Abstract:
In the traditional linear multivariate Granger causality test, conditional variables are introduced to determine whether the causal relationships exist between every two variables or not. Its disadvantage is that the selection of conditional variables is often subjective and lacks reasonable rules. In order to deal with the problem, this paper proposes an improved nonlinear multivariate Granger causality test method with selecting conditional variables. It uses support vector regression to construct test equations to adapt to nonlinear conditions. After analyzing the relationship between two variables and building a preliminary structure, it selects the conditional variable, and then conducts a nonlinear multivariate Granger causality test based on the selected conditional variable. Two kinds of topological structures are introduced to avoid the repeated inspection on some real relationships that do not produce pseudo causality problems. The experimental results on numerical simulation and wastewater treatment benchmark simulation model show that the proposed method can adapt to nonlinear conditions. After screening the condition variables to reduce the interference of independent variables and introducing topology, this method has more accurate test results and better performance in calculation strength.