Abstract: In order to reduce the downtime and maintenance time of a DC system and improve the stability of converter valve operation, a state evaluation method for UHV converter valve based on multi-state data equalization and extreme gradient boost (XGBoost) is proposed. Firstly, for the main components of the thyristor converter valve, the four characteristic indexes of the thyristor component, the valve cooling component, the valve arrester and the external environment are extracted. A data preprocessing method based on isolated forest and synthetic minority oversampling technology is proposed, where the outlier samples in the data set are removed, and then the minority samples are oversampled to achieve the effectiveness and balance of the data set in each state. The preprocessed data are utilized to train the XGBoost classifier, and the optimal hyperparameters of the model are obtained based on the K-fold cross-validation and grid search method. Finally, the measured data of a converter station in Jiangsu Province are taken as an example to verify the proposed method, and the results show that the accuracy rate of the evaluation model is 97.1%, which is more accurate than traditional methods for judging the operating state of the converter valve. In addition, the model can reflect the characteristic contribution of each state quantity, and provide a basis for the maintenance of the converter valve.