In order to fully explore the temporal characteristic relation between power grid operation information and area control error, and improve the prediction accuracy of area control error, an area control error prediction method based on principal component analysis (PCA) and extreme gradient boosting (XGBoost) was proposed. The feature selection of the relevant variables affecting the area control error was carried out with PCA, and the coupling between the features was eliminated. The principal components of the extracted variables were input into XGBoost, and the mapping relationship between the current power grid operation data and the future ACE in high-dimensional space was determined through the training model parameters. Therefore, an area control error prediction model based on PCA-XGBoost is established. Through the actual data verification of a regional power grid, compared with other methods, the proposed area control error prediction model has significant advantages in forecasting accuracy and generalization ability.