面向油藏动态分析的规则引擎研究与实现

发布时间：2018-11-01 20:14

【摘要】：油藏动态分析是油田开发中一项重要工作,传统动态分析辅助软件通常将分析规则以“硬编码”的方式嵌入系统,具有可扩展性和可维护性差的固有缺点,同时无法保存油田开发中的重要知识。当动态分析的指标发生变化时,需要重新修改程序代码,这样不仅工作量大而且容易出错,软件维护成本高。另外,油藏动态分析涉及业务范围广泛,针对每项分析内容,需要重复编写低水平代码,软件复用性差。规则引擎作为人工智能领域的一项重要研究成果,是一个可嵌入系统任意位置具有智能推理的小组件,能够复用到不同系统中,降低了软件的开发成本。同时,将分析逻辑采用规则的形式单独存放,便于系统维护,也有利于总结和推广油田开采过程中积累的经验知识。本文以油藏动态分析和规则引擎为研究对象,详细探讨了适用于油藏动态分析的规则引擎的实现及应用。首先,从业务层面和系统层面研究油藏动态分析知识的特点,提出了“事实模板-条件元素-推理规则”三层的知识表示模型,为基于规则的推理和知识的存储管理提供模型基础。其次,详细介绍了RETE算法的原理,针对RETE算法本身网络结构,通过对Beta节点左右存储区分别建立索引的方式提高模式匹配的效率;基于实际应用中动态分析具有过程性的特点,通过建立虚拟工作内存区并逆向加载事实的方式降低匹配过程中内存使用率。然后,基于开源的规则引擎CLIPS6.3版本,通过增加功能和提高性能两方面对其进行改造得到适用于油藏动态分析的规则引擎。在功能上,通过添加可视化知识管理模块,弥补了CLIPS知识管理方面的不足;在性能上,通过改进的RETE算法解决了油田大数据量推理效率不高的问题。最后,将实现的规则引擎应用于某采油厂实际项目中,证明了实现的规则引擎的可行性和有效性;通过与传统动态分析方案的比较可知基于规则引擎的新方案以少许性能损失获得了更好的可扩展性和可维护性。
[Abstract]:Reservoir performance analysis is an important work in oilfield development. The traditional software for dynamic analysis usually embed the analysis rules into the system in the way of "hard coding", which has the inherent shortcomings of poor scalability and maintainability. At the same time, the important knowledge of oilfield development can not be preserved. When the index of dynamic analysis changes, it is necessary to modify the program code again, which is not only heavy workload but also error-prone, and the cost of software maintenance is high. In addition, reservoir performance analysis involves a wide range of business, for each analysis content, it is necessary to repeatedly write low-level code, software reuse is poor. Rule engine, as an important research result in artificial intelligence field, is a small component with intelligent reasoning in any location of embedded system. It can be reused to different systems and reduce the cost of software development. At the same time, the analytical logic is stored separately in the form of rules, which is convenient for the maintenance of the system, and also conducive to summing up and popularizing the experiential knowledge accumulated in the course of oilfield production. Taking reservoir dynamic analysis and rule engine as research objects, this paper discusses in detail the realization and application of rule engine suitable for reservoir dynamic analysis. Firstly, the characteristics of reservoir dynamic analysis knowledge are studied at the operational and system levels, and the knowledge representation model of "fact template, conditional element and inference rule" is proposed. It provides a model basis for rule-based reasoning and knowledge storage management. Secondly, the principle of RETE algorithm is introduced in detail. Aiming at the network structure of RETE algorithm, the efficiency of pattern matching is improved by indexing the left and right storage areas of Beta nodes separately. Based on the process characteristic of dynamic analysis in practical application, the memory utilization rate in matching process is reduced by establishing virtual working memory area and loading the facts in reverse. Then, based on the CLIPS6.3 version of the open source rule engine, the rule engine for reservoir performance analysis is obtained by improving the performance and function of the engine. In function, the lack of CLIPS knowledge management is made up by adding visual knowledge management module, and the problem of low reasoning efficiency of large amount of data in oil field is solved by improved RETE algorithm in performance. Finally, the realized rule engine is applied to the actual project of a certain oil production plant, which proves the feasibility and effectiveness of the realized rule engine. Compared with the traditional dynamic analysis scheme, the new scheme based on rule engine achieves better scalability and maintainability with little performance loss.
【学位授予单位】：东北石油大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TE331;TP18

【参考文献】