Prediksi Endapan Asphaltene Menggunakan Machine Learning Dengan Metode Bayesian Network

Dino, Irfan (2021) Prediksi Endapan Asphaltene Menggunakan Machine Learning Dengan Metode Bayesian Network. Other thesis, Universitas Islam Riau.

Text 163210109.pdf - Submitted Version Download (2MB)

Abstract

Asphaltene deposits are one of the causes of production problems because asphaltene deposits are able to change the characteristics of formations and cause plugging on surface facilities. In solving the problem, it takes calculation to know the value of variables that affect the precipitation of asphaltene with the value of asphaltene precipitation formed. Prediction is made by an Artificial intelligent method with machine learning using the Bayesian Network classification method. The Bayesian network method is a probabilistic graphical model (PGM) that has a direct acyclic graph (DAG) structure that shows a one-way opportunity relationship on each variable, each variable is described as a node and the causality of each variable is associated with an arrow or called an arc or edge. In building the Bayesian Network, structures are built taking into a statistical approach called conditional probability. Bayesian Network application using Python as a programming language and using Jupyter Notebook as an open-source to process code and data, and also use pgmpy as a python library in making some model. The model will be created first by determining the structure of the graph based on structure learning by using the hill-climbing algorithm by using BDeu score as the scoring method, then the Bayesian network model will be built according to the structure that has been determined by using training data with different weights, then doing learning parameters using Bayesian estimation parameters based on probability approach from data based on graph structure. Then the Bayesian network model that has been built will be tested for its ability to predict asphaltene deposits by inference using the variable elimination method, the weight of the test data used varies by weight by 10%,15%,20%,25%, and 30% of the total attribute value of the dataset and selected randomly. The ability of the model built to predict asphaltene deposits is then measured performance with matric threshold accuracy, while the accuracy value obtained from each test varies with an average accuracy value of 64%.

Actions (login required)

View Item