Design and Implementation of a Digital Twin for Live Petroleum Production Optimization
historical data that a well has . Lapse Time is defined as a period of time we don ’ t receive any signal from the well , therefore Lapse Time / Monitor Time ratio is to determine the proportion of data availability over the life of a well . Similar to Lapse Time / Monitor ratio , Zeros ratio is used to see the actual variance in data . Frequency of data is also a key factor to the success of our models , the more granular data we have the better our models ’ performances are . A set of thresholds will be applied on each of the properties mentioned above . Sensors that meet these thresholds will be examined to see if they are sufficient for our models . Wells that include these qualified sensors are included in a cohort .
An example of Monitored Time threshold is shown below where Monitored Time threshold is set at 90 days which consequently reduces the number of wells in the cohort from 217 to 56 wells .
Fig . 5 : Cohort selection based on monitored time .
Cleanup : Data coming from several wells over a period of time may include various challenges such as abnormal states of operation of the well ( for example : shut-in , maintenance job , inconsistent performance ), faulty signal , signal lapses , signal names that are inconsistent from convention . It is necessary to identify these inconsistencies . Some of these can be identified and eliminated systematically , such as identification of dummy values of a signal , removal of outliers , or identifying that a well is shut-in . There are other abnormalities that require human review , for instance , a signal that was named inconsistently , or a metadata element that does not fit the template .
Transformation : It is not beneficial to evaluate meaning out of data , or generate simulations at the frequency of live data . Signals may be recorded at different timestamps and variable frequencies . In the work presented in this paper , data was resampled to a daily frequency to match the frequency of the production rates of the wells . Further , the variation related features such as the oscillation frequency / wavelength , level of stability when compared to the normal signal was captured through a rolling normalized coefficient of variance . Unstable states were
- 44 - March 2021