USING ANOMALY DETECTION WITH MACHINE LEARNING BY ASSESSING THE BOTTOM HOLE PRESSURE CHANGES

Date

2023-12-08

Author

Khalil, Ahmed Magdy A. E.

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

176
views

272
downloads

Cite This

For achieving sustainability in producing from petroleum systems, it is vital to have a field model as close as possible to the real situation in order to monitor production and injection performance. In this study, it is aimed to develop a machine learning model that can be used as a representative performance indicator and coupled with an anomaly detection method. Machine learning consists of a supervised learning DNN model with four inputs and one output. Inputs are time, oil, water, and gas rates from well production. Output is the forecasted pressure performance of the well. Anomaly detection is developed by observing the predicted next day forecast and thirtieth day forecast given the previous data, after which it is compared with the actual pressures corresponding with the dates of prediction. Using this algorithm, it will help us forecast pressure or production, evaluate proposed scenarios, and identify anomaly pressures due to unaccounted reservoir characteristics between wells (like faults, permeability, or porosity changes). Additionally, the given model algorithm will be able to be tested on many different wells in a relatively short time, giving headroom to build an efficient development plan. Proposed scenarios can consist of new infill wells, changes in operational conditions, or changes in injection/production patterns. The thesis developed and tested an algorithm to forecast and find anomaly. It showed accurate and robust results that match with the results of a commercial simulator in spite of the complexity of the reservoir and the fitting procedure of the machine learning model.

Subject Keywords

Petroleum systems, Pressure Performance, Machine Learning, Anomaly Detection, Proxy Model

URI

https://hdl.handle.net/11511/107790

Collections

Graduate School of Natural and Applied Sciences, Thesis