There is a growing number of oil production wells in the world that use rod pump units as an extraction system. In fact, this lifting method is become the preferred one for unconventional wells that are producing at the late-stage period, yet with still attractive rates of a few hundred bopd. Dynamometry basically consists of the visual interpretation of the shape of the load graph based on the position of the piston of the subsoil pump. This task is carried out by an operator and his experience is used for the correct interpretation that can be contrasted. With additional tests, diagnosis becomes very important because it allows optimizing production by adjusting rest and production times, reducing operation and maintenance costs, avoiding failures and unscheduled stops, but many times there are not enough trained personnel for interpretation. In this context, dynamometry has been refined both in the acquisition and in the interpretation of dynamometric records.
Sustainability Applied to Unconventional Oil and Gas Field Exploration and Development
Top1. Introduction
Mechanical pumping is the most widespread artificial lift system in oil fields and contributes a significant volume to global oil production. For this reason, different techniques have been developed for the diagnosis and optimization of its operation, from the simplest ones such as the manometric test to the most sophisticated ones such as dynamometry. Dy- namometry is the most widely used and accepted technique for the diagnosis of mechanical pumping units, its bases have been developed by Gilbert (1936) and Fagg (1950) more than 50 years ago. This technique is based on the interpretation of the shape of the dynamometric chart, which is the graph of the measured or estimated load versus position on the piston in the subsurface pump. The traditional method of interpretation consists of visual inspection performed by trained and experienced personnel, but many times due to the number of records generated primarily by continuous
With the development of image processing techniques and neural networks, automatic diagnosis methods based on one-to-one architecture and multiclass classification have been implemented. But extracting just one operating condition from a dynamometric chart wastes all the information that the dynamometric chart can provide. In general, most research related to the automatic diagnosis of dynamometric charts report metrics greater than 99% under the one-to-one architecture for operating conditions ranging from 6 to 30 classes. Works such as those carried out by Boguslawski, Boujonnier, Bissuel-Beauvais, Saghir and Sharma (2018) and Nazi, Ashenayi, Lea and Kemp (1994) allow reporting percentage values that indicate the probability of belonging to one or more operating conditions, even though these models have not been trained for the simultaneous diagnosis of several operating conditions. However, they open the possibility of diagnosing conditions simultaneously from a dynamometric chart. The contributions of this research to the task of automatic computer diagnosis are listed below:
- •
In this work, a data set is assembled that is made up of a dynamometric chart as an instance (x) and an objective (y) which is a label that contains all the operating conditions identified in the chart.
- •
Regarding training, the proposed models seek to simultaneously identify all operating conditions in a dynamometric chart, going from one to one to one to one to one to many architecture. In this way, more information is extracted from the dynamometric chart and consequently the operation of the unit.
- •
To evaluate the performance of these multiple classification models, it is proposed to make a modification to the traditional confusion matrix, with the objective of extracting metrics that allow measuring the performance of deep learning architectures.
Top2. Previous Concepts
Concepts such as the basics of the mechanical pumping system, dynamometry and dynamometric charts are exposed.
2.1. Rod pump system
Mechanical pumping is the simplest, most efficient, and adaptable artificial lift system for most types of fluids. For this reason, it is the most widespread oil extraction system in the world. Mechanical pumping is a system that provides the energy necessary to lift the fluid from the bottom of the well to the surface. It consists of three parts: a surface unit, a system of rods that connect the surface installation with the bottom, and a pump. subsoil. Surface equipment has a drive unit which may be a gas or electric motor, a gearbox, counterweights, an outrigger, a “horse head”, reins, a polished bar, etc. In the Figure 1 Conventional type pumping unit shown.
Figure 1. Conventional rod pump unit