Data-driven and unsupervised machine learning for comprehensive quality evaluation and intelligent optimization of multi-stage design in horizontal wells within shale reservoir

A critical strategy for shale reservoir development is the comprehensive reservoir evaluation and the fine division of fracturing grades. However, the diversity of evaluation parameters limits hydraulic fracturing optimization. Therefore, we propose an adaptive-category-gaussian-mixture-model (AC-GMM) based on a geology engineering framework, combining reservoir quality (RQ) and completion quality (CQ) to classify the composite quality index (CQI). The classification serves as the basis for an intelligent algorithm developed for fracturing design. Taking three typical wells from the Lucaogou Formation in the Junggar Basin in China as examples the following research results are summarized. First, the AC-GMM model can finely identify the fracturing grades, achieving a conformity rate of over 90 % with the field production data. Second, the paper obtains three types of fracturing grades (I, II, III) and further refines them into four grades (I, II₁, II₂, III), the grade I considers both high RQ and CQ, while grade II only regards the better of the double quality, and prioritizes the better CQ. Third, the intelligent algorithm groups similar qualities into the same stage, achieving up to 96 % intra-stage homogeneity, significantly enhancing hydraulic fracturing efficiency for long horizontal wells. Our work provides a data-driven framework for optimizing multi-stage fracturing designs in shale reservoirs.