A Comparative Study on Different Machine Learning Algorithms for Petroleum Production Forecasting

Abstract

In the recent years, machine learning and its subset, deep learning, have been quickly developed and applied with great success in various areas of petroleum engineering. Different machine learning algorithms for petroleum production forecast are studied in this work for efficiency comparison purpose. Historical production data from an oil well currently producing in the oilfield X, Southern Vietnam has been first preprocessed to construct eight different predictive models for production forecast on the oil well under consideration. The algorithms under consideration in this work are: (1) the classical machine learning algorithms, including random forest, gradient boosting, k-nearest neighbor, support vector regression; and (2) the deep learning algorithms, including multilayer perceptron, convolutional neural network, long short-term memory, and gated recurrent unit. The results from this comparative study show that in spite of their simplicity, some classical machine learning algorithms, especially the support vector regression shows its high efficiency in performing the prediction tests. In addition, it can be found from this work that preprocessing of the historical production data is crucial to the success of the application of artificial neural networks to production forecasting.