A new method for multivariable nonlinear coupling relations analysis in complex electromechanical system

Abstract

Coupling relations analysis of monitoring variables in complex electromechanical system is a powerful and useful means for abnormal detection, false monitoring information identification and fault diagnosis. However, due to the characteristics of multivariable, nonlinear and non-stationarity in the actual production process, it is difficult to achieve multivariable coupling modeling of complex electromechanical systems. In this paper, a new coupling modeling method is proposed by combining causality analysis with RBF neural network. First, considering the multivariate and nonlinearity of complex system, the conditional Granger and nonlinear Granger is combined to analyze the causal relations of process variables and obtain the cause variable set of any variable in complex system. Second, the RBF neural network is applied to achieve the nonlinear fitting and the parameter is optimized to ensure the fitting precision. Finally, the effectiveness of the proposed method is verified by an analysis of one case study of real compressor groups data set in chemical production system. This new approach can handle general coupling modeling problems and obtain a quantitative nonlinear coupling model by determine the dependent and independent variables in system and the functional relationship between them. Which dedicated to studying quantitative functional relationship of variable coupling, not just considering the direction or strength of coupling as in the existing, and does not need any prior knowledge about the physical structure. Thus, the proposed method can be effectively used in coupling modeling of complex electromechanical systems and formulate the foundation of anomaly detection, information quality assessment, and failure propagation mechanism, as well as other engineering applications.