The actual operation of wind turbines is complex. Due to the time-varying wind speed, harsh outdoor environments, high failure rates of various components, and human power restrictions, the SCADA system of wind turbines contains a certain proportion of abnormal data in the measured operation data, and the types of abnormalities are complex. If raw data is used for data abnormal state analysis, it will cause significant deviation in the analysis results, leading to a decrease in the accuracy of SCADA data abnormal states. Therefore, an intelligent judgment method for SCADA data abnormal states based on time-varying hidden Markov models is proposed. Using sliding window technology and data augmentation state matrix to smooth SCADA data and obtain denoised data, providing high-quality data for intelligent determination of abnormal data states in the future. Classify and process data based on the features of denoised data. Using a time-varying hidden Markov model to compare the attributes of normal data and the attributes of classified sub datasets, if they are consistent, it indicates that the data state is normal; if they are inconsistent, it indicates that the data is abnormal, thus achieving intelligent judgment of SCADA data abnormal states. The experimental results show that this method can accurately determine the abnormal state of SCADA data and ensure the safe and stable operation of the SCADA system.