Abstract: There are many problems in complex mountain wind farms,such as large errors of wind resource evaluation and low accuracy of wind power prediction caused by poor quality of observed wind speed data. Because of the strong intermittent,fluctuating,and nonstationary characteristics presented by the wind speed in complex mountain wind farms,conventional quality control methods cannot effectively improve data quality. For this situation,an integrated learning algorithm（PVL）based on particle swarm optimization improved variational modal decomposition improved by particle swarm optimization and long short-term memory is proposed and applied to the quality control of wind speed observations in complex mountainous areas to improve the quality of wind speed data. In order to assess the feasibility and applicability of the proposed method,the 10 minutes wind speed observed in five observation tower of a complex mountain wind farm in Guangxi from 2015 to 2016 were examined. Otherwise,we compared this method to spatial regression test（SRT）and inverse distance weighting method（IDW）. The results show that the method can more effectively flag suspicious data,and it also has the advantages of high identification accuracy,strong adaptability to different terrains and wind conditions.