Application of an enhanced decision-tree learning approach for prediction of petroleum production.

摘要

Prediction of oil well production is important in estimating economic benefit of a well. However, this prediction task is difficult because of the complex subsurface conditions of wells. In response to the above problems, advancements in data mining technology, in recent years, has improved the ability for discovering information within a database that can then be used to support decisions. Data mining technology is a powerful AI tool that effectively extracts information from massive observational data sets, as well as discovers new and meaningful knowledge for the user.;As well, we explore modeling petroleum production data using the NDT model. We experiment in the modeling process by introducing different strategies and different parameter combinations. First, an overall oil production model is developed using three geoscience parameters (permeability, porosity and first shut-in pressure). Second, two different models, with different input parameters, are developed to predict production in the post water flooding stage only. The results of the above models indicate that the mechanisms used are somewhat superficial and these configurations may not allow the data-driven models to classify and predict oil production. Finally, a trend model is developed in an attempt to improve the effectiveness and accuracy of the predictive model. The result shows that the trend model demonstrates an improved performance and is comparable to the artificial neural network.;In this thesis, we adopt an existing neural based decision-learning (NDT) model, which can obtain explicit information on the processing involved in generating predictions of oil production. In our experiment, the NDT model, which uses a neural network to extract the underlying attribute dependencies, was evaluated in comparison to the conventional C4.5 model on different kinds of data set. The results generated by the NDT model are found to be satisfactory.