一种用于油气储量评估中渗透率预测新模型

A new model for permeability prediction in appraisal of petroleum reserves

致密砂岩储层因受成岩作用和裂缝分布等地质因素严重影响,其渗透率值很难用带有储层其他参数的显式来准确求取.根据测井解释理论,渗透率是储层地质因素的一种综合影响,从而也能看作是多种测井曲线的一种综合响应,可通过拟合测井曲线来预测.LightGBM模型在数据拟合方面表现出众,其计算效率被证明比传统拟合模型的更高,为此本文采用该模型来预测渗透率.由于该模型在建模时使用了较多的超参数,难以保证预测结果为最优,所以本文采用PSO技术对其进行优化,进而提出PSO-LightGBM.本文以姬塬油田西部长4+5段致密砂岩储层为验证对象,并通过设计两个实验来验证提出模型的预测能力.实验结果显示提出模型的预测结果误差和PSO-XGBoost的误差非常接近,都为最小,但PSO-XGBoost的耗时却约是提出模型的21倍.实验结果证明PSO-LightGBM能够在不失精度的情况下快速预测致密砂岩储层渗透率,是一种高效的渗透率拟合预测模型,在测井解释智能化研究方向上具有推广应用价值.

Since the status of tight sandstone reservoirs is universally impacted by many geological factors such as diagensis and fractures,permeability of the reservoirs is difficult to be determined accurately via a certain explicit expression which consists of other reservoir parameters.According to the theory of logging interpretation,permeability is a kind of comprehensive response on many geological factors of the reservoirs,and then also could be regarded as a comprehensive response on many logs,implying it could be analyzed by logs in a fitting way.LightGBM is an excellent model in the aspect of fitting,and its computational effectiveness has been proved better than those of other classic fitting models,thus selected as a predictor for permeability.However,LightGBM employs many hyper-parameters during modeling stage,resulting in a problem that the quality of predicted results cannot be guaranteed.In view of that,PSO technique is adopted to optimize LightGBM,so that PSO-LightGBM is proposed.The tight sandstone reservoirs of member of Chang 4+5 in western Jiyuan oilfield are validation targets,and two experiments are designed to validate prediction capability of the proposed model.Validations manifest that the predicted results produced by the proposed model and PSOXGBoost have the similar errors,all of which are the smallest ones,while the computational time of PSO-XGBoost is approximate 21 times longer than that of the proposed model.The validation information demonstrates that based on the guarantee of accuracy of predicted results,the proposed model has capability to rapidly figure out permeability,presenting it is high-effective on permeability prediction of tight sandstone reservoirs and also has better generalization and applicable prospect in the field of AI study of logging interpretation.