Artificial Intelligence-based Predictive Technique to Estimate Oil Formation Volume Factor

This investigation presents a powerful predictive model to determine crude oil formation volume factor (FVF) using state-of-the-art artificial intelligence (AI) techniques. FVF is a vital pressure-volume-temperature (PVT) parameter used to characterize hydrocarbon systems and is pivotal to reserves calculation and reservoir engineering studies. Ideally, FVF is measured at the laboratory scale; however, prognostic tools to evaluate this parameter can optimize time and cost estimates. The database utilized in this study is obtained from open literature and covers statistics of crude oils of the Middle East region. Multiple AI algorithms are considered, including Artificial Neural Networks (ANN) and Artificial Neural Fuzzy Inference Systems (ANFIS). Models are developed utilizing an optimization strategy for various parameters/hyper-parameters of the respective algorithms. Unique permutations and combinations for the number of perceptron and their resident layers is investigated to reach a solution that provides the most optimum output. These intelligent models are produced as a function of the parameters intrinsically affecting FVF; reservoir temperature, solution GOR, gas specific gravity, bubble point pressure, and crude oil API gravity. Comparative analysis of developed AI models is performed using visualization/statistical analysis, and the best model is pointed out. Finally, the mathematical equation extraction to determine FVF is accomplished with the respective weights and bias for the model presented. Graphical analysis is used to evaluate the performance of developed AI models. The results of scatter plots showed most of the points are lying on the 45 degree line. Moreover, during this study, an error metric is developed comprising of multiple analysis parameters; Average absolute percentage error (AAPE), Root Mean Squared Error (RMSE), coefficient of determination (R2). All models investigated are tested on an unseen dataset to prevent a biased model's development. Performance of the established AI models is gauged based on this error metric, demonstrating that ANN outperforms ANFIS with error within 1% of the measured PVT values. A computationally derived intelligent model provides the strongest predictive capabilities as it maps complex non-linear interactions between various input parameters leading to FVF.

Development of Oil Formation Volume Factor Model using Adaptive Neuro-Fuzzy Inference Systems ANFIS

The oil formation volume factor is one of the main reservoir fluid properties that plays a crucial role in designing successful field development planning and oil and gas production optimization. The oil formation volume factor can be acquired from pressure-volume-temperature (PVT) laboratory experiments; nonetheless, these experiments' results are time-consuming and costly. Therefore, many studies used alternative methods, namely empirical correlations (using regression techniques) and machine learning to determine the formation volume factor. Unfortunately, the previous correlations and machine learning methods have some limitations, such as the lack of accuracy. Furthermore, most earlier models have not studied the relationships between the inputs and outputs to show the proper physical behaviors. Consequently, this study comes to develop a model to predict the oil formation volume factor at the bubble point (Bo) using an adaptive neuro-fuzzy inference system (ANFIS). The ANFIS model was built based on 924 data sets collected from published sources. The ANFIS model and previous 28 models were validated and compared using the trend analysis and statistical error analysis, namely average absolute percent relative error (AAPRE) and correlation coefficient (R). The trend analysis study has shown that the ANFIS model and some previous models follow the correct trend analysis. The ANFIS model is the first rank model and has the lowest AAPRE of 0.71 and the highest (R) of 0.9973. The ANFIS model also has the lowest average percent relative error (APRE), root mean square error (RMSE), and standard deviation (SD) of -0.09, 1.01, 0.0075, respectively.

Integration of Self Organizing Map and Date Driven Methods to Predict Oil Formation Volume Factor: North Africa Crude Oil Examples

In this paper we introduce for the first time an innovative approach for deriving Oil Formation Volume Factor (Bo) by mean of artificial intelligence method. In a new proposed application Self-Organizing Map (SOM) technology has been merged with statistical prediction methods integrating in a single step dimensionality reduction, extraction of input data structure pattern and prediction of formation volume factor Bo. The SOM neural network method applies an unsupervised training algorithm combined with back propagation neural network BPNN to subdivide the entire set of PVT input into different patterns identifying a set of data that have something in common and run individual MLFF ANN models for each specific PVT cluster and computing Bo. PVT data for more than two hundred oil samples (total of 804 data points) were collected from the north African region representing different basin and covering a greater geographical area were used in this study. To establish clear Bound on the accuracy of Bo determination several statistical parameters and terminology included in the presentation of the result from SOM-Neural Network solution. the main outcome is the reduction of error obtained by the new proposed competitive Learning Structure integration of SOM and MLFF ANN to less than 1 % compared to other method. however also investigated in this work five independents means of model driven and data driven approach for estimating Bo theses are 1) Optimal Transformations for Multiple Regression as introduced by (McCain, 1998) using alternating conditional expectations (ACE) for selecting multiple regression transformations 2), Genetic programing and heuristic modeling using Symbolic Regression (SR) and cross validation for model automatic tuning 3) Machine learning predictive model (Nearest Neighbor Regression, Kernel Ridge regression, Gaussian Process Regression (GPR), Random Forest Regression (RF), Support Vector Regression (SVM), Decision Tree Regression (DT), Gradient Boosting Machine Regression (GBM), Group modeling data handling (GMDH). Regression Model Accuracy Metrics (Average absolute relative error, R-square), diagnostic plot was used to address the more adequate techniques and model for predicting Bo.

