Prediction of the Incrustating Trend in Oil Extraction Pipelines: An Approach Based on Neural Decision Trees

The oil and gas industry assesses the tendency of mineral deposit formation based on the principle of chemical equilibrium of the fluid based on existing production data. Instead of using this approach, the present work has used artificial intelligence to develop predictions of the incrustating tendency within oil extraction pipes using physicochemical analyzes on the extracted oil, using the processing capacity of current computers and the use of artificial neural networks of deep learning with the objective of determining how reliable a prediction based on artificial intelligence can be. Simultaneously, contemporary evaluation methods require on-site inspections that mostly provide remediation measures involving the consumption of labor and financial resources. Consequently, a new method for predicting the embedded trend in pipes based on an artificial neural network using decision trees as classifiers is proposed. The neural network model is trained based on an extensive database of the characteristics of the oil and the incrustation generated in the pipeline to obtain a predictive model. Subsequently, the model generates a decision tree by selecting within the database that information relevant to the solution of the problem and excluding the rest. The results of the experimentation and simulation were satisfactorily compared, obtaining a success rate of 83,26% when evaluated with a dataset dedicated only to the validation phase. Finally, the incrustating trend detection model using decision trees proved to be an applicable technology in the field of engineering within the field of gas and oil belonging to the Ecuadorian industry.