A Robust, Multi-Solution Framework for Well Location and Control Optimization

Unconventional hydrocarbon resources: geological statistics, petrophysical characterization, and field development strategies

“…In addition, clustering has been usually adopted for determining the representative set of samples for realization. Also, the Net Present Value (NPV) is usually taken as an objective function (Jesmani et al 2020;Salehian et al 2021aSalehian et al , 2021bSalehian et al , 2020Chen et al 2012;Shirangi and Durlofsky 2016;Wang et al 2012;Sefat et al 2016;Moraes et al 2019;Mohsin Siraj et al 2017;Brito and Durlofsky 2021;Lu and Reynolds 2020;Fonseca et al 2020).…”

Multi-solution well placement optimization using ensemble learning of surrogate models

“…Haghighat Sefat et al (2016) used SPSA andLu et al (2017) used StoSAG), or multi-level well placement and control problems (e.g. Li et al (2013) and Salehian et al (2020b) used SPSA). Salehian et al (2020a) recently developed a multi-solution optimization framework and showed its value in providing multiple solutions with close-to-optimum objective values.…”

Robust integrated optimization of well placement and control under field production constraints

“…However, access to the reservoir simulation source code is required to calculate the gradient, which makes them impractical for use with commercial (black box) simulators. The approximate gradient-based algorithms overcome the above issues by stochastically estimating the gradient of a black-box objective function using a reasonably sized ensemble of simultaneous perturbations of control variables, where an ensemble of less than 10 perturbations per iterations can usually provide a good balance between the gradient accuracy and computation time (Haghighat Sefat et al, 2016;Salehian et al, 2020b). These algorithms have been successfully employed to solve large-scale well placement [e.g.…”

“…Jesmani et al (2020)) is another method that can potentially alleviate this problem, especially when the number of iterations is large. A systematic approach is to tailor the realization selection process to the objective of the subsequent optimization (Wang et al, 2012;Haghighat Sefat et al, 2016;Salehian et al, 2020b). Wang et al (2012) recommended projecting all model realizations to two-dimensional space while each dimension is selected from static (e.g.…”