PSO-GBDT识别致密砂岩储集层岩性研究——以姬塬油田西部长4+5段为例

Lithology Prediction of Tight Sandstone Reservoirs Using the PSO-GBDT:A Case Study of the Chang 4+5 Members in the Western Jiyuan Oilfield

交会图在致密砂岩储集层应用上难以奏效,其主要原因是多种储集层岩性具有相似测井响应特征而难以在交会图版中被有效分辨。众多机器学习技术可有效分辨属性相似度高的数据,为此识别性能出众的GBDT(gradient boosting decision tree,梯度提升决策树)常被用来解决致密砂岩储集层岩性识别问题。但GBDT使用较多超参数致训练模型难以最优化,本文选用PSO(particle swarm optimization,粒子群算法)来解决优化问题,进而提出PSO-GBDT模型。本文以姬塬油田西部长4+5段致密砂岩储集层为研究对象,通过设计两个实验来验证提出模型的识别能力。实验结果表明,PSO-GBDT岩性识别准确率分别为(90.37%,88.20%)和(93.48%,90.16%),高于其他验证模型。该模型能有效解决致密砂岩储集层岩性识别问题,在岩性识别研究上具有良好的推广应用前景。

The crossplot generally is ineffective on lithology prediction of tight sandstone reservoirs since many lithologies presenting with similar logging responses are rather difficult to be recognized effectively by using crossplots. Some machine learning techniques are capable to effectively discriminate the data with highly similar properties. Thus, the GBDT(gradient boosting decision tree), one of the most excellent techniques, is generally applied to deal with the lithology prediction of tight sandstone reservoirs. However, the application of many hyper-parameters by the GBDT has resulted in the difficult optimization of the training model for processing data. Thus, the PSO(particle swarm optimization) is adopted, and the PSO-GBDT is furtherly proposed to solve the optimization of training model in this paper. The tight sandstone reservoirs of the Chang 4+5 members in the western Jiyuan oilfield have been selected to design two experiments for validating lithology prediction capability of the PSO-GBDT model. Experimental results manifest that the lithology prediction accuracies obtained by using the PSO-GBDT model which are(90.37%, 88.20%) and(93.48%, 90.16%), respectively, are higher than those obtained by using other validation models. This means that the proposed PSO-GBDT model has capability to effectively solve the problem of lithology prediction of tight sandstone reservoirs, and has a very good generalization and applicable prospect in the study field of lithology prediction.