A Deep Learning Approach for the Prediction of Oil Formation Volume Factor

Accurate knowledge of Pressure-Volume-Temperature (PVT) properties is crucial in reservoir and production engineering computational applications. One of these properties is the oil formation volume factor (Bo), which assumes a significant role in calculating some of the prominent petroleum engineering terms and parameters, such as depletion rate, oil in place, reservoir simulation, material balance equation, well testing, reservoir production calculation, etc. These properties are ideally measured experimentally in the laboratory, based on downhole or recommended surface samples. Faster and cheaper methods are important for real-time decision making and empirically developed correlations are used in the prediction of this property. This work is aimed at developing a more accurate prediction method than the more common methods. The prediction method used is based on a supervised deep neural network to estimate oil formation volume factor at bubble point pressure as a function of gas-oil ratio, gas gravity, specific oil gravity, and reservoir temperature. Deep learning is applied in this paper to address the inaccuracy of empirically derived correlations used for predicting oil formation volume factor. Neural Networks would help us find hidden patterns in the data, which cannot be found otherwise. A multi-layer neural network was used for the prediction via the anaconda programming environment. Two frameworks for modelling data using deep learning viz: TensorFlow and Keras were utilized, and PVT variables selected as input neurons while employing early stopping which uses a part of our data not fed to the model to test its performance to prevent overfitting. In the modelling process, 2994 dataset retrieved from the Niger Delta region was used. The dataset was randomly divided into three parts of which 60% was used for training, 20% for validation, and 20% for testing. The result predicted by the network outperformed existing correlations by the statistical parameters used for the same set of field data. The network has a mean average error of 0.05 which is the lowest when compared to the error generated by other correlation models. The predictive capability of this network is found to be higher than existing models, based on the findings of this work.

Development of a Computationally Intelligent Model to Estimate Oil Formation Volume Factor

This investigation presents a powerful predictive model to determine crude oil formation volume factor (FVF) using state-of-the-art computational intelligence (CI) techniques. FVF is a vital pressure-volume-temperature (PVT) parameter used to characterize hydrocarbon systems and is pivotal to reserve evaluation studies and reservoir engineering calculations. Ideally, FVF is measured at the laboratory scale; however, prognostic tools to evaluate this parameter can aid in optimizing time and cost estimates. The database utilized in this study is obtained from open literature and covers statistics of crude oils of Pakistan, Iran, UAE, and Malaysia. Resultantly, this allows to move step forward towards the creation of a generalized model. Multiple CI algorithms are considered, including Artificial Neural Networks (ANN) and Artificial Neural Fuzzy Inference Systems (ANFIS). Models for CI are developed utilizing an optimization strategy for various parameters/hyper-parameters of the respective algorithms. Unique permutations and combinations for the number of perceptron and their resident layers is investigated to reach a solution that provides the most optimum output. These intelligent models are produced as a function of the parameters intrinsically affecting FVF; reservoir temperature, solution GOR, gas specific gravity, and crude oil API gravity. Comparative analysis of various CI models is performed using visualization/statistical analysis and the best model pointed out. Finally, the mathematical equation extraction to determine FVF is accomplished with the respective weights and bias for the model presented. Graphical analysis using scatter plots with a coefficient of determination (R2) illustrates that ANN equation produces the most accurate predictions for oil FVF with R2 in excess of 0.96. Moreover, during this study an error metric is developed comprising of multiple analysis parameters; Average Absolute Error (AAE), Root Mean Squared Error (RMSE), correlation coefficient (R). All models investigated are tested on an unseen dataset to prevent the development of a biased model. Performance of the established CI models are gauged based on this error metric, which demonstrates that ANN outperforms the other models with error within 2% of the measured PVT values. A computationally derived intelligent model proves to provide the strongest predictive capabilities as it maps complex non-linear interactions between various input parameters leading to FVF.

Development of Proxy Models for Screening Water Flood and Gas Flood Candidates

Fluid-flood and other improved oil recovery techniques are becoming prominent in global petroleum production because a large proportion of production is from mature oil fields. Although water flooding and gas injection are well established techniques in the industry, several of the screening criteria in literature are discipline which could sometimes be subjective. This work used experimental design techniques to develop proxy models for predicting oil recovery under water-flood and gas-flood conditions. The objective of the study is to develop a quantitative screening method that would allow for candidates to be evaluated and ranked for water flood or gas injection. The model was applied to some field cases and compared with published models and the well-known Welge Analysis method. The coefficient constants for the oil formation volume factor for water flooding and gas injection was 0.0139 and 0.0434 respectively. Similarly, the coefficient constants for water injection and gas injection for the generated proxy model was -2.34* 10-8 and -6.1 *10-5 respectively. The results show that the proxy models developed are quite robust and can be used for first pass screening of water and gas flood candidates.