Combined Data Reconciliation and Parameter Estimation

This paper presents a nonlinear approach for data reconciliation. The advantages of this approach over the conventional extended Kalman filtering (EKF) are namely, less linearization errors are generated in the process, and secondly the choice of the objective function is flexible. In this work, two probability density functions, namely the Logistic and Lorentz distribution are proposed as the objective function to be optimized in this approach. These two functions are proven to be statistically robust, which is an advantage over the conventional weighted least squares function. The extended Kalman filter and the modified nonlinear approach are implemented and verified via two case studies, namely a simulation case study and more importantly an experimental case study of a heat exchanger. The results obtained from the simulation case study demonstrated a reduction of approximately 50% error of the nonlinear approach over the extended Kalman filter.