Fuzzy Inference for Well Log Lithology Classification

In the oil and gas industries, lithology classification (characterization of rock samples) is used to improve accuracy in locating new reservoirs. Compared to geochemical analysis of physical rock samples in laboratory settings, wireline logging (measuring petro-physical properties of rocks using a variety of sensors lowered down a given borehole) provides a more cost-effective method of obtaining rock data for classification and characterization. However, this is a challenging task due to the complexity of well log data, and often expert knowledge is required to analyze and interpret this data. Within the remit of automatic well log data analysis, a new fuzzy inference system for sedimentary rock class classification is presented here. The performance of this system was evaluated on the data provided in the FORCE 2020 Machine Learning Competition. The system, which only used three fuzzy rules to map four well logs to three rock classes, had a classification accuracy of 74%, which was slightly lower than the classification accuracy of three non-linear classifiers: XGBoost (82%), Random Forest (82%) and Multi-Layer Perceptron (83%). These results are encouraging towards building a simple and human-explainable expert system to improve oil mining efficiency.