IIC Journal of Innovation 16th Edition | Page 56

Design and Implementation of a Digital Twin for Live Petroleum Production Optimization
Table 1 : Details of simulation parameters and relationship to field data . Types of Simulation Parameters
Knowns Unknowns Approximations
Field data type
Sensor data and Metadata
Field data unavailable Sample field data available . Live data unavailable
Simulation input type
Static data and dynamic data
Static data and dynamic data
Static data
Example sensors
● Injection pressure
● Gas injection rate
● Design operating pressures
● Well deviation
● Sales gas pressure
● Produced gas to oil ratio
● Produced water cut
● Reservoir pressure
● Productivity index
● Tubing friction factor
● Gas injection depth
● Solution gas to oil ratio
● Location of pipe sensors
● Fluid specific gravities
● Valve pressure loss coefficients
● Less sensitive parameters ( such as ambient temperature )
Range of input
Exact operating range as observed in field data
Wide range of input based on physical possibility
Narrow range of input
The commercial physics-based simulator employed in the case presented is a transient simulator . This implies that some of the input parameters can be entered as a time series , and the response to changes in these parameters can be obtained as a time series output . This detail is a key component of the simulation setup because the startup of a simulation and the post-processing associated with writing outputs to files has a non-trivial overhead on the simulation schedule . To reduce the input-output overhead and transition associated delays between simulations , dynamic parameters such as Reservoir pressure , Productivity index etc . are stepped through time series .
In the current case , 240 simulation cases were connected together as a time series , and input into a simulation queue as a single item . Each item read from the queue generates a commercial simulator input file with 240 cases in series . The number of cases per simulation file has been heuristically arrived at as an optimum scenario . If a very large number of cases are input per file ,
IIC Journal of Innovation - 51 -