Prof. Nikos Varotsis Director of the Fluid Analysis Laboratory of the School of Mineral Resources Engineering, Technical University of Crete. Former Vice Rector for Finance Planning and Development of the Technical University of Crete (2005-2010) and Department Head (1999-2003).
Before joining the University, he served as Head of the PVT and Fluid Analysis Research and Development Group for Schlumberger in Paris. He holds a Dipl. of Chemical Engineering from NTUA, MEng and a PhD degrees in Petroleum Engineering from Heriot-Watt University, Scotland.
Acting Consultant since 1990 on Phase Behaviour, Fluid Analysis topics for various international oil companies and service companies. He authored over 80 papers for the SPE conventions and journals as well as for other international scientific journals and conferences.
Petroleum Engineers and geo-scientists are faced with complex, non-linear problems as they attempt to predict and optimize recovery from hydrocarbon reservoirs. As the efficiency in production becomes an increasingly important issue in the oilfield, operators are realizing that in their large amount of data that they routinely collect lies a vast source of important information. State-of-the-art machine learning tools (MLTs) are nowadays more and more utilized to exploit available databases, such as PVT, core analysis data, well logs, well test interpretation data, seismic data, etc, for deriving very useful correlations which can describe hidden patterns without the need to provide in advance the correlating function form. In addition, Machine Learning Tools are also used for solving problems which otherwise are very time consuming or simply impossible to solve with conventional techniques.
In this presentation, three MLT applications, developed against a very large PVT database, for predicting reservoir fluids phase behavior and analysis are described. PVT Expert is a series of ANNs constructed to simulate accurately the complete set of PVT properties for both reservoir oils and gas condensates. For the second application, a quality assurance evaluation tool was developed, which by checking the affinity of an unknown fluid with the fluids included in the dataset against which the PVT Expert was developed as well as the confidence by which the ANN models have learned the general trend in the area around the unknown test case, assesses the quality of the predicted PVT data. Finally, a combined approach was used by training an ANN to provide, for each test case input, the appropriate parameters of a conceptual deterministic model for the accurate prediction of GOR and Bo based on an input provided in real time by bottom hole tool measurements.
The three above MLTs have been successfully implemented in commercial software packages which are utilized worldwide by the oil & gas industry.