Knowledge Discovery in Databases and Multiphase Flow Metering: The Integration of Statistics, Data Mining, Neural Networks, Fuzzy Logic, and Ad Hoc Flow Measurements Towards Well Monitoring and Diagnosis

Abstract

The usual approach to the interpretation of producing wells is based on mechanistic models for the simulation of steady state and transient flow regimes. However, there are significant reservations about convergence problems, computational limits, the need for extensive tuning on field data, the instability of boundary conditions, the limited applicability of existing multiphase flow models, and the uncertainties associated with choke valve models. The current industry standards are critically reviewed within this framework. The real-time monitoring of producing wells is recognised as the best way of optimising field performance. Monitoring a producing well implies the ability to track, in real-time, any changes in fluid composition, flow rates, or pressure and temperature profiles. Multiphase Flow Metering (MFM) plays a key role in this scenario. Such information, combined with the critical analysis of historical data from the well itself or from analogue wells, allows diagnosis of the system and prediction of future trends. However, field data per se' do not necessarily generate knowledge. This is particularly true for large databases, which are difficult to manipulate to provide suitable inputs for wellbore simulators. This paper suggests how MFM, Knowledge Discovery in Databases (KDD) and Fuzzy Logic (FL) can offer an alternative approach to the analysis of producing wells. KDD is the automated extraction of patterns representing knowledge implicitly stored in large information repositories. Distributed, ad-hoc field measurements (including MFM and downhole measurements) can be processed via data cleaning, data integration, data mining, artificial intelligence, and pattern evaluation. FL can then manage the resulting information in terms of flow assurance and production optimisation. The same techniques can also be extended to the reservoir and the production network, for an integrated approach to production system analysis

A Hybrid Artificial Intelligence Method for the Optimization of Integrated Gas Production System

“…The conventional stochastic programming mainly addresses randomness of uncertain parameters with discrete or continuous probability functions. However, uncertainty associated with gas production system is not random but rather a matter of the imprecise description of information collected from the gas field 6 . In this design problem, fuzzy mathematical programming can be a more efficient and flexible tool to quantitatively accommodate uncertainty in the presence of vague information.…”

Application research of data mining on reservoir characterization

“…A lot of works in terms of oil description could be regarded as problem of classification and regression, that is why data mining is viewed as one of the ten core techniques in development of oil industry 1 J . Data mining is successfully applied into dynamic monitoring [4] of oil, best identification of practical work in petroleum engineering^ and reservoir characterization [6 ] in foreign countries.…”

Application of Data Mining on Reservoir

“…A lot of works in terms of oil description could be regarded as problem of classification and regression, that is why data mining is viewed as one of top core techniques in development of oil [2]. Data mining is successfully applied into dynamic monitoring of oil [3], best identification of practical work in petroleum engineering [4] and reservoir characterization [5] in foreign countries.…”