Satellite anomaly detection method and system based on prototype negative sample mixed comparison

The invention provides a prototype negative sample mixed comparison satellite anomaly detection method and system, and the method comprises the steps: inputting a satellite telemetry data set to a trained deep neural network model, obtaining the anomaly score of data, determining an anomaly score threshold value according to an anomaly sample proportion epsilon obtained through training, sorting the anomaly scores according to an ascending order, and obtaining an anomaly score threshold value; the samples of the front epsilon are judged to be abnormal; the training process of the deep neural network model comprises the steps of clustering samples with similar semantics into the same group to obtain pseudo labels of the samples, performing sample feature mixing by taking the distance between an anchor point and a prototype to which a negative sample belongs as a weight, generating a difficult-to-negative sample, and guiding neural network learning; and constructing an abnormal fractional function by adopting t