multidimensional time series classification systems and methods of the parameters

摘要

systems and methods for classifying multidimensional time series of parameters the invention refers to traditional systems and methods implementing manual extraction of characteristics of variable duration series that result in complexity and require co-knowledge of domain. building classification models requires large labelled data and is computationally expensive. the ways in which this invention is realized implement learning models for multi-dimensional time series classification tasksperforming the extraction of the characteristics of the entity parameters by means of a non-super-visioned encoder and building a non-temporal linear classifier model. a fixed-dimensional characteristic vector is emitted using an unsupervised pre-trained encoder, which acts as a characteristic extractor off-shelf. the extracted ca-characteristics are concatenated to learn a non-temporal li-near classification model and weight is assigned to each characteristic during learning,which helps determine the parameters relevant to each class. the mapping of parameters for the target class is considered when restricting the linear model to use only a subset of large number of features.