Closed-Loop Field Development Optimization: Reduce Risk when Drilling New Wells

In oil and gas reservoir management, drilling new wells is typically the most expensive part of operations. In practical settings, there are limited number of rigs, and after drilling each group of new wells, new spatial & temporal data become available. Can we learn from the new data after drilling each new well to make better decisions about the next wells? It turns out that we need a physics-based supervised learning framework here as system parameters are nonlinearly related to the new data.

Closed-loop field development (CLFD) is the comprehensive reservoir optimization framework where the data from each new well drilled is incorporated to ultimately update decision parameters for next wells. CLFD involves three major steps (1) solving an optimization problem for the number, type, location and controls, and drilling sequence of new wells based on “current” geological knowledge, (2) sequentially drilling new wells and collecting data, and (3) model calibration based on all data, where geostatistics & inverse problem methods are utilized. This process is repeated until the optimal number of wells is drilled.

At each step of CLFD, the overall future development/operating plan is optimized; i.e., the location for the next well is determined with the knowledge that it is one well within a sequence. Results showed that the use of CLFD increased the NPV for the true model by 10%-70% relative to that achieved by optimizing over a large number of prior realizations.

Closed-loop reservoir management (CLRM) which is the optimal operation of existing wells, is in fact, a subset of CLFD. Optimization step of CLRM only includes continuous operational settings of existing wells. The optimization step of CLFD, however, is very general and depending on the resources, one can formulate and optimize decision parameters of infill wells (number/type of new wells, their location and control, and drilling sequence) together with continuous operational settings of existing wells. The model calibration step of CLFD is also more general as it involves integrating both spatial & temporal flow data. In CLRM, by contrast, only integration of temporal flow data is considered.