基于SAO-LightGBM算法的致密砂岩储层孔隙度预测方法

One method to predict porosity in tight sandstone reservoirs based on SAO-LightBGM algorithm

孔隙度是评价储层物性的关键参数,四川盆地中部NC地区钻井取心资料有限,储层孔隙度直接获取难度大,而基于常规测井资料的传统孔隙度预测方法误差大、精度低。为了明确NC地区致密砂岩气藏储层物性特征,以上三叠统须家河组四段储层为研究对象,提出了一种改进的机器学习算法SAO-LightGBM;使用该算法分析了孔隙度与地球物理测井参数之间的深层次潜在关系,指出了研究区储层孔隙度与声波时差、密度、中子孔隙度、地层电阻率和自然伽马具有较强的相关性,并基于以上测井参数建立了孔隙度预测模型。研究结果表明:①采用SAO优化算法独特的双重种群机制、高效的探索与利用策略,可以快速寻找到LightGBM的最优超参数组合,提升了模型的预测能力;②在测试数据集上,SAO-LightGBM的平均绝对误差为3.37%,决定系数为0.92。结论认为,较之于其他常规模型,SAO-LightGBM具有更为可靠的预测能力,能够高效完成目标层位孔隙度的预测工作,对NC地区的储层研究和后期勘探开发具有指导作用。

Porosity is a decisive parameter in the evaluation on reservoirs'physical properties.Only a little data has been acquired from drilling and coring in NC area,central Sichuan Basin.In particular,it was a direct challenge to receive this parameter.And traditional methods for porosity prediction based on conventional logging data give rise to a large error or poor accuracy.So,taken the reservoirs of Xujiahe 4 Member as examples,an improved machine-learning algorithm,namely SAO-LightGBM,was put forward to clarify the physical properties of tight sandstone reservoirs in this area.Then,the algorithm was utilized to analyze the latent relationship of porosity to logging parameters,indicating its strong correlation with acoustic time difference,density,neutron porosity,resistivity and natural gamma.Finally,a prediction model was built on account of these parameters.Results show that(i)owing to the exclusive dual swarm mechanism together with both efficient enquiry and utilization strategy,SAO algorithm can quickly search an optimal hyper-parameter combination of LightGBM,so as to scale up this model's prediction ability;and(ii)both mean absolute error and determination coefficient of SAO-LightGBM are 3.37%and 0.92,respectively.In conclusion,boasting more reliable ability in comparison with other regular models,SAO-LightGBM can predict the porosity very well,which plays a guiding role in reservoir research and later exploitation in NC area.