DYNAMIC DATA RECTIFICATION BY RECURRENT NEURAL NETWORKS VS TRADITIONAL METHODS

Recurrent neural networks are used to demonstrate the dynamic data rec tification of process measurements containing Gaussian noise. The perf ormance of these networks is compared to the traditional extended Kalm an filtering approach and to published results for model-based nonline ar programming techniques for data reconciliation. The recurrent netwo rk architecture is shown to provide comparable, if not superior, resul ts when compared to traditional methods. The networks used were traine d using conventional nonlinear programming techniques in a batch fashi on.