Layer | Fill | Outline |
---|
Map layers
Theme | Visible | Selectable | Appearance | Zoom Range (now: 0) |
---|
Fill | Stroke |
---|---|
Collaborating Authors
Results
Intelligent Prediction of Differential Pipe Sticking by Support Vector Machine Compared With Conventional Artificial Neural Networks: An Example of Iranian Offshore Oil Fields
Jahanbakhshi, Reza (Young Researchers and Elites Club, Science and Research Branch, Islamic Azad University,Tehran, Iran) | Keshavarzi, Reza (Young Researchers and Elites Club, Science and Research Branch, Islamic Azad University,Tehran, Iran) | Aliyari Shoorehdeli, Mahdi (K.N. Toosi University of Technology) | Emamzadeh, Abolqasem (Islamic Azad University)
Summary Differential pipe sticking (DPS) is one of the most conventional and serious problems in drilling operations that imposes some extra costs to companies. This phenomenon originates mainly from improper mud properties, bottomhole assembly (BHA) (contacting area), still pipe time, and differential pressure between the formation and the drilling mud. Investigation on various conditions that lead to DPS makes it possible to develop some preventive treatments to avoid this problem's occurrence. In the past, statistical methods were applied in this area, but recently artificial neural network (ANN) approaches are frequently being used. ANNs have some priorities over conventional statistical methods such as the model-free form of predictions, tolerance to data errors, data-driven nature, and fast computation. On the other hand, the designed ANNs have some shortcomings and restrictions as they are developed to predict problems. In this paper, to solve most of the existing disadvantages of ANNs, a novel support-vector machine (SVM) approach has been developed to predict a DPS occurrence in horizontal and sidetracked wells in Iranian offshore oil fields. The results from the analysis have shown the potential of the SVM and ANNs to predict DPS, with the SVM results being more promising.
- North America > United States > California (0.28)
- North America > United States > Texas (0.28)
- Asia > Middle East > Saudi Arabia > Arabian Gulf > Arabian Basin > Arabian Gulf Basin > Foroozan Field (0.93)
- Asia > Middle East > Iran > Lavan Island > Arabian Gulf > Arabian Basin > Arabian Gulf Basin > Salman Field (0.93)
- Asia > Middle East > Iran > Arabian Gulf > Arabian Basin > Arabian Gulf Basin > Sirri Field > Sirri D Field > Shu'aiba Formation (0.93)
- (4 more...)
Abstract Stuck pipe has been recognized as one of the most challenging and costly problems in the oil and gas industry. However, this problem can be treated proactively by predicting it before it occurs. The purpose of this study is to implement the two most powerful machine learning methods, Artificial Neural Networks (ANNs) and Support Vector Machines (SVMs), to predict stuck pipe occurrences. Two developed models for ANNs and SVMs with different scenarios were implemented for prediction purposes. The models were designed and constructed by the MATLAB language. The MATLAB built-in functions of ANNs and SVMs, and the MATLAB interface from the library of support vector machines were applied to compare the results. Furthermore, one database that included mud properties, directional characteristics, and drilling parameters has been assembled for training and testing processes. The study involved classifying stuck pipe incidents into two groups - stuck and non-stuck - and also into three subgroups: differentially stuck, mechanically stuck, and non-stuck. This research has also gone through an optimization process which is vital in machine learning techniques to construct the most practical models. This study demonstrated that both ANNs and SVMs are able to predict stuck pipe occurrences with reasonable accuracy, over 85%. The competitive SVM technique is able to generate generally reliable stuck pipe prediction. Besides, it can be found that SVMs are more convenient than ANNs since they need fewer parameters to be optimized. The constructed models generally apply very well in the areas for which they are built, but may not work for other areas. However, they are important especially when it comes to probability measures. Thus, they can be utilized with real-time data and would represent the results on a log viewer.
