Abstract: The zero-point shifting of yaw angle seriously affects the performance of the wind turbine, and the premise of elimination is to detect it accurately and quickly. Based on wind energy capture principle, this paper proposed a yaw angle zero-point shifting diagnostic method using machine learning algorithms. Firstly, isolated forest (IF) outlier detection algorithm was presented to preprocess the data; secondly, the non-parametric sparse Gaussian process regression (SGPR) model was established to evaluate the yaw angle zero-point shifting; finally, the proposed method was verified by actual operation data of wind turbines from multiple wind farms, and randomly collected datasets were used to analyze the dependence of different diagnostic models on the amount of data. Verification results show that: the proposed IF+SGPR diagnostic method is highly accurate, requires a small amount of data, and can quickly detect the yaw angle zero-point shifting. It can be applied to different types of wind turbines in different wind farms, and has a high universality.