A fundamental parameter in the exploration and development of unconventional shale reservoirs is total organic carbon (TOC). To achieve reliable TOC values, it requires a labour intensive and time-consuming laboratory experiment. On the other hand, models have been proposed using geophysical well logs as input variables with little attention paid to the contribution of mineralogical parameters in the evaluation of TOC. In this paper, a novel stacking machine learning technique is examined to generate accurate TOC predictions from the mineral content of the shale rock in the Sichuan Basin. The stacking machine learning model involves first-level models of multivariate adaptive regression spline (MARS), random forest (RF) and gradient boosted machine (GBM) known as base learners, while MARS was further used in the next step as the meta learner model. The research result indicated that the stacking TOC model outperformed the single applied models of MARS, GBM and RF. The proposed stacking TOC model generated estimates having the least error statistics of 0.29, 0.54 and 0.54 for MSE, RMSE and MAPE respectively while producing the highest correlation of 0.86 during the model validation stage. Therefore, stacking machine learning approach permits an improved estimation of TOC from the mineralogy of the rock.
From upstream to downstream, all phases are equally important when developing an oil field. To extract the hydrocarbons from the reservoir in a controlled manner, a connection between the wellbore and the fluid contained in the formation is needed. The most common way to make this connection is by perforating tunnels along the cemented metal casing and the rock behind it. These channels will allow the formation fluid to flow to the wellbore and consequently to the surface. The characteristics of these tunnels are relevant for production and need to be carefully chosen. Not selecting the right perforating system and or technique will affect the variables that dictate the production rate. In this paper, well perforation techniques were reviewed. Perforating parameters, such as depth of penetration, tunnel clean-up, entrance hole, and pressure balance, among many others, can have a severe impact on the final production results. We will demonstrate that the challenges faced by the reservoir engineers, during the designing phase, in choosing among the existing options are just the beginning of the integrated and collaborative workflow. The balance between the "best" perforating technique and the expenses of using it, must be carefully analyzed. Usually, the value difference between the charges, guns, and service providers is the smallest portion of the entire budget. This factor makes the choice of the gun system a multi-disciplinary matter.
Reservoir geomechanics has already proven to play an important role in reservoir management studies. However, the computational costs of these studies usually hinders a thorough evaluation of geomechanical effects. In this paper, we present the development of a massively parallel, multiscale reservoir geomechanics simulator currently in use on industry-grade geomechanical studies. The viscoplastic formulation allows for accurate modelling of the geomechanics effects, at the same time that results on an efficient computational model. The massively parallel distributed memory implementation of a linear system framework takes the most advantage of high performance computing (HPC) infrastructures, making use of clusters of multicore nodes. A Preconditioned Conjugate Gradient solver is implemented based on this framework. An additive, coarse-space preconditioner, based on the Multiscale Finite Element (MSFE) method, allows for an efficient, fit-for-purpose, linear system solution strategy. Because the viscoplastic formulation results on a symmetric system matrix that does not change across the simulation in time (only the right hand side does), the MSFE basis-function can be built only once, hence avoiding expensive computations. Our reservoir geomechanics simulator is capable of simulating stress and strain behaviour on large real field, geologically complex, case applications, e.g. from the Brazilian pre-salt. We present different studies, which involve the investigation of geomechanical effects, namely, subsidence, thermomechanics and cap rock integrity. In these studies, we demonstrate the scalability of the simulator in real field models with up to almost 100 million elements, running on up to more than 600 computing cores. The usage of state-of-the-art simulation approaches, combined with modern HPC strategies, enables reservoir geomechanics studies which were once hampered by computational limitations. This implementation will also allow for even more computationally intensive workflows which require many simulations, e.g. uncertainty quantification.
In this paper, we revisit the issues related to dynamics and mechanisms of the pump jack as a surface unit of the sucker rod pumping systems. In particular, we employ both the traditional study methods based on trigonometric relationships and current Newton-Raphson iteration based general solution techniques for the sets of nonlinear governing equations of angles within the four-bar linkages. The nonlinear computational methods enable the study of transient behaviors of pump jacks with non-uniform motor speeds and further analytical studies of the positions, velocities, and accelerations as well as the polish rod force and motor torque in comparison with the actual experimental measures documented in Echometer TAM software and other case studies. These current nonlinear approaches have paved a way for more in-depth studies of the entire downhole system as a specific viscoelastic dynamical system with respective stiffness and damping coefficients for individual wells. In this study, we also identify a common misunderstanding about the structural unbalance B as illustrated in API Specification 11E Page 47 which must be multiplied by the sine function of the polar angle of the Walking beam measured from its horizontal position. This key finding has been confirmed and validated with the comparison between results from both traditional and proposed nonlinear approaches and experimentally measured data collected by Echometer equipment and TAM software.
Stuck-pipe phenomena are relatively rare in drilling operations in the oil & gas industry, but can have disastrous economic consequences, causing costly time delays and sometimes even the loss of expensive machinery. In this work, we develop an event-based prediction model that relates the occurrence of precursor events to the stuck-pipe phenomena. To this aim, the detectors of various types of precursor events that typically anticipate stuck-pipe occurrences are first designed based on the available mudlog data. A Hidden Markov Model (HMM) is then developed to relate these precursor events to actual drilling problems, producing different levels of alarm, with the ultimate goal of predicting stuck pipes. The model has been tested on a dataset of wells with different characteristics, showing positive results.
