The　shape　of　power　spectra　both　for　typhoon　and　non-typhoon　winds　directly　affect　the　response　level　of　the　structure.　The　conventional　deterministic　spectrum　models　fail　to　reproduce　the　randomness　of　the　structure　response　and　could　underestimate　the　real　vibration　amplitudes,　especially　for　typhoon　winds　featured　with　stronger　gustiness　due　to　internal　circulation　and　thermodynamic　effects.　Based　on　long-term　observation　data　captured　by　the　structural　health　monitoring　system　installed　at　Xihoumen　suspension　bridge,　the　parameters　of　wind　power　spectra　in　along-wind,　cross-wind　and　vertical　directions　for　typhoon　and　non-typhoon　winds　are　extracted,　respectively.　The　statistical　characteristics　and　correlations　among　these　parameters　are　examined.　The　data-driven　stochastic　power　spectra　models　are　then　proposed　by　utilizing　the　moment-based　theoretical　solutions　and　Monte　Carlo　technique,　respectively.　The　mean　wind　speed　is　incorporated　in　these　models　which　allows　the　random　simulations　of　power　spectra　with　respect　to　different　mean　wind　speeds.　The　proposed　stochastic　model　is　finally　applied　to　generate　a　large　number　of　wind　power　spectra　adapted　to　the　mean　extreme　wind　speed　return　period　curves　in　typhoon　and　non-typhoon　mixed　climates.　It　is　suggested　that　the　present　stochastic　power　spectra　model　can　be　applied　to　estimate　the　random　response　of　structures　at　different　return　periods　due　to　typhoon　and　non-typhoon　winds,　which　can　be　extended　to　conduct　the　performance-based　design　and　uniform-risk-based　design　of　structures　in　wind　engineering.