Maximizing the annual energy production has become an issue for power producer. Conventionally, a yaw control of wind turbine systems uses nacelle mounted anemometers. On the other hand, there is a possibility that yaw misalignment will increase because of the difference between the wind direction measured by the nacelle mounted anemometers and the inflow wind direction. In this study, we proposed a method to estimate the wind direction measured by a doppler LiDAR using the deep learning with the wind direction measured by the nacelle mounted anemometer and the wind turbine operation information as input. The proposed method was evaluated for upwind and downwind turbines, and the results showed that it was suitable for estimation of the doppler LiDAR wind direction. Future work includes the implementation of yaw control test using the deep learning model on an actual wind turbine to verify the improvement of annual energy production. In addition, it is necessary to develop a method to realize early deployment of the constructed deep learning model to other wind turbines and other sites.