The invention belongs to the field of power plant safety control systems, and particularly relates to a multi-dimensional data anomaly detection method and device based on XGBoost. The method is characterized by comprising a first step of data acquisition and cleaning, a second step of performing standardization processing on cleaned data, and unifying dimensions between data of different dimensions; step 3, feature extraction and dimension reduction; step 4, exception detection model training: training the dimension reduction data by using an XGBoost method, and establishing a prediction model of equipment exception; and step 5, abnormity on-line detection is carried out, and if a given threshold value is exceeded, it is determined that abnormity occurs. The method is suitable for processing and predicting important abnormal events of equipment, the thought and technology of ensemble learning are fully utilized, important features in multi-dimensional data information detected by an equipment sensor are effectively utilized, and then online abnormal detection based on real-time measuring point data of a power plant is achieved. The method is large in collected data volume, small in analysis error and high in early warning result accuracy.