The objective of this study is to predict the last 10% of the hydrocarbon production data from historical information, expressed as time series.
The presentation will illustrate how two machine learning algorithms can be used and compared to select the best prediction of the last 10% of hydrocarbon production for a field/play.
The hydrocarbon production data for this dissertation is complete, i.e., the information spans from the beginning of production data till the end. I will use the Volve dataset from Equinor and the combined onshore (land) and offshore (marine) production from Texas that was accessed from the Railroad Commission of Texas.
The aim is to train 90% of the hydrocarbon production data to predict the last 10%, using two Python algorithms ARIMA (the Autoregressive Integrated Moving Average) and LSTM (Long Short-Term Memory). The algorithm that is the closest to the last 10% will be selected for the result comparing the real data versus the modelled data. An analysis of both methods will be detailed and some best practices for Machine Learning will also be described.
I recently completed a Master’s degree in Data Science from Sussex University in the United Kingdom.
I spent the most of my career was as a program manager for Schlumberger Research and Software Development for Seismic Depth Imaging, including Earth Model Building, and Seismic Reservoir Characterization.
I have a PhD in seismic depth imaging and initiated a depth domain inversion, as opposed to the time domain inversion for reservoir characterization from Total and Institut Franҫais du Pétrole. I enjoy connecting domains that were previously disjointed.