Oil and Gas Well Loading

September 17, 2016
Experiment to predict Oil and Gas well liquid loading
&nbsp; <img style="float: right; display: inline;" src="http://www.electronhacks.com/wp-content/uploads/2016/09/123_1233474900.jpg" width="347" height="629" align="right" /> In the lifecycle of a Gas well “Lift” often becomes an issue. When the well is first drilled often pressure is sufficient to lift liquids (Oil and Water) but as wells age they can reach stages where the gas velocity is not always great enough to lift the liquid. In this exercise we use a specific example of “well loading” to train Azure Machine Learning. Machine Learning can then return to us the probability of the well having “Loading” issues. If actions are taken quickly to mediate loading issues improved efficiency and reduced chance of costly intervention techniques. <a href="https://www.youtube.com/watch?v=Zy3l7gNg7Pg">The complete video tutorial for this project can be found here.</a> This trend illustrates well loading. In fact this is the exact data we feed to Machine Learning later. Green is pipeline pressure, you can se an event of high pressure caused the well to go into a spiral of loading problems. Loading can generaly be detected by the simultaneous decrease in production and tubing pressure (red and black) while casing pressure rises (blue). <img src="http://www.electronhacks.com/wp-content/uploads/2016/09/Well-loading-example.png" /> &nbsp; The data is actually fairly simple. feed in the pressures and flow rate along with averages. Keep in mind that Machine Learning won’t be able to look back into history so we have to intentionally send the averages. Also for training we send in a true or false column for classification. We are giving the correct answer to Machine Learning and after training is complete it should be able to come up with it’s own classification. <img src="http://www.electronhacks.com/wp-content/uploads/2016/09/WellLoadingData.png" /> <a href="https://www.hackster.io/fileark/detecting-well-liquid-loading-with-azure-iot-ml-and-pi-68aa4a">Here is a write up of the overarching project on Hackster.</a>