Non-supervision-based intelligent power grid anomaly detection method

The invention provides an unsupervised smart power grid anomaly detection method, which is used for detecting power utilization anomaly behaviors, power utilization anomaly modes or network attacks in a power grid based on measured value statistical correlation, and a framework of a smart power grid attack detector comprises DBN modeling, mutual information for feature extraction and RBM for data training. The DBN and the mutual information are applied to an intelligent power grid test system composed of a plurality of measurement values, and the RBM is used for capturing a whole system mode extracted by a DBN model in an unsupervised mode; the smart grid attack detector extracts knowledge from the SCADA system; according to the method, starting from actual application, the key point is that characteristic quantity selection and sample preprocessing of massive online data are carried out on power equipment and a power network, and two different data mining methods are introduced to process the abnormal proble