Rock Squeezing Prediction By a Support Vector Machine Classifier
Shafiei, A. (Department of Earth & Environmental Sciences, University of Waterloo) | Parsaei, H. (Department of Systems Design Engineering, University of Waterloo) | Dusseault, M.B. (Department of Earth & Environmental Sciences, University of Waterloo)
ABSTRACT: Redistribution of in situ stresses so they exceed rock strength leads to yielding of the intact rock material around a tunnel after excavation, causing large plastic deformations often referred to as ground squeezing. This tunneling problem typically occurs during deep tunneling in weak rock such as shales and weak schists where volumetric dilatancy accompanies the process of rock yield and deterioration. In this article, a decision support system to assist a tunnel engineer in making a decision on tunnel route design, selection of excavation technique or mitigation measures is presented. A support vector machine-based supervised classifier is proposed which employs the Q tunneling index and depth of the tunnel to predict risk of rock squeezing. Performance analysis using extensive field data obtained from several tunnels around the world indicated that the developed system is more accurate than heuristic systems currently in use. The proposed system provides a posterior probability as a support for the decision being made that can be used to assess the acceptability level of the prediction. 1 INTRODUCTION Redistribution of in situ stresses so they exceed rock strength leads to yielding of the intact rock material around a tunnel after excavation, causing large plastic deformations often referred to as ground squeezing. This tunneling problem typically occurs during deep tunneling (depth > 200m) in weak rock such as shales and weak schists where volumetric dilatancy accompanies the process of rock yield and deterioration. For example, squeezing ground conditions are reported in the Taloun tunnel in Iran, the Bolu tunnel in Turkey and tunnel projects in Japan, Venezuela, Germany, Austria, Switzerland, India, Nepal and the USA. A number of parameters and criteria have been proposed in the literature to assess ground squeezing potential, especially for deep tunnels designed in weak rock masses.
- Europe (1.00)
- North America > United States (0.88)
- Asia > Middle East > Iran (0.37)
- (2 more...)
- Geology > Geological Subdiscipline > Geomechanics (1.00)
- Geology > Rock Type > Sedimentary Rock > Clastic Rock > Mudrock > Shale (0.45)
Abstract In waterflood management, there exists several models to describe a petroleum reservoir for predicting the future production rates using scheduled injection rates. Most of them have the ability to estimate how much the injectors impact some specific producers, namely, the interwell connectivities between the injectors and the producers. Knowing these values not only reduces the cost of water injection, but can also increase the oil production. In this paper, we construct four different models for the interaction between a group of injectors and a producer and then dynamically estimate the parameters of these models, along with the interwell connectivities using an Iterated Extended Kalman Filter (IEKF) and Smoother (EKS). We then use the Generalized Choquet Integral (GCI) to aggregate the estimated interwell connectivities. The GCI is optimized to minimize mean-square errors in future forecasted production rates. This is done by using Quantum Particle Swarm Optimization (QPSO) to search for the optimal set of fuzzy densities which are required by the GCI. Several experiments are conducted to show the improved average performance of our approach on a set of data from a real reservoir. I. Introduction Flooding an oil field with extraneous water has been a widely accepted method for increasing a reservoir's oil recovery since the 1950's. Water is injected into dedicated injection wells strategically located throughout the reservoir, in order to displace the remaining oil towards the producing wells. If properly designed and operated, a waterflood can double the reservoir's oil recovery. In almost all waterflood operations, measured injection and production rates are the most abundant available data. They are considered to be correlated to each other in some very complicated way, and many methods have been previously proposed to infer the interwell connectivities (referred as the "Injector-Producer-Relationship (IPR)?? in this paper) between each producer and its surrounding contributing injectors using only these data. In all those works, the reservoir is modeled as a dynamical system in which the injection rates act as the system's inputs and the production rates are the system's outputs. Heffer et al. [7] used Spearman rank correlations to relate injector-producer pairs and associated these relations with geomechanics. Panda and Chopra [18] used artificial neural networks to determine the interactions between injection and production rates. Albertoni and Lake [1] estimated the interwell connectivity based on a linear model using a multiple linear regression (MLR) method. Yousef et al. [28], [29] improved this work by building a more complex model, named "capacitance model," to describe the relationship between injection and production rates. Lee [12] further generalized the capacitance model to the Distributed Capacitance Model (DCM) by taking into account that the reservoir between some injector-producer pairs is highly heterogeneous or includes some high permeability channels, fractures or faults. Liu and Mendel [13] modeled the reservoir using continuous impulse responses which were characterized as a two-parameter auto-regressive model between each single injector and a single producer. All of these approaches not only can estimate the IPR values using only measured injection and production data, but can also predict future production rates given future scheduled injection rates; however, different approaches focus on different aspects of modeling the reservoir and utilize different estimation methods to estimate the model's parameters. For example, Lee's DCM basically was derived from a total mass balance with compressibility (the same as the capacitance model) and then its parameters were estimated using constrained quadratic programming. Liu and Mendel's model, referred as the LMM in this paper, was developed based on domain expert knowledge, and its parameters were estimated using the Extended Kalman Filter (EKF).
- North America > United States > California (0.68)
- Asia (0.68)
- Research Report (0.46)
- Overview (0.46)