Process technology in real time data verification and reconciliation for optimal oil production under the presence of uncertainties

Abstract

Measurements collected during oil production can be used to improve the modeling and optimization of oil well systems. However, these measurements often suffer from errors due to various types of noise, causing the observed data to violate operational constraints like mass and energy balance. Data reconciliation has long been a topic of great interest in a wide range of process technologies. By using the constraints of a system, data reconciliation adjusts the measurements to minimize the bias between the measurements and the corresponding quantities, further helping to estimate the real output of the system. On the other hand, model fitting attempts to improve the model of a system using observed data.

The research focuses on two types of oil well systems: gas lift oil wells and electric submersible pump (ESP) lift oil wells. The PhD research study set out to explore methods of data reconciliation and model fitting for oil well production systems. Dynamic observation was collected from oil well simulators. The estimation of uncertainty is attained in some methods. Some approaches are developed based on offline batch data, while others can be implemented in real time. The phrase ‘batch data’ will be used in this study to describe the collected measurement.

Filtering techniques are one of the most common approaches in data reconciliation. Reconciled variables include both the system output and the system state. The first principles models of oil wells contain the relationships between variables and the dynamics of oil operation. Thus, the model and the prior quantities are vital to be used as process nonlinear constraints for performing data reconciliation. Papers B and D include the application of nonlinear filters to different oil production systems. Machine learning is another way to solve data reconciliation problems. A physics-informed neural network (PINN) was proposed based on first principles models of an oil production system in paper C.

Parameter estimation is a continuing concern within model fitting. Dynamic parameter estimation in oil production systems has received scant attention in the research literature. Maximum likelihood estimation and Bayesian analysis are two dominant approaches to address the parameter estimation problem. Compared with maximum likelihood estimation, Bayesian analysis offers important insights into the probability of parameters. Based on Bayesian inference, the parameter estimation problem becomes the acquisition of the joint posterior probability distribution. The estimation of parameters and their uncertainty is described using a probability density function. The first principles models of gas lift oil fields and ESP lift oil fields are nonlinear time series models, whose inference is difficult to evaluate analytically. To avoid the difficulty of mathematical calculation, Monte Carlo methods can be employed using sampling. In papers A, B, and D, Markov chain Monte Carlo (MCMC) methods are applied and compared in diverse scenarios for different oil well fields.