PREDICTING INITIAL PRODUCTION RATE USING MACHINE LEARNING (K-NEAREST NEIGHBOR)

• Main Author: Andriani, Safira
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/40040
• Institution: Institut Teknologi Bandung

Predicting the initial production rate is an intrinsic part of project forecasting, which influences the economic viability of the opportunity and the producing life subsequent facilities. However, predicting the accurate result of the initial production rate is facing a challenge as the function in determining it involves calculations which require a full feature data that is commonly not available during the early stage. As a result, it is not possible to find the exact prediction result of initial production rate. A practical alternative is the use of machine learning that is ideal for repetitive calculations and does not require the availability of full-featured data. In this paper, K-nearest Neighbor (KNN) calculation is used as the machine learning method to predict initial oil production rate. This method calculates the distance or delta between the training data and parameters of interest and search for the K-nearest value. The approach involves gathering several input data, which are net pay, porosity, permeability, water saturation, and initial production rate as the training set. During this stage, the K value will be determined and will be used for further prediction in the prediction phase. By using Euclidian equation for calculating distance and integrating those data in a simple KNN model calculator, this method aims to give better result in predicting initial oil production rate compared to a deterministic approach where minimum data is available. Most of the principles here apply to fields outside this area and can also be adapted to predict other value related in the oil and gas industry where the result follows a particular pattern. It is concluded that this method is promising for further use. However, a further study is needed using various field data to ensure this method is reliable for any reservoir and input data.