|Theme||Visible||Selectable||Appearance||Zoom Range (now: 0)|
Abstract Drilling a directional well becomes an essential process in the oil and gas industry to ensure better reservoir exposure and less wellbore collision risk. In the high-volume drilling market, cost-effective mud motors are dominant. The motor is capable of delivering the desired well curvature by switching between rotating and sliding operations. Therefore, to follow a predefined well trajectory, it is a critical mission to determine the optimal operation control sequence of the motor. In this paper, a method of training an automatic agent for motor directional drilling using the deep reinforcement learning approach is proposed. In designing the method, motor-based directional drilling is framed into the reinforcement learning with an automatic drilling system, also known as an agent, interacting with an environment (i.e., formations, wellbore geometry, equipment) through choices of controls in a sequence. The agent perceives the states such as inclination, MD, TVD at survey points and the planned trajectories from the environment, and then decides the best action of sliding or rotating to achieve the maximum total rewards. The environment is affected by the agent's actions and returns corresponding rewards to the agent. The rewards can be positive (such as drilling to target) or negative (such as offset distance to the planned trajectory, cost of drilling, and action switching). To train our agent, currently, a drilling simulator in a simulated environment is created with layered earth model and BHA directional responses in layers. Other attributes of the drilling system are assumed to be constant and handled automatically by the simulator. The planned trajectory is also provided to the agent while training. The directional-drilling agent is trained for thousands of episodes. As a result, the agent can successfully drill to target in this simulated environment through the decisions of sliding and rotating. The proposed workflow is known as the first automated directional drilling method based on deep reinforcement learning, which makes a sequence of decisions of rotating and sliding actions to follow a planned trajectory.
Amr, Salma (Raisa Energy) | El Ashhab, Hadeer (Raisa Energy) | El-Saban, Motaz (Raisa Energy) | Schietinger, Paul (Raisa Energy) | Caile, Curtis (Raisa Energy) | Kaheel, Ayman (Raisa Energy) | Rodriguez, Luis (Raisa Energy)
This paper proposes a set of data driven models that use state of the art machine learning techniques and algorithms to predict monthly production of unconventional horizontal wells. The developed models are intended to forecast both producing locations (PLs) and non-producing well locations (NPLs). Furthermore, results of extensive experiments are presented that were conducted using different methodologies and features combinations. Results are measured against conventional Arps's decline curve analysis showing significant boost in prediction accuracy for both NPLs and PLs. The most accurate model outperforms Arps's-based estimates by almost 23% for NPLs and 36% for PLs. Results also show that using data from multiple basins in training models for another basin yields gains in accuracy, especially for basins with relatively small data.
Abstract This paper describes the processes involved in the developing of a learning culture, and implementing organizational learning software as a pilot project. The chosen location for the pilot project was Chevron Niugini's Drilling Department at the Iagifu Ridge Camp in the Southern Highlands of Papua New Guinea. The Pilot was part of a larger joint effort by Mobil, BP, Texaco and Chevron (MoBPTeCh) to understand how to more effectively capture and share experience in well operations. Over a period of six months, Organizational Learning System software was customized to suit the unique drilling operations in Papua New Guinea. To accommodate the system, a cultural change, relative to "learning", was proposed for the office and the rigsite personnel. It was necessary that the ownership of becoming a learning organization was accepted by all personnel involved. The rigsite is the initial location where an organization will determine if the well plan requires modification. The real- time rigsite understanding, verification and assurance of the plan changes is the driver for continued learning. Subsequently, incorporating lessons into future well planning is essential for anticipated improvement. Both of these actions have been crucial to the success of the implementation of the system. All changes and decisions are documented in the Organizational Learning System, with future well plans, created in the office, reflecting these documented alterations. This paper examines the keys to obtaining a learning culture both in the office and at the rigsite, as well as highlighting other benefits obtained by implementing the Organizational Learning System software at Chevron Niugmi. P. 101
Invisible lost time (ILT) has been estimated to contribute significantly to well construction costs. Reduction of drilling connection time can lead to significant savings, especially in offshore drilling projects. The overall time of a drilling connection can easily be measured using conventional sensors on the rig. However, to gain further insight into the source of operational inefficiencies it is beneficial to breakdown the overall drilling connection process into smaller sub-processes and quantify the time spent doing each. Current sensor technology does not allow us to easily measure these sub-processes on the rig. In this paper, we outline a deep learning-based computer vision method to analyze and measure such sub-processes within the overall drilling connection process using a real-time video feed. Video of the drilling connections is streamed in real-time using a novel IT framework. In a proof of concept exercise, we applied image recognition techniques to enable us to breakdown and classify the sub-processes during the drilling connections. Convolutional neural networks (CNNs) - a deep learning algorithm - are used to analyze these video images frame by frame to identify rig activities. The workflow includes an analytics pipeline to capture video on the rig, transfer of data to the office, video recognition analysis, tabulation of results, and delivery of this result to a graphical interface for cognitive analytics. We performed experiments wherein CNNs were used on the video images to diagnose rig activities with good accuracy. Incorporating the temporal information using a recurrent network has been shown to further improve analysis accuracy. The described workflow is a suggested method for automating the detection of inefficiencies on drilling rigs using rig floor video and machine learning. For the demonstration case, we successfully analyzed video captured on an offshore rig and derived high value useful information using machine learning. This video analysis information can be further combined with conventional drilling sensor data to boost accuracy to identify and confirm rig activities and to quantify invisible lost time during each connection. The presented techniques demonstrate the potential of video analytics as a reliable, low-cost means to effectively "micro-analyze" various rig operations with the aim to identify and replicate best composite performance.
The pdf file of this paper is in Russian.
Drilling of oil and gas wells is a basic process in production of hydrocarbons. Up to now drilling projects are often being designed on the base of segmental data and primary analysis of drilling experience, which is ineffective under modern drilling conditions. A possible solution of this problem consists in computer analysis of data on earlier drilled wells with the use of modern information technologies. Authors propose methods for prediction of probable troubles occurring while drilling new wells on the base of information on well stock of the oilfield using efficient data mining tools – artificial neural networks. Structural and parametric identification of a neural network solving the formalized problem of functions approximation has been conducted. Results of computational experiments on real data show effectiveness of software developed.