| Hydrocarbons being the most substantial commodity is the widely used energy resource in the world.It is quite contemporary and common to estimate the hydrocarbon potential from poor-quality reservoirs rock in Pakistan.Being an emerging economy,the exploration and development sector of oil and gas companies in Pakistan needs to understand the proper reservoir distribution and geology trend of these poor-quality reservoirs for the economic appraisals.Therefore,to diagnose a poor-quality and consolidated reservoir,it is essential to build the best suitable rock physics model and template that has been previously neglected.Implementing this approach,goals to ensure that the huge operational costs for companies is reduced drastically.The study required to examine the nature of missing or inconsistency data that was related to the production logs and be able to predict the levels of production in the oil and gas reservoirs.In addition,the traditional interpretations are time consuming due to the dependency on the manual approaches because it entails human involvements.In addition,the cost on the manual interpretation is also quite high which makes it challenging for the exploration and production(E&P)sector.For example,when the estimation of porosity is required density and sonic log maybe during logging process is affected by bad hole condition.Additionally,the specific log may be missing for the particular zone of interest.In these scenarios,an option is to obtain extra log data from a new drilling site or rerunning wireline logging to acquire the needed type of log for that has been already well drilled.Nevertheless,a new well drill or stopping the well production to rerun required well log cause huge extra cost and some kind of logs are not able to measure owning in casing.Therefore,the dependency on the manual approaches raises a lot question regarding the precision of the interpretation as well as cost and time.These problems can be solved by employing the novel and advanced automated machine learning approaches which can efficiently increase the accuracy of the interpretations by utilizing less time and cost.The estimation of petrophysical and elastic parameters,data conditioning,and evaluation of unknown parameters such as shear wave velocity logs can be efficiently identified via automated machine learning approaches.Furthermore,reservoir modeling is an important process to comprehend and evaluate a target reservoir and a model of reservoir apply a simulation of a petroleum reservoir for an area of reservoir engineering for getting ready for a field development plan.Wireline logging data is the most important role to play in reservoir modeling.Therefore,in this study,novel and advanced automated machine learning tools are utilized to accomplish the two following goals which were previously ignored;(a)data processing of well-logs to remove redundant values through machine learning tools to predict the missing logs such as shear sonic log(DTS)and density log(RHOB),and(b)development of a reliable rock physics model and rock physics template for the fields of Lower Indus Basin,SE Pakistan.Pakistan holds small and scattered reserves of Hydrocarbons almost throughout its length and width.Sindh province of Pakistan is known for its hydrocarbon resource potential as it boasts a major chunk of Pakistan's total Hydrocarbon production.The area selected for this study is a gas discovery area located in the North Eastern part of Sindh and the lease is called Sawan gas field.The Sawan gas-field is located in the Thar-Desert,at the southern eastern part of the Sukkur in the Sindh province of Pakistan.The Sawan gas-field is located in the Thar-Desert,at the southern eastern part of the Sukkur in the Sindh province of Pakistan.Sawan gas field is one of the significant gas producing areas with early–late Cretaceous Lower Goru Formation acting as the potential reservoir here.Sawan was and still is one of the biggest discoveries of gas reserves in Pakistan.To achieve these objectives of the study,this research proposes machine learning(ML)techniques to train models that streamline the data of the manually interpreted process,subsequently,minimizing human efforts and saving time of the manual interpretation and enhancing the log quality by machine learning techniques and to establish a suitable rock physics workflow that helps to build a rock physics model and template for a Lower Goru sand reservoir in Pakistan.This destination objective is answer and explore several key questions pertinent to the application of machine learning techniques to enhance the geophysical log responses and build a robust rock physics model and template.The study employs different rock physics models,using published empirical and theoretical approaches,to understand the logs response and relate them to the effects of porosity,lithology and fluid type.Various ML methods were employed to train and enhance the model's prediction such as automated ML workflow,encompassing quality control(QC),outlier detection,and rebuilding of well logs.It integrates contemporary ideas for outlier detection,predictive regression,classification,multi-dimensional scaling(MDS),and novel well ranking based on inter-well similarity.Similarly,to construct a rock physics model;the second objective of the study,which is required to accurately understand the geological properties of a well and support activities including oil and gas exploration,a petrophysical interpretation must be performed.This is performed by an experienced petrophysicist and follows a workflow with a number of steps running consecutively,each using the results of previous steps.Petrophysics combines well logs data for evaluating,predicting and establishing the formation lithology,porosity,hydrocarbon saturation and permeability.Petrophysics is also used to estimate the economic feasibility of a well.Well log analysis for geophysics diverges from conventional log analysis in many aspects.In surface seismic usage,the vertical dimension of the zone of interest is always much larger than that of the hydrocarbon production zones.Geophysicists need elastic rock properties over the entire interval through which the seismic waves passed from the topographic surface to the total depth.This section of the paper summarizes the methods used for quality