Method and device for evaluating data accuracy based on complex electromechanical system coupling relation model

Abstract

The invention discloses a method and device for evaluating data accuracy based on a complex electromechanical system coupling relationship model. The method comprises: preprocessing real-time monitoring data of a complex electromechanical system; determining a reason variable set of each monitoring variable based on multivariable Granger causality analysis; based on the monitoring variable and thereason variable set thereof, performing nonlinear relationship simulation by using a neural network, and determining a nonlinear mapping relationship between the monitoring variable and the reason variable set thereof; and obtaining a coupling relationship model among the monitoring variables, and realizing monitoring data accuracy evaluation by utilizing the model. According to the method, a causal influence mechanism between variables is considered, by combining the advantages of causal analysis and machine learning, nonlinear simulation is performed on the coupling causal relationship of the complex electromechanical system, so that effectiveness evaluation of the monitoring data is realized, and the problems of long monitoring data evaluation time consumption, low evaluation accuracyand the like caused by modeling difficulty / model complexity in data accuracy evaluation based on a model in the prior art are solved.

Description

Technical field [0001] The invention belongs to the field of complex electromechanical system data monitoring and analysis, and relates to a method and device for evaluating data accuracy based on a complex electromechanical system coupling relationship model. Background technique [0002] The complex electromechanical systems represented by chemical production and nuclear power generation have the characteristics of high coupling and high correlation between monitoring variables. At the same time, as the production environment changes and process adjustments, there is a dynamic coupling relationship between the variables. Therefore, exploring the coupling constraint relationship and coupling characteristics between the variables in the system, analyzing the evolution of the coupling characteristics, and constructing the coupling relationship model between the monitoring variables in the system are the basis for system monitoring data anomaly detection and accuracy evaluation. [0...

AI Technical Summary

Problems solved by technology

Support vector machines can better deal with nonlinear data modeling of small samples, and have been widely researched and applied in pattern recognition, classification, etc.; Copula has obvious advantages in dealing with nonlinear structural problems, and can deal with tail-related problems of data. However, the effect of dealing with chaotic time series is general; Bayesian network analysis often requires data to satisfy normal distribution or near-normal distribution, and the actual system state information does not obey the normal distribution. Data capacity has been widely used, but because the traditional neural network does not consider the causal relationship between variables in the actual process, it is only a simple nonlinear mapping of input and output, which cannot fully express the non-linearity in the actual system. linear relationship

[0005] As one of the important methods of data accuracy evaluation, the monitoring data accuracy evaluation based on the coupling relationship model has two main disadvantages: 1) the construction of the coupling model is difficult and the model is complex, which leads to poor real-time evaluation of data accuracy; 2) the traditional There is a lack of causal impact analysis in the process of modeling from the aspect of machine learning to achieve data accuracy assessment, and the system coupling relationship is not fully reflected, resulting in low accuracy of assessment results

Images