Summary In this paper, deep learning artificial neural networks (ANNs) are used to analyze the features of downhole dynamometer cards and identify the sucker rod pumping system conditions. A description model for the dynamometer cards, using Fourier descriptors, was established for card feature extraction. Then, neural networks were trained to generate failure prediction models to recognize downhole faults of the rod pumping systems. The failure prediction models were validated and tested with a large database of previously interpreted cards. The proposed model is trained by using 4,467 dynamometer cards—29.2% of these cards represent sucker rod pumping systems of normal conditions, while the rest (70.8%) represent faulty sucker rod pumping systems. Genetic algorithms (GAs) were used to search for the best deep ANN structure that gives highest accuracy for the testing data. Accuracy of the proposed ANN model was measured with 1,915 cards that were not used in developing the ANN. The proposed model identified the sucker rod system failure successfully with very high accuracy (99.69%).
The objective of this work is to further explore the potential application of Machine Learning algorithms in production prediction and ultimate recovery. Intelligent Machine Learning Approaches such as Gradient Boosted Trees (GBT), Adaboost, and Support Vector Regressor (SVR) are applied to detect the most important features contributing to cumulative production prediction within the first 12 producing months. The models are applied on a data set composed of 5 wells in the Volve field in the North Sea. The collected data was then filtered and used to structure and train the different Regression algorithms and fine tune the appropriate hyperparameters. The different models were All models were evaluated by measuring the Mean Absolute Error (MAE). The generalization and precision performance of the proposed models are established by comparing the forecasted outcome after cross validation with field data. The optimized model can predict production response with high accuracy. The data-fitting process comprises of splitting the data into training using 70% of the data set, 15% validation, and 15% testing. Constructing a regression model on the training set and validating it with the test set. Recurrent application of a "cross-validation" process produces important information concerning the robustness of any regression-modeling method. Six parameters were considered as input factors to build the model. Factors affecting production prediction included on stream hours, average choke size, bore oil volume, bore gas volume, bore water volume, average wellhead pressure were used as input. The outcome showed that the developed model provided better prediction compared to analytical models with a 11.71% MAE prediction for SVR. This novel data mining application could be trained on any dataset to help predict future production performance at any conditions in any given scenario.
This study presents a novel data-driven approach for calculating multiphase flow rates in electrical submersible pumped wells. Traditional methods for estimating flow rates at test separators fail to identify production trends and require additional costs for maintenance. In response, virtual flow metering (VFM) has emerged as an attractive research area in the oil and gas industry. This study introduces a robust workflow utilizing symbolic regression, extreme gradient boosted trees, and a deep learning model that includes a pipeline of convolutional neural network (CNN) layers and long short-term memory algorithm (LSTM) layers to predict liquid rate and water cut in real time based on pump sensors' data. The novelty of this approach lies in offering a cost-effective and accurate alternative to the usage of multiphase physical flow meters and production testing. Additionally, the study provides insights into the potential of data-driven methods for VFM in electrical submersible pumped wells, highlighting the effectiveness of the proposed approach. Overall, this study contributes to the field by introducing a new, data-driven method for accurately predicting multiphase flow rates in real time, thereby providing a valuable tool for monitoring and optimizing production processes in the oil and gas industry.
Electrical submersible pumps (ESPs) are considered the second-most widely used artificial lift method in the petroleum industry. As with any pumping artificial lift method, ESPs exhibit failures. The maintenance of ESPs expends a lot of resources, and manpower and is usually triggered and accompanied by the reactive process monitoring of multivariate sensor data. This paper presents a methodology to deploy the principal component analysis and extreme gradient boosting trees (XGBoosting) in predictive maintenance in order to analyze real-time sensor data to predict failures in ESPs. The system contributes to an efficiency increase by reducing the time required to dismantle the pumping system, inspect it, and perform failure analysis. This objective is achieved by applying the principal component analysis as an unsupervised technique; then, its output is pipelined with an XGBoosting model for further prediction of the system status. In comparison to traditional approaches that have been utilized for the diagnosis of ESPs, the proposed model is able to identify deeper functional relationships and longer-term trends inferred from historical data. The novel workflow with the predictive model can provide signals 7 days before the actual failure event, with an F1-score more than 0.71 on the test set. Increasing production efficiencies through the proactive identification of failure events and the avoidance of deferment losses can be accomplished by means of the real-time alarming system presented in this work.
Steam injection is a popular technique to enhance oil recovery in mature oil fields. However, the conventional approach of using a constant steam rate over an extended period can lead to sub-optimal performance due to the complex nature of the problem and reservoir heterogeneity. To address this issue, the Markov decision process can be employed to formulate the problem for reinforcement learning (RL) applications. The RL agent is trained to optimize the steam injection rate by interacting with a reservoir simulation model and receives rewards for each action. The agent’s policy and value functions are updated through continuous interaction with the environment until convergence is achieved, leading to a more efficient steam injection strategy for enhancing oil recovery. In this study, an actor-critic RL architecture was employed to train the agent to find the optimal strategy (i.e., policy). The environment was represented by a reservoir simulation model, and the agent’s actions were based on the observed state. The policy function gave a probability distribution of the actions that the agent could take, while the value function determined the expected yield for an agent starting from a given state. The agent interacted with the environment for several episodes until convergence was achieved. The improvement in net present value (NPV) achieved by the agent was a significant indication of the effectiveness of the RL-based approach. The NPV reflects the economic benefits of the optimized steam injection strategy. The agent was able to achieve this improvement by finding the optimal policies. One of the key advantages of the optimal policy was the decrease in total field heat losses. This is a critical factor in the efficiency of the steam injection process. Heat loss can reduce the efficiency of the process and lead to lower oil recovery rates. By minimizing heat loss, the agent was able to optimize the steam injection process and increase oil recovery rates. The optimal policy had four regions characterized by slight changes in a stable injection rate to increase the average reservoir pressure, increasing the injection rate to a maximum value, steeply decreasing the injection rate, and slightly changing the injection rate to maintain the average reservoir temperature. These regions reflect the different phases of the steam injection process and demonstrate the complexity of the problem. Overall, the results of this study demonstrate the effectiveness of RL in optimizing steam injection in mature oil fields. The use of RL can help address the complexity of the problem and improve the efficiency of the oil recovery process. This study provides a framework for future research in this area and highlights the potential of RL for addressing other complex problems in the energy industry.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.