Fast forward modeling simulation of resistivity well logs using neural networks

A neural network based approach has been developed to simulate dual laterolog and micro-laterolog responses in a 1D vertical well. Since the well site interpretation needs faster forward modeling capability, our goal is to train a neural network to a desired accuracy and substitute it for rigorous forward modeling to increase the speed of the inversion process. A modular neural network (MNN) has been trained with synthetic data. The training and testing models cover a resistivity range of 0.01 to 5000 ohm m and a thickness range of 0.5 to 50 feet. The MNN can simulate the tool response in our test case with better than 94% accuracy in most cases and the speed is almost 300 times faster than the speed in the traditional forward modeling simulation. The application of this research will be used for fast estimation of formation resistivity for 1D earth models.