Oil is an important strategic material and civil energy. Accurate prediction of oil consumption can provide basis for relevant departments to reasonably arrange crude oil production, oil import and export, and optimize the allocation of social resources. Therefore, a new grey model FENBGM(1,1) is proposed to predict oil consumption in China. Firstly, the grey effect of the traditional GM(1,1) model was transformed into a quadratic equation. Four different parameters were introduced to improve the accuracy of the model, and the new initial conditions were designed by optimizing the initial values by weighted buffer operator. Combined with the reprocessing of the original data, the scheme eliminates the random disturbance effect, improves the stability of the system sequence, and can effectively extract the potential pattern of future development. Secondly, the cumulative order of the new model was optimized by fractional cumulative generation operation. At the same time, the smoothness rate quasi-smoothness condition was introduced to verify the stability of the model, and the particle swarm optimization algorithm (PSO) was used to search the optimal parameters of the model to enhance the adaptability of the model. Based on the above improvements, the new combination prediction model overcomes the limitation of the traditional grey model and obtains more accurate and robust prediction results. Then, taking the petroleum consumption of China's manufacturing industry and transportation, storage and postal industry as an example, this paper verifies the validity of FENBGM(1,1) model, analyzes and forecasts China's crude oil consumption with several commonly used forecasting models, and uses FENBGM(1,1) model to forecast China's oil consumption in the next four years. The results show that FENBGM(1,1) model performs best in all cases. Finally, based on the prediction results of FENBGM(1,1) model, some reasonable suggestions are put forward for China's oil consumption planning.
Extensive reviews and cross-comparison studies are essential to analyze the emerging developments in a specific field of research. In the past decade, hydrocarbon exploration and exploitation from the shale reservoirs have been the most discussed and researched area around the globe. A dramatic development in shale formations became the game-changer, especially in the US. On the other hand, Artificial Intelligence (AI) and Machine Learning (ML) are playing an important role in the rapid development in all the industries through automating most of the routine operations.
The oil industry is also getting equal benefits of AI and ML for the reservoir development planning and its operational accuracy through a series of automated systems. For the field development, computerized static and dynamic simulation models are generated based on several Petrophysical and Geomechanical properties gathered through different resources that are quite time-taking and expensive. AI and ML have made this process much easier, faster, and economical by means of learning through uncounted experiences from already explored and developed reservoirs, their rock properties, and the cross-ponding fluid flow behavior under different circumstances and hence, predicts accordingly.
This article provides a comprehensive literature review in the area of AI and ML applications to model Petrophysical and Geomechanical properties using different approaches and algorithms. Also, a systematic publication counts in each field of subject study per year in different literature databases are presented that infect reflects the trending interest in this subject. Finally, multiple AI and ML techniques are discussed in detail which have been tested in the last decade for the sake of achieving higher accuracy in Petrophysical and Geo-Mechanical simulation models.
Accurate prediction of wax deposition is of vital interest in digitalized systems to avoid many issues that interrupt the flow assurance during production of hydrocarbon fluids. The present investigation aims at establishing rigorous intelligent schemes for predicting wax deposition under extensive production conditions. To do so, multilayer perceptron (MLP) optimized with Levenberg-Marquardt algorithm (MLP-LMA) and Bayesian Regularization algorithm (MLP-BR) were taught using 88 experimental measurements. These latter were described by some independent variables, namely temperature (in K), specific gravity, and compositions of C1-C3, C4-C7, C8-C15, C16-C22, C23-C29 and C30+. The obtained results showed that MLP-LMA achieved the best performance with an overall root mean square error of 0.2198 and a coefficient of determination (R2) of 0.9971. The performance comparison revealed that MLP-LMA outperforms the prior approaches in the literature.
Geochemical parameters are useful properties to enhance hydrocarbon exploration certainty. Though, attaining these parameters, for instance total organic carbon (TOC), volatile and residual hydrocarbon (S1 & S2) is a challenge for geologists due to the high cost and time consumption. Therefore, addressing this issue has become an interesting subject for many researchers. As a result, on the ground of conventional well logs, vast kinds of methods, for example, back propagation artificial neural network (BPANN), have been introduced to solve this problem. Implementing these kinds of methods brings scientists tremendous amounts of information related to the richness of organic matter in a meantime. However, the precision of the aforementioned method is inadequate and BPANN is affected negatively by local optimum. Therefore, current study cope with this issue and alleviate the uncertainty, Least Squares Support Vector Machine (LSSVM) and Adaptive-Neuro Fuzzy Inference System (ANFIS) algorithms cooperating with the particle swarm optimization (PSO) were suggested as a suitable method to increase the precision of estimating geochemical factors. The data bank for this research was attained from available sources of Shahejie formation from Bohai bay basin located in China, which consists of geochemical and well logging information. Outputs of this study illustrated that ANFIS-PSO and LSSVM-PSO have a great ability to estimate geochemical parameters. The values of R2 obtained for these two models in order to predict the output parameters of TOC, S1 and S2 are equal to 0.6846 & 0.785, 0.6864 & 0.778, and 0.7343 & 0.8128, respectively. The statistical comparison between these models shows that LSSVM-PSO shows a better performance compared to another model. Also, a new attempt was implemented to evaluate the impacts of input parameters on the outputs and the results of sensitivity analysis suggest that transit interval time had the greatest effect on the output parameters.
