Nonlinear Model Identification and Data Reconciliation Using Kernel Principal Component Regression

Reconciliation of process data is an important preprocessing technique, the main purpose of which is to obtain accurate estimates of variables and model parameters. Reconciliation requires a process model which is generally developed using first principles. For many complex processes, the development of such models is difficult and time-consuming. In this work we propose a novel alternative method for steady state data reconciliation of nonlinear processes which does not require a functional model between variables to be specified a priori. A nonlinear model relating the variables is developed from a given data set, while simultaneously obtaining accurate estimates of the measured variables. The method we propose combines concepts drawn from Kernel Principal Components Regression with an error-in-variables model parameter estimation technique. Simulation studies demonstrate that the proposed approach is able to improve the accuracy of measured variables. The identified nonlinear model is also useful for reconciling future measurements of the process.