Application of decision tree, artificial neural networks, and adaptive neuro-fuzzy inference system on predicting lost circulation: A case study from Marun oil field

Abstract

Abstract One of the most prevalent problems in drilling industry is lost circulation which causes intense increase in drilling expenditure as well as operational obstacles such as well instability and blowout. The aim of this research is to develop smart systems for estimating amount of lost circulation making able to use appropriate prevention and remediation methods. To obtain this aim, a large data set were collected from 61 recently drilled wells in Marun oil field in Iran to be used for developing relevant models. After that, using the extracted data set consisting of 1900 data subset, intelligent prediction models including decision tree (DT), adaptive neuro-fuzzy inference systems (ANFIS), artificial neural networks (ANN) and also hybrid artificial neural network namely genetic algorithm-multi-layer perception (GA-MLP) were developed to make a quantitative prediction on lost circulation. The model outputs are then analyzed by various performance indices such as variance accounted for (VAF), root mean square error (RMSE), performance index (PI) and coefficient of determination (R2). Eventually, it is found that developed models are highly applicable in lost circulation prediction. Concluding remark is that DT model having determination coefficient of 0.9355 and RMSE of 0.091 is superior comparing to other developed models and hybrid ANN (GA-MLP) exhibits lowest prediction performance among other implemented models.