In the process of shale gas drilling, geo-steering plays an important role in shale gas drilling. This paper analyzes the constituent elements of shale formation, and selects the most suitable constituent elements of shale formation. A particle swarm optimization algorithm based on improved inertia weight and acceleration factor is proposed to optimize the parameters of support vector machine. The lithology identification model of shale formation is established based on IPSO-SVM. According to the experimental analysis based on the field historical data, the recognition rate of IPSO-SVM is increased by 17.79%, 10.17% and 8.05%, respectively, compared with SVM, GA-PSO and PSO-SVM. In terms of running time, the running time of IPSO-SVM is 13.76s and 9.5s shorter than that of GA-PSO, PSO-SVM, respectively. By comparing the experimental results of different models, IPSO-SVM has the advantages of strong robustness, strong reliability, high accuracy and fast convergence speed. It provides a theoretical basis for precise geo-steering and finding the optimal shale layer.
This study focuses on predicting acoustic and mechanical rock properties using random forest and feed forward neural network models to evaluate the likelihood of developing efficient ways of handling absence of rock properties at offset locations. The Random Forest algorithm was used for direct prediction of the sonic data without considering the depth range of the facies; while Feed forward Neural network was used to predict the sonic data with emphasis on the lithofacies depths. The accuracy of these approaches was used in choosing the best and the most robust model for predicting sonic data when estimating formation strength and mechnical properties. Acoustic log was predicted after training a combination of caliper log, gamma log, depth, density log and resistivity log from offset wells. 5 hidden layers that accounts for the data structural complexities was included in the model architecture. A multilayer perceptron network was adopted for the Random forest algorithm to handle linear combinations of input data set. Diverse error computations were used to evaluate the performance of the model. Lastly, mechanical properties and sanding potential was evaluated using standard relations and appropriate depositional conditions. Random forest algorithm gave the best prediction accuracy of more than 96%, but the Feed forward network has the lower mean absolute error and mean squared error of 2.75 and 5.93 respectively. Generally, the predicted compressive and shear wave velocity show increase of values with depth, a behavior that is capable of identifying payzone characteristics. This was validated by the distinction seen within the 200 feet gas sand formation in the deeper portion of the studied well (9600-9800 feet). Potential failure portions of the wells, a common feature in the field, were inferred from the sanding potential computed using the predicted mechanical properties value.
There are numerous correlations and thermodynamic models for predicting the natural gas hydrate formation condition but still the lack of a simple and unifying general model that addresses a broad ranges of gas mixture. This study was aimed to develop a user-friendly universal correlation based on hybrid group method of data handling (GMDH) for prediction of hydrate formation temperature of a wide range of natural gas mixtures including sweet and sour gas. To establish the hybrid GMDH, the total experimental data of 343 were obtained from open articles. The selection of input variables was based on the hydrate structure formed by each gas species. The modeling resulted in a strong algorithm since the squared correlation coefficient (R2) and root mean square error (RMSE) were 0.9721 and 1.2152, respectively. In comparison to some conventional correlation, this model represented not only the outstanding statistical parameters but also its absolute superiority over others. In particular, the result was encouraging for sour gases concentrated at H2S to the extent that the model outstrips all available thermodynamic models and correlations. Leverage statistical approach was applied on datasets to the discovery of the defected and doubtful experimental data and suitability of the model. According to this algorithm, approximately all the data points were in the proper range of the model and the proposed hybrid GMDH model was statistically reliable.
