基于线性回归与神经网络的储层参数预测复合方法
发布时间：2018-12-09 13:10
【摘要】：为提高储层参数的预测精度,提出一种利用测井资料,结合多元线性回归和神经网络预测储层参数的新的复合方法,具体分两步:(1)通过多元线性回归分析建立回归值y'的计算模型,将y'作为储层参数的初步预测值;(2)通过RBF神经网络建立y'的残差Δd的预测模型,将预测结果Δd作为y'的非线性误差补偿,最终建立储层参数解释模型,y=y'+Δd。基于该方法,结合测井资料和岩心数据,建立了鄂尔多斯盆地某致密砂岩气田M3井区S_2~2、T_2~2段孔隙度和含水饱和度的测井解释模型,结果显示,新方法建立的模型预测值与S_2~2、T_2~2段实际岩心孔隙度、含水饱和度值的平均相对误差均小于17%,明显优于单独根据多元线性回归分析或RBF神经网络建立的解释模型,预测精度更高。
[Abstract]:In order to improve the prediction accuracy of reservoir parameters, a new composite method is proposed to predict reservoir parameters by using logging data, combining multiple linear regression and neural network. It is divided into two steps: (1) the regression value 'calculation model is established by multiple linear regression analysis, and y' is regarded as the preliminary prediction value of reservoir parameters; (2) the prediction model of Y 'residual 螖 d is established by RBF neural network, and the prediction result 螖 d is regarded as the nonlinear error compensation of y'. Finally, the reservoir parameter interpretation model, YY' 螖 d, is established. Based on this method, a logging interpretation model for the porosity and water saturation of SSP _ 2O _ 2T _ 2O _ 2 section in M _ 3 well area of a tight sandstone gas field in Ordos Basin is established by combining well logging data and core data. The results show that, The predicted values of the new model and the actual core porosity and the average relative error of the water saturation values of the two sections are all less than 17, which is obviously superior to the interpretation model established solely on the basis of multiple linear regression analysis or RBF neural network. The prediction accuracy is higher.
【作者单位】： 中国地质大学(北京)能源学院;
【基金】：国家自然科学基金(41272132) 中国地质调查局地质调查项目(2-1-2010-18-A,12120115002901-04)资助
【分类号】：P618.13
[Abstract]:In order to improve the prediction accuracy of reservoir parameters, a new composite method is proposed to predict reservoir parameters by using logging data, combining multiple linear regression and neural network. It is divided into two steps: (1) the regression value 'calculation model is established by multiple linear regression analysis, and y' is regarded as the preliminary prediction value of reservoir parameters; (2) the prediction model of Y 'residual 螖 d is established by RBF neural network, and the prediction result 螖 d is regarded as the nonlinear error compensation of y'. Finally, the reservoir parameter interpretation model, YY' 螖 d, is established. Based on this method, a logging interpretation model for the porosity and water saturation of SSP _ 2O _ 2T _ 2O _ 2 section in M _ 3 well area of a tight sandstone gas field in Ordos Basin is established by combining well logging data and core data. The results show that, The predicted values of the new model and the actual core porosity and the average relative error of the water saturation values of the two sections are all less than 17, which is obviously superior to the interpretation model established solely on the basis of multiple linear regression analysis or RBF neural network. The prediction accuracy is higher.
【作者单位】： 中国地质大学(北京)能源学院;
【基金】：国家自然科学基金(41272132) 中国地质调查局地质调查项目(2-1-2010-18-A,12120115002901-04)资助
【分类号】：P618.13
【相似文献】
相关期刊论文 前10条
1 杨斌,肖慈s，
本文编号：2369410
本文链接：https://www.wllwen.com/kejilunwen/diqiudizhi/2369410.html
