Application of Artificial Intelligence in the Petroleum Industry: Volume Loss Prediction for Naturally Fractured Formations
Recently, artificial intelligence has gain popularity in the drilling industry since more wells are drilled in hostile environments. One of the most difficult problems have been encountering the drilling operation is the problem of lost circulation. The complexity of the lost circulation problem is due to the interaction between the parameters that are causing this issue. The aim of this work is to create artificial intelligence models to predict lost circulation, equivalent circulation density (ECD), and rate of pentation (ROP) prior to drilling for naturally fractured formations.
Lost circulation events from 500 wells were collected and analyzed to comprehend the impact of each drilling parameter on lost circulation. The data were cleaned and outliers were removed. Partial least square (PLS), a supervised machine learning algorithm, was utilized to create three models to estimate mud losses, ECD, and ROP before drilling. The models went through a cross-validation process to validate them. In addition, the models were tested with new data that were not used in the process of creating the models.
The results showed that the three models can predict mud losses, ECD, and ROP within a reasonable margin of error. Testing the models with new data of 30 wells drilled showed that the models' predictions closely track the actual values from the real data. Moreover, the new models were compared with previously developed models for naturally fractured formations. The new models showed better predictions for the actual values than the previously developed models, suggesting the ability of the new models to predict mud losses, ECD, and ROP within an acceptable error. In addition, a 10% sensitivity analysis was conducted for all models to quantify and understand the effect of each parameter on every model. Mud weight (MW) had the highest impact on the ECD and mud losses models revealing that in order to minimize mud losses and ECD, the first action should be trying to use as low MW as possible. On the other hand, weight on bit (WOB) showed the highest positive influence on the ROP model and total flow area (TFA) of the nozzles showed the highest negative impact on the ROP model. Thus, the models developed in this study can be used to regulate the drilling parameters to minimize mud losses.
The methodology used in this study to develop estimation models for mud losses, ECD, and ROP can be applied to create predictive models in other formations if the required data are available.
A. T. Al-Hameedi et al., "Application of Artificial Intelligence in the Petroleum Industry: Volume Loss Prediction for Naturally Fractured Formations," Proceedings of the SPE/IATMI Asia Pacific Oil and Gas Conference and Exhibition (2019, Bali, Indonesia), Society of Petroleum Engineers (SPE), Oct 2019.
The definitive version is available at https://doi.org/10.2118/196243-MS
SPE/IATMI Asia Pacific Oil and Gas Conference and Exhibition 2019, APOG 2019 (2019: Oct. 29-31, Bali, Indonesia)
Geosciences and Geological and Petroleum Engineering
Keywords and Phrases
Forecasting; Fracture; Gasoline; Infill drilling; Learning algorithms; Petroleum industry; Respirators; Sensitivity analysis; Supervised learning; Well testing, Drilling operation; Drilling parameters; Equivalent circulation density; Hostile environments; Lost-circulation problems; Naturally-fractured formations; Partial least square (PLS); Supervised machine learning, Well drilling
International Standard Book Number (ISBN)
Article - Conference proceedings
© 2019 Society of Petroleum Engineers (SPE), All rights reserved.
01 Oct 2019