تخمين نمودارهاي پتروفيزيكي و نفوذپذيري نسبي برداشت نشده مخزن آسماري ميدان اهواز با استفاده از نمودارهاي پتروفيزيكي و نتايج آزمايشات مغزه موجود با روشهاي آماري، شبكه عصبي و نروفازي

چكيده انگليسي :

Estimating Well petrophysical logs of Asmari Reservoir Formations from Ahwaz Oil Field Using Available Petrophysical Well Logs and Core Sample Tests by Statistical Neural Network and Combined Hybridic Neuro Fuzzy Methods Leila Moghadasi Leila moghadasi@gmail com Department of Mining Engineering Isfahan University of Technology Isfahan 84156 83111 IranDegree M Sc Language FarsiSupervisor s Name and Email addressNader Fathian PourJamshid MoghadasiAbstractEstimating reservoir petrophysical parameters such as porosity permeability water and oil saturation is vital to anyin situ hydrocarbon reserveevaluation In the case of highly heterogeneous reservoirs common states of mostIranian oil reservoirs prediction of these parameters is not only complicated but in some cases is a seriouschallenge to the upstream oil industries The conventional means of estimating these parameters is measuring them via coresamples taken from wells in thelaboratory Although laboratory methodsof estimating petrophysical parametersis still a common practice but theproblem of being costly and moretime consuming is not significantlyresolved yet With the development of newgeneration well logging tools and systems the cost of estimating these parameters through well log data hassignificantly decreased However due to being indirect estimator and dependent on some other environmentalvariables it is a necessity to employ state of the art intelligence methods of estimating and predicting techniques toimprove their accuracy and effectiveness compared to measurements made on core samples Determination oflithology and petrophysical parameters of reservoirs using different well logs has become the central part ofanypetroleum exploration and development program in oil and gas upstream industry In the current study thehorizontal porosity and permeability as the most important petrophysical parameters ofAsmari reservoir formation inAhvaz oil field have been estimated using fourversions of intelligent techniques called conventional artificial neuralnetworks artificial neural networks based on PCA transformation statistical NeuralNetworks based onBootstrapping and combined neural and fuzzy logic approach called neurofuzzyinference networks The Ahvaz oil field is formsan anticlinestructure striking parallel to the Zagros rangesand is located in the south tosouthwest of Dezful Embayment extending 67 by 6km in length and width respectively Data preprocessing including removing outlier data and multivariate statistical analysis between input and outputdata were carried out in order to understand the variational behavior and patterns among multivariate data Well logdata includinglog depth caliper conductivity sonic natural gamma density and neutron on one hand and estimatedparameters as water saturation percent of shale volume and type of lithology as inputs on other hand as the input ofneural network predictors of drilled core horizontal porosity and permeability taken from a total number of 19exploratory cored wells have been usedto design train and optimize different artificial neural networks explainedearlier Results show that neural networks with a structure of one hidden layer are more effective than networks withmultiple hidden layers The optimum network parameters were found to employ TanSigas activation function TrainLMas training function and LearnGDMas learning function Also it was found that the optimal network forestimating permeability should include11 inputs with an overall correlation fit of 93 in the training phase and 77 in the validation stage with 24 neurons in one hidden layer Also the optimal network for estimatingporosity shouldinclude 9 input logswith 22 neurons in the hidden layer resulting in 75 correlation fit in training phase and 71 5 in the validation process Finally and through comparing results obtained from above mentioned four networksindicates that the estimates obtained from statistical neural network based on bootstrapping approach outperfor

مقدسي، ليلي

تخلخل , شبكه هاي عصبي مصنوعي كلاسيك , ميدان نفتي اهواز , شبكه هاي عصبي تركيبي