Multivariable clustering and fusion time series combination prediction method

Abstract

The invention discloses a multivariable clustering and fusion time series combination prediction method; aiming at solving the problems that an existing neural network model does not have a specific learning mechanism and cannot fully mine data structure feature information, from the multivariable directed coupling perspective, and in combination with the advantages of a graph convolutional neuralnetwork and a long-term and short-term memory network, the invention provides a multivariable clustering and fusion time series combination prediction method. The method comprises the following steps: firstly, exploring a causal transfer relationship between variables based on coupled Granger causal measure analysis; secondly, establishing a directed weighted network according to a variable causality analysis result, extracting node and edge weight characteristics of the directed weighted network, and embedding the weight of a target variable into a graph convolutional neural network for training to realize accurate classification of monitoring variables; finally, taking the non-target monitoring variable time series contained in the community where the target monitoring variable is located as input, and predicting the target monitoring variable based on the long-term and short-term memory neural network. The method is applied to verification of a compressor unit monitoring sequence in a chemical production system, and results show that the method is superior to a traditional node classification method in the aspects of prediction accuracy and calculation complexity, and the proposed method can also maintain high prediction capability in an abnormal state of the system.

Description

technical field [0001] The invention relates to the technical field of time series prediction, in particular to a multivariate clustering and fusion time series combined prediction method. Background technique [0002] The complex electromechanical system of the process industry represented by the energy and chemical industry involves the exchange of various media such as energy, information, and matter, and has the characteristics of harsh process conditions, large-scale equipment, continuous processing, and fine production control requirements. The mutual coupling of the monitoring variables in the system essentially forms a network diagram representing the dynamic changes of the complex electromechanical system, and the basis for constructing a complex network is to clarify the representational relationship between each node and its edges and even weights. Therefore, many scholars Many studies have been conducted on the correlation between variables. For example, Sui et ...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the above-mentioned problems in the prior art, the object of the present invention is to provide a combined time series forecasting method of multivariate clustering and fusion. There is no specific learning mechanism for the existing neural network model, and the data structure feature information To mine insufficient problems, from the perspective of multi-variable directed coupling, combined with the advantages of graph convolutional neural network and long-term short-term memory network, using graph convolutional neural network to train complex networks based on coupled Granger causality measure analysis, It aims to discover the community structure of the target monitoring variable in the complex network, so as to reasonably divide the composition of the network nodes, so as to realize the accurate prediction of the time series of the target monitoring variable

Images