Data-driven evolutionary algorithm for oil reservoir well-placement and control optimization

Well placement and control scheme optimization is crucial for hydrocarbon, groundwater geothermal development, generally involves a large number of discrete correlated decision variables. Meta-heuristic algorithms have showed good performance in solving complex, nonlinear non-continuous problems. However, numerical simulation runs are involved during the process. In this work, novel efficient data-driven evolutionary algorithm, called generalized differential algorithm (GDDE), proposed to reduce on well-placement Probabilistic neural network (PNN) adopted as classifier select informative promising candidates, most uncertain candidate based Euclidean distance prescreened evaluated with simulator. Subsequently, local surrogate model built by radial basis function (RBF) optimum surrogate, found optimizer, simulator accelerate convergence. It worth noting that shape factors RBF PNN optimized via hyper-parameter sub-expensive problem. The results show study very problem two-dimensional reservoir joint Egg model. convergence curves reveal significantly reduced around 20 percent process comparison conventional algorithm. can help better making computationally expensive simulation-based