A new fracture prediction method by combining genetic algorithm with neural network in low-permeability reservoirs

摘要

Natural fractures in low permeability reservoirs are the dominant flow paths for the whole flow system. It is the difficult point to identify and predict natural fractures by using conventional logging data. Back propagation neural network is an effective method for natural fracture prediction, but it has the defect that convergence rate is slow and the objective function easily falls into the local minimum value. The genetic algorithm population search method can realize the optimal allocation if given network weights and threshold, which can improve back propagation neural network's defect of over-reliance on the gradient information and achieve the minimum global error. In this paper, by analyzing in-depth the relationship between observed fractures in the cores and well logging data, we proposed a new method of fracture identification in terms of deep-shallow laterolog curves as well as their amplitude difference and micro-electrode logging curves as well as its amplitude difference. The method was verified by oilfield dynamic monitoring data. In the application of the method to Xinli oilfield, the optimized standard samples were chosen to train the designed genetic algorithm back propagation neural network, and then the genetic algorithm-back propagation neural network for Xinli oilfield was established to predict fractures in the target reservoir. The prediction has good consistency with the oilfield's actual development performance, which proves the reliability of the new method.