The invention discloses a transformer state monitoring data anomaly detection method. The method includes the following steps that S1, multi-dimensional transformer state monitoring data is preprocessed to obtain standardized state monitoring data; S2, a sliding window is built for the standardized state monitoring data, and k-means clustering is carried out on the data in the sliding window; S3, abnormal data and abnormal moments in the sliding window are screened out to construct an abnormal dataset; S4, whether the obtained state monitoring data is transformer running state abnormal data or sensor noise is judged according to the abnormal dataset in the step S3, the normal moments of the running state of a transformer are detected at the same time if the state monitoring data is the transformer running state abnormal data, and the sensor noise is removed if the state monitoring data is the sensor noise. The method has the advantage of achieving abnormal state real-time detection.