Simulation based optimization of petroleum production problems: Development of a special purpose B&B for a non-convex MINLP
MetadataVis full innførsel
This work is concerned with the upstream and operational planning of petroleum fields. The short-term production optimization problem is modeled here as a simulation based non-convex MINLP problem. A special purpose B&B algorithm fit for manipulation is developed to allow sophisticated operations on each node of the B&B tree, such as running heuristics or implement non-convexity measures designed specifically for this problem.Two types of heuristics are included to help obtain good feasible solutions quickly. A generic feasibility pump heuristic is modified to fit non-convex production allocation problems, creating high quality solutions. This heuristic can also function as a standalone solution method. In addition, two problem specific heuristics are made based on problem case knowledge and data analysis, giving good solutions fast. Specific non-convexity measures are included to avoid eliminating interesting parts of the solution space and help push towards finding the global solution. Pruning poses a challenge when solving non-convex problems with B&B, therefore negative gaps are allowed in order to assess the impact on the solution found. Further, the influence of including several starting points is studied. Two different versions of the algorithm with different emphasis are designed, the first one on solution time, and the second more sophisticated one on solution quality. The results show quite a dramatic increase in solution time for the sophisticated version. The proposed algorithm is compared to other existing methods for optimizing the production allocation problem. In comparing emphasis has been on the tradeoff between solution time and quality, in addition to integration of the method into the daily work process of operating an offshore production platform. The version focusing on solution time is considerably faster, although both versions prove better than the other methods in terms of solution quality.