実油田シミュレーションモデル上でのニューラルネットワークを用いた原油生産計画の最適設計

抄録

This paper investigates the applicability of Artificial Neural Networks (ANN) in the optimization process of lifting rate allocation for ESP wells. In the optimization, oil recovery is maximized according to the production performance predicted using a reservoir simulation of a real oil field.
Lufeng 13-1 Oil Field in the South China Sea has strong natural bottom-water drive and its development has reached a mature production stage. All production wells are currently operated with Electric Submersible Pumps (ESP) and the averaged water cut has exceeded 90%. The limited handling capacity of the surface processing facilities cannot allow all the wells to lift liquid at their ESPs' maximum rate. The lifting rate allocation to the wells, therefore, needs to be optimized for maximizing the oil recovery in a certain production period.
Generally, a large number of reservoir simulation runs are required for solving this type of optimization problems since all possible cases of lifting rate allocations need to be examined. In this study, oil recovery volume was estimated with a trained ANN on the basis of allocated liquid rates at the wells as the initial screening. The ANN was constructed with back-propagation method treating the results of 100 simulation results as the training data sets. Two patterns of training data sets were examined; one with random rate allocation and the other with systematic rate allocation. Blind tests for the estimation accuracy presented that the ANN trained with the systematic data sets showed better results than that with the random ones. Oil recovery factors under all the possible cases with different liquid rate combinations were estimated using the trained ANN. The top fifty cases were selected for the final examination by numerical simulation. The best case yielded 5.5% increase of produced oil volume from the base case, in which lifting rates were equally reduced to 89% of the maximum lifting capacities to meet the facility capacity. The investigation results demonstrated that the efficiency of the optimization was remarkably improved with the use of ANN on the determination of optimum liquid lifting rates in terms of oil recovery.