论文详情
回顾了油气人工智能研究进展,分析其面临的一些关键问题。将油气人工智能研究分成两个层级,即学术型油气人工智能研究和工业级油气人工智能研究,两者面临不同的问题和挑战。对于学术型油气人工智能应用场景,主要是关心算法及其相关理论应用,着重于解决智能点的局部问题;对于工业级人工智能应用场景,更多的要关心数据治理、数据集、平台、多源多尺度多模态数据融合建模、数据驱动与机理模型融合建模以及机器学习模型的可解释性等问题。针对数据驱动与机理模型融合问题,提出3种途径,即算法融合、评价方法融合、数据集融合,并给出实验验证。针对油气人工智能模型的可解释性问题,指出工业级油气人工智能必须具有可解释性,并提出初步解决方案,包括建模前、建模中、建模后的多级解释模型。最后,作者认为,探寻工业级人工智能理论和应用场景发展之路,必须厘清人工智能时代“物理世界”、“数字世界”、“人类认知世界”、“机器认知世界”和“机器正在改造的世界”之间的互动关系。
Research of oil and gas artificial intelligence can be divided into two levels,academic and industrial research,which faces different problems and challenges.Academic oil and gas artificial intelligence application scenarios are mainly concerned with algorithms and their related theoretical applications,focusing on solving the local problems of intelligent points.Industrial-grade artificial intelligence applications are mainly concerned with data sets,platforms,multi-source multi-scale data fusion modeling,data-driven and mechanism model fusion modeling,and machine learning model explanatory issues.In this study,three suggestions are put forward for data-driven and mechanism model fusion:algorithm fusion,evaluation method fusion and data set fusion,and experimental verification is given.In view of the problems of oil and gas artificial intelligence models,the author illustrates that industrial-grade oil and gas artificial intelligence must be explanatory and puts forward some preliminary solutions,including multi-level interpretation,pre-modeling,in-modeling,and post-modeling.Finally,the author suggests that,to explore the development of industrial-grade artificial intelligence theory and application scenarios,we must clarify the interaction between the “physical world,” “digital world,” “the world recognized by humans,” “the world recognized by machines,” and “the world in which machines are being transformed.”