Data Pre-Conditioning For Predictive And Interpretative Algorithms:Often Ignored But Extremely Important

Frost, Elton (Baker Hughes, Inc) | Quinn, Terrence (Baker Hughes, Inc)



 Much has been written about methods for estimating and interpreting log measurements. These methods are highly dependent on the quality of the original acquired data sets. Wireline and logging-while-drilling (LWD) technologies have advanced to a level where today’s analysts frequently assume the acquired measurements are correct, unless problems are encountered when integrating the data. The assumption is generally valid, but starts to fail when conditions within the borehole being surveyed degrade to the point of falling outside the physical measurement limitations of the instruments. Data reconstruction/estimation can take many forms including translation applications of regional trends, transformation of one type of measurement into another type, extrapolation of offset well data to the well of interest, use of offset openhole data combined with cased hole data in predicting the measurements on adjoining wells when only the cased hole data is available, extraction of measurements from seismic, etc.. Methods involved in these endeavors vary from empirical algorithms, to regional trend analysis, to statistical inference and to neural nets. Successful application of any method requires data that are representative of the formations when acquired under optimal conditions. Interpretation algorithms applied to the data are no different, in that data quality is assumed for the analysis models to function correctly. Proprietary internal and third-party external interpretation packages have problems when the data quality suffers. Proper data reconstruction requires an understanding of the quality of the acquired data (calibrations, accuracy, etc.), instrument configuration in the toolstring, acquisition methodology and the condition of the wellbore environment when the data was acquired. We will examine the application of several methods used in pre-conditioning data and data reconstruction, along with some novel statistical methods the authors employ with examples in various environments. Validation of the reconstructed data sets is also demonstrated.

Duplicate Docs Excel Report

None found

Similar Docs  Excel Report  more

None found