Neural-net-based Modeling Used In The ASP Complicated Flooding Systems

Based on neural networks modeling non-linear functional relationships between the petrophysical properties of rocks and their electrical properties of the ASP complicated flooding systems. The learning algorithm is a kind of variable metric method. The algorithm has fast convergence rate and good precision. The research result shows that the method is suitable for the modeling and identification of nonlinear systems.At present most of the waterflooding exploitation oil field in our country are in the anaphase, how to increase the oil field outputs and improve the recovery ratio are urgent affairs. The distribution of the remnant oil is complex, and discriminating the distribution and type of the remnant oil is the key to select the method of exploiting the remnant oil. The main index which indicates the distribution of the remnant oil is oil saturation. We can only measure the resistance of the rock to detect the remnant oil in the rock because of being difficult to measure it directly. This is why we study the relationship between resistance of the rock and oil saturation. This paper used the neural network method to study the relationship between resistance of the rock and oil saturation, and constructed a mathematical model.People mainly apply the back propagation algorithm (BP algorithm) based on gradient descent direction of search to train the weights of the multi-layered feedforward neural network. The drawback of BP algorithm is that it convergents too slowly during the weights training of neural network, and it is difficult to reach the satisfied precision. This paper applies the variable metric learning algorithm to train the weights of neural network. The parabola interpolating method is used in the optimization of learning rate r\. This method is used to model the non-linear mathematical model of the functional relationship between the resistivity and oil saturation of the ASP complicated flooding systems. The result of this research shows that the algorithm has fast convergence and high precision.