AUTOMATED LITHOLOGY PREDICTION BASED ON WELL LOGGING DATA USING MACHINE LEARNING

• Main Author: Irfan Ibrahim, Muhammad
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/40048
• Institution: Institut Teknologi Bandung

Understanding of the reservoir characteristics and conditions is important in the development of an oil and gas project. One characteristic that needs to be known is the lithology types of the reservoir formation. Information about the lithology types affect the calculation of petrophysical parameter of the reservoir. Generally, identification of lithology types is done manually by geologists based on cuttings sightings, core samples, and well logging data, which may not be efficient enough as it takes lot of time and effort. Therefore, several approaches in supervised learning method are applied to improve the efficiency and accuracy of the identification of lithology types. In this study the supervised learning method was used so that input data and desired output data were needed to build a predictive model that would be used to identify the lithology types. Input data used are well logging data which usually available such as gamma ray (GR), spontaneous potential (SP), density, and neutron porosity. While the desired output data is obtained from core samples in the form of lithology types label. Several approaches in supervised learning method were applied to find model with sufficient accuracy, there are KNearest Neighbors (KNN) model, Support Vector Classifier (SVC) model, NuSVC model, LinearSVC model, and Stochastic Gradient Descent Classifier (SGDC) model. Data set was divided into train data set and test data set. 75% of the data is used to train the model while 25% of the data is used as test data. Several adjusting to parameters of each model was done. Prediction results of each model is compared. All models are able to achieve accuracy higher than 80%, where accuracy of these models is 94.09%, 93.18%, 80.91%, 84.09%, and 84.55% for KNN, SVC, NuSVC, LinearSVC, and SGDC model sequentially. Performances in predicting each label is also compared. KNN and SVC model offer good overall performances. The study shows that the KNN model is the best model in predicting lithology types. KNN model can be applied to predicting lithology with improved efficiency and accuracy.