No abstract

Porosity, a critical property of petroleum reservoirs, is a key controlling factor of the reservoir storage capacity. It has been conventionally measured from core plugs. Empirical correlations, statistical, and machine learning methods have been employed for indirect estimation of porosity. The results obtained from these approaches are available only after acquiring drilling and wireline logs. Obtaining porosity estimates in real time, ahead of wireline logging, can help in making critical decisions and enabling early assessment of reservoir quality. We present the results of a machine learning approach to predicting porosity from advanced mud gas data. Datasets integrating advanced mud gas data with porosity were gathered from seven wells to prove this concept. The mud gas data includes light and heavy flare gas components. Optimized artificial neural network (ANN) models were applied to the datasets and multivariate linear regression (MLR) models were used as benchmarks. Each well dataset was split into training and validation subsets using a randomized sampling approach with the ratio of 70:30. A 100 ppm cut-off was applied to the total normalized gas. To evaluate the performance of the models, we use correlation coefficients (R) and mean squared error (MSE). The ANN models consistently outperformed the MLR models in all the datasets. The ANN models have training and validation correlation coefficients of up to 0.89 and 0.88, respectively, compared to an average of 0.79 and 0.77 for the MLR models. The training and validation MSEs for the ANN models are as low as 0.0135 and 0.021, respectively, compared to those of the MLR models in the range of 0.0007 and 0.03, respectively. These results indicate the nonlinearity of the relationship between porosity and the gas components. Furthermore, it can be deduced that the approach is feasible and better results are achievable. The randomized sampling ensures that each data point has an equal chance to be used for either training or validation without bias. The cut-off applied to the normalized total gas is a standard practice to eliminate the background gas effect in the mud gas data. This study provides an opportunity to utilize mud gas data beyond the traditional fluid typing and petrophysical correlation purposes. The presented approach has the capability to complement existing reservoir characterization approaches in providing reservoir quality assessments at the early stage of exploration. We plan to apply state-of-the-art machine learning models and perform sensitivity analysis on the gas components in the future to increase the accuracy.

The utility of advanced mud gas (AMG) data has been limited to fluid typing and petrophysical correlations. There is the need to extend the utility to real-time reservoir characterization prior to wireline logging and geological core description. Our first attempt to predict reservoir rock porosity within a well yielded good result. This study improves on the previous effort by utilizing big data obtained from combining various wells in the study area. We used machine learning (ML) methodology in the absence of established physical relationship between AMG data, comprising light and heavy flare gas components, and reservoir rock porosity. More than 20,000 data points collected from representative wells were used to prove the concept of predicting the porosity in an interval or section of any well within the study area. Optimized models of artificial neural network (ANN), decision trees (DT) and random forest (RF) were applied to the combined dataset. The combined dataset was randomly split into training and validation subsets in 70:30 ratio. The 30% validation subset simulates a missing well interval or section. Comparing the results of the ANN, DT, and RF models using statistical model performance evaluation metrics, the RF model outperformed the others. The RF model gave a training and validation correlation coefficient (R-Squared) values of 0.94 and 0.83 respectively compared to 0.36 and 0.35 for the ANN and 0.84 and 0.73 for the DT models respectively. However, the p-value and mean errors show that the models are statistically acceptable. Having showed in a previous research that a multivariate linear regression model could not handle the complexity in the relationship between porosity and the flare gas components, these results have further confirmed the robustness of nonlinear solutions based on the ML methodology. We conclude that the ML approach to predicting reservoir rock porosity from advanced mud gas data is feasible and better results are achievable with more research. This study has confirmed the feasibility of predicting porosity based on a dataset of combined wells and the huge benefit in extending the utility of AMG data beyond the traditional workflows. This approach is capable of complementing existing reservoir characterization processes in assessing reservoir quality at the early stage of exploration. Future work will investigate the impact of integrating the AMG with surface drilling parameters to possibly increase the prediction accuracy.

Background gas is the baseline gas measurement due to the recycled gas dissolved in or expelled from the drilling mud additives. It occurs more in oil-based mud systems than in water-based. A cut-off is usually applied on the mud gas data to remove the background gas effect in traditional mud gas analyses. This imposes an overhead on modeling procedures. This study investigates the effect of applying the cut-off on the performance of machine learning algorithms. A case of porosity prediction using advanced mud gas data is considered in this study. Using data from six wells, we implemented two experiments to compare the performance of artificial neural networks (ANN) with and without the cut-off. The first experiment applies a cut-off of 100 ppm on the total normalized gas while the second uses the entire data without the cut-off. The comparative results are benchmarked with those of a multivariate linear regression (MLR). Each well dataset was split into training and validation subsets using a randomized sampling approach in the ratio of 70:30. The results compare each of the MLR and ANN models individually and over all the datasets without and with the cut-off applied. The ANN models show better or same performance on the datasets without the cut-off in four out of six cases (67%). This shows that the ANN models may be less affected by the presence of the background gases in the mud gas datasets. It could be preliminarily concluded, based on the data used in this study, that it might be unnecessary to apply cut-offs on the mud gas data for ML algorithms due to their capability to handle noisy data. This conclusion is, however, subject to more extensive studies while ensuring consistency. Avoiding the application of the cut-off will remove the unnecessary overhead and provide more data for effective ML model training. While the results of this preliminary study somewhat agree with the traditional practice of applying a cut-off on advanced mud gas data, more extensive experiments will be conducted in our future work to further validate the conclusion. The background gas is traditionally considered noisy. In ML modeling, it could provide more information to further explain the nonlinear relationship between the input features and the target variable, hence improving the predictive capability.

In our previous study, we presented the preliminary results of the first attempt to predict reservoir rock porosity from advanced mud gas (AMG) data within the wellbore. The objective was to investigate the feasibility of generating a porosity log while drilling prior to wireline logging and core description processes. Knowing that porosity remains a critical property of petroleum reservoirs, this work improves on the previous research to predict porosity within a field. The methodology leveraged the machine learning (ML) paradigm in the absence of established physical relationship between AMG data, comprising light and heavy flare gas components, and reservoir rock porosity. More than 15,000 data points collected from representative wells in a field were used to prove the possibility of predicting the missing porosity in a well within the field. Optimized models of artificial neural network (ANN), decision trees (DT) and random forest (RF) were applied to the combined dataset. The dataset was randomly split into training and validation subsets in 70:30 ratio simulating the complete and missing sections respectively. Comparing the results of the ANN, DT, and RF models using statistical model performance evaluation metrics, the RF model consistently outperformed the others. In one of the test cases, the RF model gave a correlation coefficient (R-Squared) value of 0.84 compared to 0.46, and 0.78 for ANN and DT models respectively. The RF model also has a mean squared error (MSE) of 0.001 compared to 0.02 and 0.01 respectively for ANN and DT models. Having showed in a previous publication that a multivariate linear regression model could not handle the complexity in the relationship between porosity and the flare gas components, these results have further confirmed the robustness of nonlinear solutions based on the ML methodology. It can be deduced that the ML approach to predicting reservoir rock porosity from advanced mud gas data is feasible and better results are achievable with more research. This study has confirmed the feasibility of predicting porosity at the field scale and the huge benefit in utilizing AMG data beyond the traditional fluid typing and petrophysical correlation processes. The presented approach has the capability to complement existing reservoir characterization processes in assessing reservoir quality at the early stage of exploration. Future work will investigate the impact of integrating the AMG with surface drilling parameters to possibly increase the prediction accuracy.