Applications of Neural Networks in Multiwell Field Development

Abstract

Abstract A common field development task is the optimal placement of the wells. One of the key parameters in the success of a new well is its physical location in the reservoir. When several wells are planned hundreds even thousands of placement combinations must be considered in order to optimize specific production goals. The large number of degrees of freedom makes using a standard reservoir simulator to evaluate all these placement combinations prohibitively time consuming and expensive. This paper demonstrates a neuro-simulation technique that forms a bridge between an accurate reservoir simulator and a fully-trained predictive artificial neural network (ANN). The paper outlines how the coupling of soft and hard computing techniques allows for a rapid and efficient selection of well locations while maintaining reasonable accuracy. In this method, several key well scenarios are selected by engineering judgement and/or randomly. These scenarios are then evaluated using a numerical reservoir simulator. The simulation results form the basis for training an ANN. When the ANN is trained for a specific reservoir configuration, it forms a fast predictive tool for optimizing the locations of the new wells in the reservoir. Thousands of possible scenarios are evaluated using the ANN with an insignificant computational effort. Then, all the predictions are ranked and sorted to provide the best possible scenario for placing the new wells. It is the combination between ANN and genetic concepts that leads to an efficient optimization process. Simulated reservoir examples with various shapes, properties, and number of existing new wells are considered in the paper. The results of the paper demonstrate the effectiveness of neuro-simulation by reducing the computational effort and maintaining accurate predictive capabilities. The paper also presents the guidelines for using neuro-simulation to optimize the placement of the wells.