Dynamic data reconciliation with simultaneous time-varying parameter estimation in real time: application to an electric submersible pump lift oil production

Abstract Data reconciliation techniques have been the subject of many classic studies in the data conditioning process. By reconciling the measurements, accurate estimation of the system output and unmeasured variables is provided. However, accurately determining measurement noise and parameter uncertainty in real time remains a significant challenge. How to simultaneously estimate parameters in the system has been attracting considerable interest. So far, very little attention has been paid to time-varying parameter estimation in oil production systems. In particular, estimation of parameter dynamics and the corresponding uncertainty without prior knowledge remains challenging. This work extends a previous study on dynamic parameter estimation by considering scenarios where parameters change both gradually and abruptly. To address these dynamics, nonlinear filtering methods are employed and compared. A comparative analysis was conducted using both quantitative metrics and visualization plots to evaluate the performance of various approaches. Under the same abrupt parameter change scenario, nonlinear filter-based methods demonstrated superior performance in parameter estimation, achieving a root mean square error of $$6.56 \times 10^{-11}$$ , compared to $$7.84 \times 10^{-11}$$ for the MCMC-based method-even without the use of prior information. Additionally, nonlinear filters showed a significant advantage in simultaneous state estimation, with a root mean square error of $$1.94 \times 10^{4}$$ , markedly lower than the $$1.47 \times 10^{6}$$ observed with the MCMC-based approach. The effectiveness of nonlinear filtering methods was further validated in scenarios involving gradual parameter changes, again without relying on prior knowledge. This work provides an important opportunity to advance the understanding of dynamic parameter estimation in the gas and oil industry, and the improved model can possibly be applied to real-time optimization and model-based control. Graphical abstract