Slide 1

ARTIFICIAL INTELLIGENCE (ANN) AND ACCURATE WELL LOG INTERPRETATION:KEY TO ACCURATE GRAVEL PACK DESIGN(A NIGER DELTA CASE STUDY)TOCHUKWU, UGOMUOH THEOPHINE20091645323PETROLEUM ENGINEERINGFUTOOutline

IntroductionLiterature ReviewObjectiveAdvanced Regression AnalysesArtificial Neural NetworkProblem StatementPhase of the solutionMethodologyResultsConclusion and Recommendation.

IntroductionReview, effects, Causes and Control of sand productionConventional Gravel sizing criteriaWhy look for alternatives?Are there viable alternatives? Yes!!!Accurate well-log interpretationArtificial Intelligence (ANN)Literature ReviewTimur (1968): Permeability, porosity and water saturation relationship for sandstone reservoirs.

Sandro Randovanovic and Milan Radojii, Predicting the results of football matches using the neural network.

Baldwin et al, Computer Emulation of human Mental Processes: Application of Neural Network simulators to Problems in Well Log InterpretationOBJECTIVE

This work seeks to bypass the conventional sieve analysis as it applies different advanced regression analyses and correlation, in order to propose a standard for the industry as well as analyze the behavior of the artificial intelligence.Advanced Regression Analyses

OLS:

RMA:

Robust:

Moving Average: Yt = C

Artificial Neural Network

Human MechanismComputerized Mechanism

Problem statementin order to accurately design gravel pack, knowledge of the formation grain size is needed. Therefore, this study seeks to provide unconventional means to formation grain size determination. Phases of the solutionPhase 1: Accurate Interpretation Of Permeability From Porosity And Water Saturation Data via Comparative Regression Analyses and Correlations

Phase 2; Experiment 1: Implementation Of The Artificial Neural Network With An Incomplete Set Of Small Distorted Data Points

Phase 2; Experiment 2: Implementation Of The Artificial Neural Network With An Incomplete Set Of Large Data Points

Phase 3: Implementation Of The Artificial Neural Network With A Large Complete Set Of Data Point

Methodology: Phase 1

Y = A + B * X1 + C * X2Multivariate Linear RegressionDATAFIT 9.1K = A * B / Scwr) (Timur, 1948)PLOTDATA GATHERINGMS EXCELMethodology: Phase 2; Experiments 1 &2, and Phase 3DATA GATHERINGMS EXCELINPUT TRAINING PARAMETERSCREATE NETWORK

TRAIN NETWORK

GENERATE PERFORMANCE & REGRESSON PLOTIF R > 0.93

STOP TRAININGELSE RETRAINIF OUTPUT COINCIDESTOP TRAININGCONTINUE ITERATIVELYCREATE NEW NETWORKALTER NEURONS AND LAYERSRETRAIN, ITERATE UNTIL R > 0.93Results: Phase 1)...................................................................................................................4.1SOURCE: DATAFIT 9.1 & PLOT

Y = 4007.27( Accurate Interpretation Of Permeability From Porosity And Water Saturation Data via Comparative Regression Analyses and Correlations

GRAPHS GENERATED FROM PLOT STATISTICAL SOFTWAREPhase 2: Experiment 1

Implementation Of The Artificial Neural Network With An Incomplete Set Of Small Distorted Data PointsPerformance and Regression Plot of Phase 2: Experiment 1

Phase 2: Experiment 2Implementation Of The Artificial Neural Network With An Incomplete Set Of Large Data Points

Performance and Regression Plots of Phase 2: Experiment 2

Phase 3Implementation Of The Artificial Neural Network With A Large Complete Set Of Data Points

Phase 3Implementation Of The Artificial Neural Network With A Large Complete Set Of Data Point

ConclusionFrom the results of phase 1, it has been proven that the optimum porosity-water saturation relationship for the prediction of permeability for any field is the correlation generated based on the data got from that field and the RMA Regression fits any data point best. (Which supports Timurs (1968) work).From the results of phase 2; experiment 1 and experiment 2, and phase 3, which comprises of the neural network training with complete, incomplete, small and large set of data points, it can be inferred that an optimum training/convergence can be obtained if the input training sets are large, irrespective of the fact that some data points might be missing.The result of Phase1, depicts the high level of intelligence of this machine network so that even though its result has some measure of error, its training speed, accuracy and reading rate emphasizes its time saving, and cost effective advantages.RecommendationsPhase one is a work in progress. In the statistical world, there exists various regression analysis tool of which only 4 was used in this work. Investigation into the use of other regression analysis and the measure of their accuracy should be research. Also, build your own correlation and test wit other available correlations. Further work can be done on the use of:Generalized regressionANCOVAANOVA etc.From the ANN training conducted, an investigation can be conducted to analyze the behavior of the artificial intelligence in training a small set of complete data and a small set of complete but highly disjointed data points. THANK YOU