We recommend you cite the published version. The publisher's URL is http://dx.doi.org/10.1109/ TNN.2008.2003223 Refereed: Yes (c) 2009 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other users, including reprinting/ republishing this material for advertising or promotional purposes, creating new collective works for resale or redistribution to servers or lists, or reuse of any copyrighted components of this work in other works.
DisclaimerUWE has obtained warranties from all depositors as to their title in the material deposited and as to their right to deposit such material. UWE makes no representation or warranties of commercial utility, title, or fitness for a particular purpose or any other warranty, express or implied in respect of any material deposited.UWE makes no representation that the use of the materials will not infringe any patent, copyright, trademark or other property or proprietary rights.UWE accepts no liability for any infringement of intellectual property rights in any material deposited but will remove such material from public view pending investigation in the event of an allegation of any such infringement.
PLEASE SCROLL DOWN FOR TEXT.
I. INTRODUCTIONN EURAL NETWORK (NN) has been extensively studied and applied as a generic and powerful black-box modeling technique to enhance complex nonlinear system modeling and identification [1]- [4]. In system identification procedure, validation is the final step to check the adequacy of identified models. Because an NN could be incorrectly designed due to many problems such as incorrect network selection, incorrect input vector selection, insufficient training, and overfitting, validation is a very important means to determine if the NN agrees sufficiently well with the observations. Generally, the basic statistics of residuals, e.g., mean and variance or standard deviation, are considered to be critical indices for the identified NNs to check whether the residuals are reduced to the lowest possible levels. However, the levels of the residuals sometimes cannot clearly and directly indicate the adequacy of the identified NNs since the original systems are always contaminated in unknown noisy environments.To properly validate linear and nonlinear models, several methods for model validation based on correlation tests have been developed that are based on the concept that if a model is valid, the residuals should be reduced to a white noise and uncorrelated to the delayed system inputs and outputs [5]. Manuscript received September 21, 2006; revised August 23, 2007 and April 04, 2008; accepted April 16, 2008 . In comparison to previous higher order correlation-test-based approaches, the new method provides an enhanced nonlinear correlation detection power and a condensed correlation illustration. It should be mentioned that almost all previous validity tests only focus on the correlation computation for residuals and inputs. In these methods, the correlation between residuals and delayed outputs is i...