Title :
A Correlation-Test-Based Validation Procedure for Identified Neural Networks
Author :
Zhang, Li Feng ; Zhu, Quan Min ; Longden, Ashley
Author_Institution :
Dept. of Economic Inf. Manage., Renmin Univ. of China, Beijing
Abstract :
In this study, an enhanced correlation-test-based validation procedure is developed to check the quality of identified neural networks in modeling of nonlinear systems. The new computation algorithm upgrades the validation power by including a direct correlation test between residuals and delayed outputs that have been quoted indirectly in the most previous approaches. Furthermore, based on the new validation procedure, three guidelines are proposed in this study to help explain the validation results and the statistic properties of the residuals. It is hoped that this study could promote awareness of why the correlation tests are an effective method of validating identified neural networks, and provide examples how to use the tests in user applications.
Keywords :
identification; neural nets; nonlinear systems; statistical testing; correlation-test-based validation procedure; identified neural network; nonlinear systems modeling; statistical analysis; Correlation functions; model validation; neural networks; nonlinear dynamical systems; residuals; Algorithms; Artificial Intelligence; Computer Simulation; Neural Networks (Computer); Nonlinear Dynamics; Reproducibility of Results;
Journal_Title :
Neural Networks, IEEE Transactions on
DOI :
10.1109/TNN.2008.2003223
