Fan fault monitoring and diagnosis method based on online learning and multivariate state estimation

The invention relates to a fan fault monitoring and diagnosis method based on online learning and multivariate state estimation. The fan fault monitoring and diagnosis method comprises the following steps: acquiring variable data related to a generator and having a sampling frequency of 5 minutes from a master control SCADA system as an original data set; preprocessing the data by a DBSCAN algorithm based on probability distribution and interval distribution, deleting abnormal samples and fault samples, and dividing a preprocessed data set into a training data set and a test data set; performing normalization operation on the training data set and the test data set; establishing an MSET-based model through an improved memory matrix construction method; carrying out fault early warning based on a self-adaptive similarity threshold value; and after abnormal early warning is triggered, an early warning result is pushed to a centralized control center. According to the method, the sample redundancy in the memory