Abstract: To evaluate the accuracy of ERA5 wind speed data and enhance its application value, this study conducted a comparative analysis between ERA5 reanalysis data and lidar wind profile data obtained during the 2023 spring voyage in the Southeast Atlantic. The comparison results indicate that ERA5 data performs well in predicting low-level wind speeds but exhibits significant deviations in the boundary layer and higher altitudes, particularly under complex weather conditions such as turbulence, wind shear, severe convection, and precipitation. To address these deviations, this study developed a deep learning-based correction method, using lidar-measured wind profile data as the calibration reference to optimize the ERA5 reanalysis wind speed data. The results show that the corrected ERA5 data aligns more closely with lidar observations. Under complex meteorological conditions, the correlation coefficient between the corrected ERA5 wind speed data and the measured wind speed increased by 57.80%, while the root mean square error decreased by 33.26%, the mean absolute error decreased by 67.10%, and the bias was reduced by 75.25%. These findings demonstrate that the deep learning correction method based on lidar-measured data can effectively improve the accuracy of reanalysis wind speed data, providing new insights and references for numerical wind field foreasting and research on air-sea interactions.