions　at　high　altitudes　are　quite　different　from　those　in　plain　areas.　Accurately　predicting　the　speed　of　vehicles　traveling　at　high　altitudes　is　of　great　significance　for　the　development　of　vehicle　safety　assisted　driving.　In　order　to　study　the　vehicles＇　speed　of　highways　in　high-altitude　areas,　and　predict　the　speed　of　vehicles　accurately,　more　than　30　000　data　were　collected　by　a　medium-sized　SUV　in　Qinghai　under　typical　adverse　environmental　conditions.　The　real-time　vehicle　status　data　(engine　speed,　engine　torque,　transmission　gear,　throttle　opening),　road　alignment　data,　and　historical　vehicle　speed　data　was　denoised　by　wavelet　method.　The　collecting　data　is　time-varying　and　nonlinear　characteristics.　A　nonlinear　auto-regression　with　exogenous　inputs　(NARX)　dynamic　neural　network　prediction　model　was　established　in　this　paper　to　fit　travel　speed.　The　network　model　after　training　has　small　error　and　high　fitting　degree.　The　accuracy　of　vehicle　speed　prediction　in　the　next　five　seconds　is　96.01%.　The　root　mean　square　error　of　prediction　is　less　than　2.21　km/h,　which　can　achieve　better　prediction　effect.　At　the　same　time,　the　transplantability　of　the　model　is　enhanced　by　taking　altitude　and　road　alignment　as　variables.　©　Published　under　licence　by　IOP　Publishing　Ltd.