Digital Concept and Machine Learning for Cost Estimation of Oil and Gas Development

Abstract

Abstract The aspiration of cost estimation for oil and gas development is to deliver sharper, accurate yet competitive estimates. Accuracy of cost estimation highly depends on certain attributes such as scope definition and project schedule. In addition to that, uncertainties and pace of change in oil and gas market such as fluctuation of oil price and supply chain, trade wars and geopolitical issues, are external challenges that will affect the accuracy of cost estimates. Strong database is required to predict the cost based on historical and future market rates. One of the solutions is using probabilistic model to have range of possible outcome for an estimate. Recently, computing techniques such as data analytics and machine learning for quick and accurate cost estimation has been seen as an alternative to overcome the uncertainty of input parameters. Digital Concept such as Machine Learning leverages on statistical approach, nonlinear equations and genetic algorithm programming is used to model the cost. The metrics are used to supply meaningful and timely management information regarding techniques and process. Cost Estimate Relationship (CER) is established to model the correlation against the technical deliverables at any stage gate process to support the target metrics. This paper presents an alternative method to estimate the cost which is applied to offshore platforms through machine learning as it compares the performance of two different machine learning algorithms: Support Vector Regression (SVR) and Artificial Neural Network (ANN). A comparison between SVR and ANN method is set to show the fundamental algorithm of each and evaluate the comparison between them to show the advantages and limitations. Performance of the algorithm is measured by the RMSE value, which in this case the predicted platform cost by the model against the actual cost observed. The results show that despite the limitation in the amount of data, the algorithms exhibit high performance with both SVR recording average error or deviation of platform cost of RM 27 million, which is the lowest RMSE value, and ANN recording RM 36 million deviation. In addition to that, both algorithms used in predicting the platform cost have returned highly accurate results when comparing them to actual platform cost generated using conventional method.