Abstract
The main objective of this project is to develop a model that, based on the data of previously drilled wells, allows the user to evaluate the mud plan and identify the probability of a stuck pipe event occurrence. This is done, by using statistical predictive models, particularly in this case, neural network models.
The proposed model takes the input data from the Volve’s dataset. The process starts with the pre-processing of more than 24.000 rows of data from the daily drilling operation reports of 30 wells. The pre-processing of the data consists of identifying the cases where the operation ran across a section successfully or when it resulted in a stuck pipe event. Depending on whether there is or not a stuck pipe event, each row will be classified indicating the chance that a stuck pipe event occurs based on the geological and the operational conditions.
In order to ease the processing of the 24.000 rows, the project workflow implements a Natural Language Processing model, which, based on an initial dataset of 1.200 rows, reads the remaining 22.800 identifying whether the row content is relevant or not for stuck pipe identification. Once all the rows are processed, the rows which represent a stuck pipe event, and those who identify a clean operation run are selected and integrated into the final dataset.
Once the data is processed, the resulting dataset consists of around 600 rows. After some data transformation, this dataset will be the input for the neural network model in the training and validation phases. Once the model is built, the model is tested with the validation dataset in order to evaluate its accuracy.
The trained model shows high accuracy, higher than 84% on the test data. After the model is trained and validated, it’s implemented into a frontend which will serve as the front user interface where he can upload the mud plan parameters, and obtain a stuck pipe event chance evaluation. Having these results, the user can improve the mud plan in order to reduce the chance that an event of this type occurs.
The main advantage of this methodology is that it allows to include historical data from the other drilled wells, this way, boosting the mud plan design with real data and reducing the chances that a stuck pipe event, that account for a high percentage of NPT’s during drilling operations occur. Also, by the use of these statistical models, applied to a specific field, it will allow to intrinsically take into account variables such as geomechanics properties in the model.