Rock particle segmentation is essential for identifying thin sections accurately, yet tuning deep learning models for this task is challenging. Existing studies have attempted auto-tuning techniques but face three limitations, including limited search space, inefficient search, and high demand for storage space. In this paper, we introduce HyperRPS, a novel multi-fidelity hyper-parameter tuning framework tailored for rock particle segmentation, which addresses these issues simultaneously. First, by augmenting the search space with diverse hyper-parameters, we unleash the potential of the segmentation model. Second, to accelerate the hyper-parameter search, we propose a tuning algorithm leveraging multi-fidelity surrogate fitting and time constraint modeling. Third, we develop a storage management technique to reduce the space requirements for asynchronous scheduling. Experimental results on both public and rock particle segmentation datasets demonstrate HyperRPS’s superior performance. Specifically, on the rock particle segmentation dataset, HyperRPS achieves notable improvements of 8.26% and 2.78% on average precision when tuning Mask2Former (Swin-B) and Mask2Former (Swin-L), respectively, outperforming existing state-of-the-art methods.

The rapid integration of artificial intelligence into oil and gas exploration and development offers transformative opportunities within the context of the global energy transition. This article highlights the key advancements and challenges in artificial intelligence applications. Machine learning algorithms enable data-driven shale sweet spot prediction, overcoming the limitations of traditional methods by capturing complex controlling factors. Intelligent core image analysis, leveraging computer vision and foundation models, enables automatic mineral identification, pore analysis, and rock structure characterization, thereby providing a comprehensive framework for microscopic reservoir appraisal. Physics-informed neural networks address the limitations of purely data-driven reservoir simulation by embedding governing seepage equations into their loss functions, thereby ensuring physical consistency and improved generalization. Multimodal architectures significantly enhance unconventional shale gas production prediction by integrating geological heterogeneity with dynamic production behavior, leading to more accurate and stable forecasts. Collectively, these AI-driven approaches underscore the importance of combining domain expertise, multi-source data, and physics-aware modeling to achieve efficient and intelligent oil and gas development.

The enrichment and migration of reservoir fluids are affected by the types and structural parameters of shale pores, which are significant components of shale reservoir evaluation. Due to issues with current assessment methods, including high subjectivity, low efficiency, and low degree of quantification, it is challenging to address the urgent needs of quick and accurate examination of shale samples. Based on this, an intelligent characterization method for shale pore structure based on semantic segmentation is proposed. Firstly, the two-dimensional gray images of shale are obtained by scanning electron microscopy (SEM) and multi-scale acquisition and processing software (MAPS). Secondly, these images are annotated by the rock mineral identification experts, and divided into organic pores, inorganic pores, fractures and organic matter. Then, a combination network Shale Seger and its training paradigm for shale pore structure analysis tasks is innovatively proposed, an intelligent recognition model of shale pores based on deep learning is constructed, as well as, a pore recognition scheme of super large image based on multi-view mosaicisis established to extract pore from MAPS images. Lastly, intelligent characterisation of pore structural features is achieved by applying image processing techniques. As of right now, this study has produced a quantitative analysis of the Gulong shale's pore structure that can statistically compute pore structure parameters like pore diameter, apparent pore ratio surface, shape factor, and so forth, as well as automatically determine the area proportion of each type of pore based on pore edge extraction and type recognition. In addition, the technique described in this article can also be extended to the evaluation of CO₂ composite fracturing, through which the change of microstructure characteristics before and after CO₂ composite fracturing can be quantitatively compared.

Imaging logging has become a popular means of well logging because it can visually represent the lithologic and structural characteristics of strata. The manual interpretation of imaging logging is affected by the limitations of the naked eye and experiential factors. As a result, manual interpretation accuracy is low. Therefore, it is highly useful to develop effective automatic imaging logging interpretation by machine learning. Resistivity imaging logging is the most widely used technology for imaging logging. In this paper, we propose an automatic extraction procedure for the geological features in resistivity imaging logging images. This procedure is based on machine learning and achieves good results in practical applications. Acknowledging that the existence of valueless data significantly affects the recognition effect, we propose three strategies for the identification of valueless data based on binary classification. We compare the effect of the three strategies both on an experimental dataset and in a production environment, and find that the merging method is the best performing of the three strategies. It effectively identifies the valueless data in the well logging images, thus significantly improving the automatic recognition effect of geological features in resistivity logging images.