Prediction of porosity amount of the Dadaş shale based on amodel using deep learning

Unconventional hydrocarbon exploration and production activities are increasingly utilizing horizontal drilling, hydraulic fracturing, and artificial intelligence technologies. The porosity levels of shales are important parameters in determining potential areas for hydraulic fracturing and drilling in unconventional reservoirs, and even identifying which levels are suitable for production before drilling begins. In this study, a data-driven machine learning method was employed to estimate the porosity of Dadaş shale using more economical and readily available data. Initially, a model was created using a dataset comprised of well-known shales worldwide. Subsequently, a prediction dataset was formed from XRD analysis data of Dadaş shale (Quartz, Feldspar, Carbonate, Clay) and Pyrolysis (TOC) data. This prediction dataset consists of analyses from 149 samples from the Doğan-1, K. Migo-2, Soğuktepe-1, Derindere-1, Akçay-1, Hançerli-3, Abdülaziz-1, B. Azıklı-1, Telhasan-1, Çalıktepe-2, and Diyarbakır 1 wells."Unconventional hydrocarbon exploration and production activities are increasingly utilizing horizontal drilling, hydraulic fracturing, and artificial intelligence technologies. The porosity levels of shales are important parameters in determining potential areas for hydraulic fracturing and drilling in unconventional reservoirs, and even identifying which levels are suitable for production before drilling begins. In this study, a data-driven machine learning method was employed to estimate the porosity of Dadaş shale using more economical and readily available data. Initially, a model was created using a dataset comprised of well-known shales worldwide. Subsequently, a prediction dataset was formed from XRD analysis data of Dadaş shale (Quartz, Feldspar, Carbonate, Clay) and Pyrolysis (TOC) data. This prediction dataset consists of analyses from 149 samples from the Doğan-1, K. Migo-2, Soğuktepe-1, Derindere-1, Akçay-1, Hançerli-3, Abdülaziz-1, B. Azıklı-1, Telhasan-1, Çalıktepe-2, and Diyarbakır 1 wells. A gradient and test study conducted with a 512, 256, 128, 64, and 64 neuron and 5-layer artificial neural network (ANN) using the Tensorflow-Keras library yielded the best average error history. The artificial neural networks (ANNs) produced high predictions with a linear correlation coefficient R2=0.96 using the parameters of this model. By reloading the model created using these ANN parameters and inputting the prediction dataset into this model via Python 3 programming language in a Jupyter Notebook, the porosity levels of Dadaş shale in the wells were successfully determined