基于贝叶斯网络的油气勘探风险预测方法——以准噶尔盆地腹部侏罗系三工河组为例

Prediction method of petroleum exploration risks based on Bayesian network:a case study of the Jurassic Sangonghe Formation in the hinterland of Junggar Basin

有效预测油气勘探风险对优化油气勘探部署、提高钻探成功率及经济效益具有至关重要的意义。在分析油气勘探风险预测方法发展状况的基础上,提出基于贝叶斯网络的油气勘探风险预测方法,论述了石油地质问题向概率预测模型的转化过程,构建了AODE模型的算法和预测步骤。以准噶尔盆地腹部侏罗系三工河组为例,开展了油气成藏地质条件的定量评价,确定了供烃、储层、圈闭、盖层与保存4项主控地质参数,建立了由203口探井参数组成的数据集。五折交叉验证结果揭示:(1)4种贝叶斯网络(朴素贝叶斯分类器、树增强贝叶斯分类器、平均一阶依赖估计器和k阶依赖贝叶斯分类器)的训练集判别正确率均大于85%,说明训练集参数分类有效;(2)测试集判别正确率均大于82%,说明预测成功率高。平均一阶依赖估计器模型的预测效果最好,准确率达到85.22%,因此采用该模型预测研究区油气勘探风险。结果揭示:平均一阶依赖估计器模型预测结果不仅在储量区内与勘探结果吻合度较高,而且在储量区外预测了3类油气资源分布有利区。

Effective prediction of petroleum exploration risks is of great significance for optimizing exploration deployment and improving drilling success rate and economic benefits.Based on the analysis of the progress of risk prediction technology,a new prediction method of petroleum exploration risks is proposed based on Bayesian network.The transformation process from petroleum geological problems to probability prediction model is discussed,and the algorithm and prediction steps of average one-dependence estimators(AODE)training model are constructed.The quantitative evaluation of geological conditions of hydrocarbon accumulation in the Jurassic Sangonghe Formation in the hinterland of Junggar Basin is conducted,four main geological parameters of hydrocarbon supply,reservoir,trap,cap rock and preservation are determined,and a data set composed of parameters from 203 exploration wells is established.The results of five-fold cross validation show that:(1)The discrimination accuracy of the training set of four Bayesian networks(Naive Bayesian classifier,tree-augmented Bayesian classifier,AODE and k-dependence Bayesian classifier)is greater than 85%,indicating that the classification of parameters in the training set is effective;(2)The discrimination accuracy of the test set is greater than 82%,indicating a high success rate of prediction.AODE has the best prediction results,with an accuracy of 85.22%,therefore,it is used to predict the risk of petroleum exploration in the study area.The prediction results of AODE model not only agreed well with the exploration results within the reserve area,but also supported to predict favorable areas of three types of oil and gas resources outside the reserve area.