• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文期刊>International Journal of Chemistry and Chemical Engineering >Artificial Neural Network and Inverse Solution Method for Assisted History Matching of a Reservoir Model
【24h】

Artificial Neural Network and Inverse Solution Method for Assisted History Matching of a Reservoir Model

机译：储层模型辅助历史匹配的人工神经网络与逆解决方法

获取原文
获取原文并翻译 | 示例
           
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

A typical inverse problem encountered in petroleum industries is referred to as history matching. In the early days, history matching was undertaken by manually changing sensitive reservoir parameters until a reasonable match between observed and simulated pressure and production data are obtained. In recent years, the use of assisted history matching has become prominent among academia and the industry and has significantly shortened the time required for manual history matching. However, assisted history matching requires several time consuming simulation runs which might take days to weeks, especially for large and complex reservoir models. In this paper, we presented a new approach to history matching. The approach employs 3-level fractional factorial design, artificial neural network and inverse solution methods to further reduce the computational time required and improve performance. In the inverse solution method of training a neural network architecture, the training input and output data are set to be historical and reservoir data, respectively. This allows to directly simulate the trained neural network and avoid the use of objective function and optimization algorithm. The efficacy of the developed approach was evaluated using a benchmark reservoir model case study which was originally developed for investigation of three-phase three-dimensional Black-Oil modelling techniques under the 9th SPE comparative study project. The proposed approach has required 27 simulation runs of randomly generated realizations. The historical data was generated by running the true case for a period of 900 days under constraints. The result of the case study has successfully demonstrated the efficacy of the proposed algorithm for history matching.
机译：石油工业遇到的典型逆问题被称为历史匹配。在早期，通过手动改变敏感的储层参数，在获得观察和模拟压力和生产数据之间合理匹配之前进行历史匹配。近年来，使用辅助历史匹配在学术界和行业中都突出，并且大大缩短了手工历史匹配所需的时间。然而，辅助历史匹配需要多次耗时的模拟运行，这可能需要几天到数周，特别是对于大型和复杂的储层模型。在本文中，我们提出了一种历史匹配的新方法。该方法采用3级分数阶乘设计，人工神经网络和逆解决方案方法，以进一步降低所需的计算时间和提高性能。在训练神经网络架构的逆解决方法中，训练输入和输出数据分别被设置为历史和储层数据。这允许直接模拟培训的神经网络并避免使用客观函数和优化算法。使用基准储层模型案例研究评估了发育方法的功效，该模型案例研究最初是为了调查第9个SPE比较研究项目下的三相三维黑油建模技术。所提出的方法已经需要27种模拟运行随机生成的实现。通过在约束下运行900天的真实情况来生成历史数据。案例研究的结果已成功展示了历史匹配算法的效果。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号