基于随机森林算法的AVO类型判别

AVO technology can be used for the identification of gas-bearing reservoirs and is of great significance for oil and gas exploration. The artificial identification of reservoir AVO types has relatively large human interference factors, with lower identification accuracy and longer time consumption. Therefore, this paper introduces the random forest algorithm, which uses techniques such as Bootstrap repeated sampling and branch and leaf node splitting to generate a large number of decision tree classifiers, and realizes the discrimination of reservoir AVO types by statistically analyzing the classification results of all decision trees. Firstly, a velocity-density model is established based on the logging data in the work area; secondly, the AVO curve is calculated using the Shuey approximation formula and the fitting polynomial corresponding to this curve is obtained; thirdly, the morphological feature parameters are extracted according to the fitting polynomial and used as the input parameters of the training data set of the random forest algorithm, and the artificial AVO type identification results are used as the output parameters to train and obtain the decision tree classifier; finally, the AVO curve feature parameters of the actual pre-stack seismic data are used as the input parameters, and the reservoir AVO types in the work area are obtained through the random forest decision tree classification and discrimination. It can be seen from the comparison results with the approximate support vector machine algorithm that the two algorithms have similar discrimination results for reservoir AVO types and both have relatively high accuracy. However, in comparison, the random forest algorithm requires fewer characteristic attributes, has stronger generalization ability, and has better universality.

AVO技术可用于含气储层的识别,对油气勘探具有重要意义。人工识别储层AVO类型人为干扰因素较大,识别精度较低且耗时较长。由此,本文引入随机森林算法,利用Bootstrap重复抽样及枝叶节点分裂等技术生成大量决策树分类器,通过统计所有决策树的分类结果实现对储层AVO类型的判别。首先,基于工区内测井数据建立速度密度模型;其次,利用Shuey近似公式计算AVO曲线并获得该曲线对应的拟合多项式;第三,根据拟合多项式提取形态特征参数作为随机森林算法的训练数据集输入参数,将人工AVO类型识别结果作为输出参数,训练并得到决策树分类器;最后,以实际叠前地震数据的AVO曲线特征参数为输入参数,通过随机森林决策树分类判别得到工区内储层AVO类型。通过与近似支持向量机算法的对比结果可以看出,两种算法对储层AVO类型判别结果相近,都具有较高的准确率,但相比之下随机森林算法所需特征属性较少,泛化性较强,具有更好的普适性。