The continuous evaluation of the measured Stand Pipe Pressure (SPP) against a modeled SPP value in real-time involves the automatic detection of undesirable drilling events such as drill string washouts and mud pump failures. Numerous theoretical and experimental studies have been established to calculate the friction pressure losses using different rheological models and based on an extension of pipe flow correlations to an annular geometry. However, it would not be feasible to employ these models for real-time applications since they are limited to some conditions and intervals of application and require input parameters that might not be available in real-time on each rig. In this study, The Group Method of Data Handling (GMDH) is applied to develop a trustworthy model that can predict the SPP in real-time as a function of mud flow, well depth, RPM and the Fan VG viscometer reading at 600 and 300 rpm. In order to accomplish the modeling task, 3351 data points were collected from two wells from Algerian fields. Graphical and statistical assessment criteria disclosed that the model predictions are in excellent agreement with the experimental data with a coefficient of determination of 0.9666 and an average percent relative error less than 2.401%. Furthermore, another dataset (1594 data points) from well-3 was employed to validate the developed correlation for SPP. The obtained results confirmed that the proposed GMDH-SPP model can be applied in real-time to estimate the SPP with high accuracy. Besides, it was found that the proposed GMDH correlation follows the physically expected trends with respect to the employed input parameters. Lastly, the findings of this study can help for the early detection of downhole problems such as drill string washout, pump failure, and bit balling.
Currently, in the oil industry, artificial lift optimization (ALO) systems are dealing with different applications including well monitor and control, reservoir management, production optimization, predictive maintenance, artificial lift, and flow assurance, multiphase pumping systems, etc. The scope of this article is to provide a detailed analysis of ALO and predictive pump maintenance applications using machine learning (ML) and artificial intelligence (AI). The oil and gas industry has experienced a lot of improvements that have impacted the businesses and economies associated with the market in recent times. Issues such as unplanned shutdown time and failure of equipment cause a huge impact on many corporations especially with the current fluctuations in hydrocarbon prices. Similarly, advanced modern technologies such as real-time analysis and predictive maintenance are designed to drive ALO systems. This paper covers several applications and techniques in which ML and AI have been applied to optimize hydrocarbon withdrawal from potentially depleted reservoirs that require some external supports to uplift the reservoir fluid from sub surface to surface using artificial lift system. In a nutshell, the applications of AI and ML for the artificial lift selection, their predictive maintenance, equipment malfunctioning detection, etc. using a self-trained system are the main topics of this paper. While reviewing each of these techniques, the workflow is also explained along with the effectiveness of utilizing each application to the current operations.
This study used six fields data alongside correlation heat map to evaluate the field parameters that affect the accuracy of bottom hole pressure (BHP) estimation. The six oil field data were acquired using measurement while drilling device to collect surface measurements of the downhole pressure data while drilling. For the two case studies, measured field data of the wellbore filled with gasified mud system was utilized, and the wellbores were drilled using rotary jointed drill strings. Extremely Randomized Tree and feed forward neural network algorithms were used to develop models that can predict with high accuracy, BHP from measured field data. For modeling purpose, an extensive data from six fields was used, and the proposed model was further validated with two data from two new fields. The gathered data encompasses a variety of well data, general information/data, depths, hole size, and depths. The developed model was compared with data obtained from two new fields based on its capability, stability and accuracy. The result and model’s performance from the error analysis revealed that the two proposed Extra Tree and Feed Forward models replicate the bottom hole pressure data with R2 greater than 0.9. The high values of R2 for the two models suggest the relative reliability of the modelling techniques. The magnitudes of mean squared error and mean absolute percentage error for the predicted BHPs from both models range from 0.33 to 0.34 and 2.02%-2.14%, for the Extra tree model and 0.40-0.41 and 3.90%-3.99% for Feed Forward model respectively; the least errors were recorded for the Extra Tree model. Also, the mean absolute error of the Extra Tree model for both fields (9.13-10.39 psi) are lower than that of the Feed Forward model (10.98-11 psi), thus showing the higher precision of the Extra Tree model relative to the Feed Forward model. Literature has shown that underbalanced operation does not guarantee the improvement of horizontal well’s extension ability, because it mainly depends on the relationship between the bottomhole pressure and its corresponding critical point. Thus, the application of this study proposed models for predicting bottomhole pressure trends.
Artificial Intelligence (AI) is becoming popular for the Rate of Penetration (ROP) estimation, hence, the need to study the best techniques and their advantages over empirical models. Various literatures were analysed to determine the prevalence of AI in ROP computation and compare the computation accuracies with empirical models. Artificial Neural Network (ANN) accounted for over 92% of the AI techniques used for ROP computation and Weight on Bit (WOB) mostly influenced the computation accuracy. The accuracy of AI algorithms is better than the empirical models thus, will improve the drilling efficiency, reduce cost and enhance the development of pad wells.
