Joint optimization method of well location and injection-production parameters based on machine learning

Abstract

To solve the problems of excessive reliance on human experience in traditional optimization methods and difficulty in obtaining the global optimal solution for sequence optimization in water-flooding reservoirs, we propose a joint optimization method of well location and injection-production parameters based on machine learning theory. Firstly, based on the characteristics of the water-flooding reservoir, the random forest algorithm is used to screen the main controlling factors affecting the oil production effect of water flooding. Then, taking well pattern form, production, injection-production ratio, etc. as input parameters, the cumulative oil production as the model output parameters, the machine learning prediction sample set is constructed through streamline numerical simulation method, and the comprehensive radial basis function (RBF) neural network is utilized to predict the development effect of water flooding. Finally, the particle swarm optimizer algorithm is applied for the joint optimization for well pattern and injection-production parameters by maximizing oil production as the optimization goal. Results show that compared with the traditional optimization methods, the new joint optimization method could automatically and synchronously optimize parameters, including well pattern form, well position, injection-production ratio, etc. The optimization scheme is better than the original ones. The water flooding performance is improved by about 12% using the new optimization method, laying a solid foundation for the intelligent and efficient development of water flooding reservoirs.