Process Data Mining for Parameter Estimation: With the DYNIA Method

Abstract

Updating the model parameters of the control system of an oil and gas production system for the reasons of cost-effectiveness and production optimization, requires a data set of input and output values for the system identification procedure. A requirement for the system identification to provide a well performing model is for this data set to be informative. Traditionally, the way of obtaining an informative data set has normally been to take the production system out of normal operational order, in the interest of performing experiments specificially designed to produce informative data. It is however desirable to use segments of process data from normal operation in the system identification procedure, as this eliminates the costs connected with a halt of operation. The challenge is to identify segments of the process data that give an informative data set. Dynamic Identifiability Analysis (DYNIA) is an approach to locating periods of high information content and parameter identifiability in a data set. An introduction to the concepts of data mining, system identification and parameter identifiability lay the foundation for an extensive review of the DYNIA method in this context. An implementation of the DYNIA method is presented. Examples and a case study show promising results for the practical functionality of the method, but also raise awareness to elements that should be improved. A discussion on the industrial applicability of DYNIA is presented, as well as suggestions towards modifications that may improve the method.