【摘要】 压裂水平井能大幅提高低渗及特低渗油气藏的最终采收率，具有重要应用前景，而其裂缝参数反演是压裂评价和产能预测的前提。试井分析方法是裂缝参数反演的重要手段之一，但目前的方法以人工反演为主，其主观性和多解性强，为裂缝参数反演结果带来了极大不确定性。为解决人工反演耗时长、受试井工作者主观因素影响强、不同试井工作者水平差异大等问题，利用特征点法及改进卷积神经网络，基于压裂水平井油藏的半解析解形成一套智能参数反演方法。对压力数据进行归一化处理，先通过训练好的卷积网络反演改造区渗透率、无因次导流能力、裂缝半长及裂缝间距，再通过特征点法得到无因次井筒储集系数及表皮系数，验证结果表明无因次井筒储集系数、表皮系数、改造区渗透率、无因次导流能力、裂缝半长和缝间距的反演值与真实值的相对误差分别为5.00%、6.01%、7.96%、6.10%、4.70%、0.25%,平均相对误差为5.00%,反演曲线与真实曲线符合程度良好，验证了该方法的可靠性。对实际现场的两次压力恢复数据进行智能反演，并与人工反演进行对比，结果显示智能反演的无因次井筒储集系数下降39.37%、表皮系数增加15%、改造区渗透率下降26.53%、无因次导流能力下降19.45%、裂缝半长下降12.82%、有效井长下降30.29%,符合现场实际情况，从而验证了该方法的实用性。更多还原

【Abstract】 Hydraulically fractured horizontal wells significantly enhance the ultimate recovery factor of low-permeability and ultra-low-permeability reservoirs, demonstrating critical application prospects in the development of such reservoirs. The inversion of fracture parameters serves as a prerequisite for fracturing evaluation and productivity prediction. The well test analysis method is one of the important methods for fracture parameter inversion, but the current approach mainly relies on manual inversion, which is highly subjective and suffers from non-uniqueness, introducing great uncertainties into inversion results. To address issues such as long inversion time, strong influence by subjective factors from well test workers, and significant disparities in professional proficiency, an intelligent parameter inversion method is developed based on the semi-analytical solution of fractured horizontal reservoirs, integrating the Feature Point Method and improved Convolutional Neural Network(CNN). The pressure data were normalized, and retrofit zone permeability, dimensionless conductivity, fracture half-length and fracture spacing were first inverted by the trained convolutional network, and then the dimensionless well reservoir coefficient and epidermal coefficient were obtained by the feature point method. The verification results show that the relative errors between the inverted and the true values of the the dimensionless wellbore storage coefficient, skin factor, stimulated zone permeability, dimensionless conductivity, fracture half-length, and fracture spacing are 5.0%, 6.01%, 7.96%, 6.1%, 4.7% and 0.25%, respectively, with an average relative error of 5.00%. The inversion curve exhibits excellent agreement with the true curve, confirming the method’s reliability. Field applications demonstrate that for two sets of pressure buildup data, the intelligent inversion reduces the inversion time by 39.37% compared to manual methods. The skin factor increases by 15%, stimulated zone permeability decreases by 26.53%, passive diversion capacity drops by 19.45%, fracture half-length shortens by 12.82%, and effective wellbore length decreases by 30.29%, which aligns with field realities, verifying the method’s practical applicability.更多还原