Summary
Geosteering techniques have been widely implemented in the oil and gas (O&G) industry for well-placement operations. These techniques allow the operators to apply real-time information in precisely controlling the wellbore direction to stay within desired reservoir zones. The industry has mainly focused on technological improvements of real-time technologies.
Nonetheless, previous work has indicated that the conventional approach for making geosteering decisions leaves much to be desired. Normal geosteering operations involve drawing inferences from behind-the-bit information to ahead-of-the-bit reservoir uncertainties and making decisions under unresolved uncertainties to optimize a single objective or multiple objectives. On the basis of a large body of research, the conventional approach—which is heavily driven by intuitions, educated guesses, and approximate methods (rules of thumb)—is unlikely to identify the optimal courses of action.
Geosteering decisions are sequential decisions made under dynamic uncertainties. A sequence of well-trajectory decisions arises as the well penetrates the formation and real-time data are gathered. To optimize decision making in such an environment requires considering future decisions and uncertainties, along with the flexibility to take action as new information is learned.
In this work, we demonstrate how sequential geosteering decisions can be optimized by use of the discretized-stochastic-dynamic-programming approach (DSDP). DSDP exploits the benefits of stochastic dynamic programming to optimize multistage, interrelated decisions under uncertainties. At the same time, the computational time is minimized so it can be applied to geosteering decisions.
Through case studies, we illustrate the application of DSDP in various reservoir structures. The results suggest that the technique could significantly improve the final results of the wells, especially if the reservoir boundaries change rapidly or in a faulted reservoir. The results are expressed as substantial increases in resulting reservoir contacts as well as reductions in well-construction costs.
Finally, we illustrate and discuss the use of DSDP to assess the value of look-ahead information. We demonstrate that merely increasing the look-ahead capability (ahead-of-the-bit distance that the tool can measure) is not sufficient. The values created from look-ahead are also strongly affected by the measurement accuracy and the flexibility to respond rapidly by making large directional changes.