基于机器学习的井位及注采参数联合优化方法

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

针对水驱油藏传统优化方法过于依赖人为经验及顺序优化难以求得全局最优解等问题,提出一种基于机器学习的井位及注采参数联合优化方法.基于水驱油藏特征,利用随机森林算法筛选影响注水开发效果的主控因素,以井网形式、产量和注采比等作为输入参数,累计产油量为输出参数,通过流线数值模拟方法构建机器学习预测样本集,综合径向基函数神经网络预测水驱开发效果.基于粒子群算法建立优化数学模型,以最大化产油量作为目标对井网形式和注采参数进行联合优化求解.结果表明,与传统优化方法相比,联合优化方法能够自动同步优化井网形式、井位和注采比等参数,优化后开发效果提升约12%,为水驱油藏的智能高效开发奠定基础.

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.