Patent attributes
A predictive contrastive representation method for multivariate time-series data processing includes: mapping temporal coding information at a current moment and future situational information by using a logarithmic bilinear model to obtain a similarity; training the similarity according to a noise contrastive estimation method and prediction situational label data, and constructing, based on a training result, a predictive contrastive loss function of the temporal coding information at the current moment and the future situational information; sampling the prediction situational label data based on a corresponding optimal loss in the predictive contrastive loss function, optimizing the predictive contrastive loss function by using a direct proportion property between the sampling probability and the similarity, constructing mutual information between the temporal coding information at the current moment and the future situational information based on the optimized predictive contrastive loss function, and performing predictive contrastive representation by maximizing the mutual information.
