Summary
Model-based production optimization has become a promising technique in recent years for improving reservoir management and asset development in the petroleum industry. A variety of methods have been developed to address production-optimization problems. However, many solutions resulting from these methods are not ready to be accepted by (operations) engineers because they are difficult to understand and implement in practice. A typical example is the erratically oscillatory, bang-bang-type solution of well-control optimization problems. For this challenge, a regularized optimization problem is formulated in this work with two purposes: to create smooth solutions and to improve the convergence speed. Furthermore, the original, high-dimensional optimization problems can be reduced to low-dimensional ones by a principal-component-analysis (PCA) -based regularization method such that some population-based methods, including genetic algorithm (GA) and particle-swarm optimization (PSO), can be used as search engines to find optimal or improved solutions that tend to have less chance of trapping in local optima. Examples show that the methodology proposed can ensure a continuous transition over time between variables (e.g., well controls) such that the generated solution is more acceptable to the (operations) engineers. Moreover, it significantly speeds up the convergence of the optimization process, allowing large-scale problems to be addressed efficiently.