Abstract: In order to determine the wind field characteristics of the plateau deep-cut canyon, a wind field measurement was carried out at a large-span bridge in the southwest mountain area. The instantaneous wind speed changes drastically, and it is difficult to accurately describe it with the constant mean wind speed. In order to solve the problem that the wind field is prominently non-stationary, the discrete wavelet transform method is used to extract the time-varying mean wind speed. A non-stationary wind speed model was established to analyze the turbulence wind characteristics, including turbulence intensity, integration scale, power spectrum density, etc., which were compared with the stationary model. The research results show that the mean wind speed conforms to the logarithmic normal distribution. As the wind speed increases, the stationarity of the fluctuating wind speed increases, and the stationarity of the non-stationary model is higher than that of the stationary model. The stationary model overestimates the turbulence characteristics of wind, and its turbulence intensity and integral scale are higher than those of the non-stationary model. The non-stationary model can significantly reduce the difference in turbulence characteristics between the non-stationary samples and the stationary samples, and can also reduce the difference between the downwind, crosswind and vertical components. The power spectrum density of the non-stationary model is applicable to the Von Karman spectrum at high-frequency, but it is significantly absent and deviated at low-frequency. The power spectrum density of the non-stationary model is overall higher than that of the stationary model. Therefore, the non-stationary model has good applicability because it can reduce the non-stationarity of the wind, its turbulence parameters distribution is stable, and it can weaken the anisotropy of wind speed fluctuation.