Design and Implementation of a Digital Twin for Live Petroleum Production Optimization
that may overload the system memory and output large files , this may lead to a system crash . A smaller number of cases input per simulation file results in underutilization of the transient capabilities of the simulator , and loss of time in the simulation input-output process .
Simulation output :
Each simulation file generates a time series output for the 240 cases as described in the simulation input section . The simulation output is updated to the No-Sql database item associated with its input . Python scripts are set up to extract the data from a No-Sql database , and post-process the time series data and extract individual case outputs corresponding to each input . These simulation outputs are matched with field observations to identify simulations representing likely operating states of the well . The “ unknown ” parameter values in simulation cases associated with output values irrelevant to field data are disincentivized in further rounds of simulation as a part of the inverse modeling process . A rough visual representation of the response comparison between field data and simulation is presented in Figure 11 .
Fig . 11 : Response comparison between field measurement and simulation output .
In the use case presented in this paper the response variables being matched with field data include :
● Oil production rate
● Wellhead pressure
● Downhole pressure
In the inverse modeling process subsequent to the simulation , further constraints are implemented on the relationship between simulation output and field responses including the historical trends that identify the likelihood of a combination of unknown parameters . Thus , the operating state of a well at a given time is estimated through this process , and this knowledge is used to generate set-point recommendations .
- 52 - March 2021