Neural Network Optimized by Genetic Algorithm for Predicting Single Well Production in High Water Cut Reservoir

Abstract

Accurate production prediction of single well is of great significance for efficient development of oilfield. At present, the average prediction accuracy of the traditional methods is about 70%. Although the prediction accuracy of the existing data-driven models is improved, they generally take less parameters into consideration. Taking a water-flooding oilfield as an example, this study selects 10 factors that affect the monthly production of oil Wells by comprehensively considering the three control factors of geology, development, and engineering, and establishes a prediction model of the monthly production of a single oil well by using the GRU-FNN neural network optimized by GA algorithm, and analyzes the influence of the model structure on the prediction accuracy. The research shows that compared with the traditional neural network method, the GRU-FNN model has higher prediction accuracy and lower training cost. The research results can provide reference for reservoir performance analysis.