control(QC)of logs,conditioning logs and estimating petrophysical properties.The results have confirmed the novel role of ML in the field of petrophysics.For example,there is only 400 m of data available in this Sawan-03 well,of which 100 m is washout section,resulting in a noisy measured Density curve.It is a far more timeconsuming procedure for the petrophysicist to rebuild the affected part of the curve based on the section,where the curve is in good condition in the well,and this is where the use of ML can be of the most advantage.It can be observed from comparing the measured density curve(black curve)against our ML model's rebuilding Density curve(red curve)that presents that our ML model could rebuild the Density curve.The fact that the Machine Learning log follows the measured log is itself an important Quality Control check as it confirms that the Machine Learning model is trained properly.Similarly,we also applied the Machine learning approach to another well that demonstrates the outcomes of our ML model for one of our selected wells.There is only 250 m of data available in this Sawan-04 well,of which 50 m is washout section,resulting in a noisy measured Density curve.It is a far more time-consuming procedure for the petrophysicist to rebuild the affected part of the curve based on the section,where the curve is in good condition in the well,and this is where the use of ML can be of the most advantage.It can be observed from comparing the measured density curve(black curve)against our ML model's rebuilding Density curve(red curve)that presents that our ML model could rebuild the Density curve.Deep Neural Network(DNN)along with the similarity metrics such as Jaccard and Overlap similarities was employed to examine the relationship between the wells and dimensionality reduction techniques including multidimensional scaling(MDS)and well-ranking process are applied to extract common geophysical responses of the wells.A higher response indicates existence of a strong similarity.This can also be verified by the superimposed of well log data.For the purpose,the research evaluates a dataset of five wells to develop a modified rock physics model and template for a Lower Goru sand reservoir in Pakistan.The study compares and evaluates different rock physic models such as Stiff sand,Friable-sand,Greenberg and Castagna,and Raymer's model to calibrate the best model,which would comprehensively assist the future researchers.According to the results,Stiff sand is quite a suitable rock physics model.Furthermore,the rock physics template(RPT)is utilized to test the reliability of the predicted model that is helpful for the formation evaluation,reservoir characterization,and prospect evaluation across different fields of the Lower Goru sand reservoir.The predicted model can be further used to estimate porosity from the seismically derived impedance in the entire Lower Goru sand reservoir,Pakistan,and worldwide which has the same geological trends and reservoir distribution.Finally,after enhancing log quality and prediction of missing logs,the study attempts to accomplish the second and last aim of the study-building a suitable model for the Lower Guru sand reservoir in Pakistan.Once the affected log has been rebuilt(such as density and sonic log)with other available logs,they are then combined to estimate the petrophysical properties,as the geological factors of any reservoir,such as porosity,lithology,and fluid characteristics,control the rock physics models.Several retrospective studies have discussed the methods of discriminating and calculating these geological factors in the Lower Goru sand reservoir.Therefore,it is imperative to build the best-modified suitable rock physics model and template to diagnose an inadequately consolidated reservoir that has been previously overlooked.Overall,the performance of the Machine Learning Model is quite encouraging that it produces the answers of accuracy that match the measured(original)results of the wells for sonic and density logs.Bearing in mind that rebuilding(removed the effect of washed out)such density and sonic curves is a challenging and timeconsuming task for a manual interpreter,the fact that our Machine Learning model can generate these results fairly and accurately in a matter of seconds is very useful.The potential benefits of this method are that it does not follow the zone-by-zone prediction of the missing logs like rock physics methods do and it outputs the uncertainties facilitated by the least-squares method.Having the potential of demonstrating shear sonic log prediction in hydrocarbon-bearing zones,which cannot be precisely predicted by the Greenberg-Castagna method that only works in brine-saturated rocks,this approach will provide improved accuracy where shear sonic logs are missing and need to be predicted for geomechanics,rock physics and other applications.A perfect coherency of modeling laid a substantial base for enhanced reservoir characterization.The measured model also proves to be useful in precisely calculating the elastic parameters such as Vp,Vs,and density in all wells.The quantitative values of cross plots,such as the average cutoff of petrophysical and elastic parameters,describe an ostensible differentiation among shale,shaly sand,and gas-bearing sand zones.The results identify that in gas-bearing sand,the VP/VS ratio is much sensitive,followed by P-impedance.The proposed model and calibrated RPT can be used to enhance seismic reservoir properties.The reliability of the model is helpful for the formation evaluation,reservoir characterization,and prospect evaluation across different fields of the Lower Goru sand reservoir.The predicted model can further be utilized to estimate porosity from the seismically derived impedance worldwide that has the same geological trends and reservoir distribution.It would also be extremely useful to apply the workflows designed in this study to other wells in the same field to test the sensitivity and quality of the model developed for this reservoir.It is also very important to use the best quality data available while implementing these workflows.It is recommended that researchers extrapolate the reservoir properties away from the well location,by inverting seismic data constrained by the rock physics models. |
PDF Full Text Request