It is difficult to determine the main controlling factors of tight oil production. In addition to the problem of uncontrollable prediction accuracy, the numerical prediction model established by the main controlling factors will also make the correctly predicted low production samples lose the value of development. Applying the optimization algorithm with fast convergence speed and global optimization to optimize the controllable parameters in the high-precision numerical prediction model can effectively improve the productivity of low production wells with timeliness, and bring greater economic value while saving development cost. Using PCA-GRA method, the sample weight and the weighted correlation ranking results of parameters affecting tight oil production were obtained. Thereupon then the main controlling factors of tight oil production were determined. Then we set up a BP neural network model with by taking the main controlling factors as input and tight oil production as output. The prediction effect of the network was good and can be put into use. The results of sensitivity analysis showed that the network was stable, and the total fracturing fluid volume had the greatest impact on the production of tight oil. Finally, by using genetic algorithm, we optimized the fracturing parameters of all low production well samples in the field data. Combined with the fracturing parameters of all high production well samples and the optimized fracturing parameters of low production wells, the optimal interval of fracturing parameters was given, which can provide guidance for the field fracturing operation.
With the advancement of technology, gas shales have become one of the most prominent energy sources all over the world. Therefore, estimating the amount of adsorbed gas in shale resources is necessary for the technical and economic foresight of the production operations. This paper presents a novel machine learning method called grey wolf optimizer support vector machine (GWO-SVM) to predict adsorbed gas. For this purpose, a data set containing temperature, pressure, total organic carbon (TOC), and humidity has been collected from several sources, and the GWO-SVM model was created based on it. The results show that this model has R-squared and root mean square error equal to 0.982 and 0.08, respectively. Also, the results ensure that the proposed model gives an excellent prediction of the amount of adsorbed gas compared to previously proposed models. Besides, according to the sensitivity analysis, among the input parameters, humidity has the highest effect on gas adsorption.
Gas field production forecast is an important basis for decision-making in the gas industry. How to accurately predict the dynamic production during gas field development is an important content of reservoir engineering research. Reservoir numerical simulation is the most common method for predicting oil and gas production. However, it requires a lot of data to build an accurate geological model which is tedious and time-consuming. At present, many scholars have used machine learning and data mining methods to predict oil and gas production, but they have not considered whether the use of increasing production measures will affect the predicted results.
Thus, ARIMA-RTS optimal smooth algorithm is the first applied to establish the prediction model of gas well production. According to the historical production data, the model is processed, the production differential autoregressive integral moving average (ARIMA) model in time series is established, then ARIMA model is combined with RTS (Rauch Tung Striebel) smoothing, and the production prediction model is constructed. RTS smoothing algorithm is an enhanced version of Kalman filter. The measurements are firstly processed by the forward filter, and then, a separate backward smoothing pass is used for obtaining the smoothing solution. The correctness of ARIMA-RTS model was verified with the actual production data. The results show that the prediction based on ARIMA-RTS model can accurately reflect the production performance of gas well. This method can effectively reduce the error caused by stimulation when predicting. When using the ARIMA-RTS model and the ARIMA-Kalman model to predict the production of the same gas well, the prediction accuracy of ARIMA-RTS model is higher than that of ARIMA-Kalman model in production wells with stimulation. Compared with that of the ARIMA-Kalman model, the mean relative error fitted by the ARIMA-RTS model is reduced by 46.3%, and the relative mean square error is reduced by 56.48%. ARIMA-RTS optimal smooth algorithm improves the prediction accuracy of gas well that uses stimulation. We therefore conclude that the ARIMA-RTS optimal smooth algorithm can help us better forecast the forecasting gas well production with stimulation, as well as other fuels output.
Well-known oil recovery factor estimation techniques such as analogy, volumetric calculations, material balance, decline curve analysis, hydrodynamic simulations have certain limitations. Those techniques are time-consuming, and require specific data and expert knowledge. Besides, though uncertainty estimation is highly desirable for this problem, the methods above do not include this by default. In this work, we present a data-driven technique for oil recovery factor (limited to water flooding) estimation using reservoir parameters and representative statistics. We apply advanced machine learning methods to historical worldwide oilfields datasets (more than 2000 oil reservoirs). The data-driven model might be used as a general tool for rapid and completely objective estimation of the oil recovery factor. In addition, it includes the ability to work with partial input data and to estimate the prediction interval of the oil recovery factor. We perform the evaluation in terms of accuracy and prediction intervals coverage for several tree-based machine learning techniques in application to the following two cases: (1) using parameters only related to geometry, geology, transport, storage and fluid properties, (2) using an extended set of parameters including development and production data. For both cases, the model proved itself to be robust and reliable. We conclude that the proposed data-driven approach overcomes several limitations of the traditional methods and is suitable for rapid, reliable and objective estimation of oil recovery factor for hydrocarbon reservoir.