General Electric Company
Label rectification and classification/prediction for multivariate time series data

Abstract:

A method of directional label rectification with adaptive graph for multivariate time-series data includes obtaining an input matrix containing sensor data and a first event matrix containing event data, identifying in the input matrix at least one feature pattern and a first corresponding time stamp, identifying in the first event matrix at least one fault signature and a second corresponding time stamp, if both a label matrix and an affinity matrix are known, then optimizing a weight matrix, else if both the label matrix and the weight matrix are known, then optimizing the affinity matrix, else optimizing the label matrix, creating a dynamically rectified event matrix by applying the label matrix, the affinity matrix, and the weight matrix to the first event matrix; and applying the dynamically rectified event matrix to forecast a future status of the asset. A system and computer-readable medium are disclosed.