基于局部加权的决策树算法在孔隙度预测中的应用

The Application of Locally Weighted Decision Tree Algorithm to Porosity Prediction

近年来,数据挖掘中的回归算法在地学领域中的运用越来越受到人们的关注,尤其是在储层孔隙度预测方面。储层孔隙度对描述储层特性和储量估计有着极其重要的意义。对于储层的孔隙度,各种测井信息可以看作是其影响因素,但是又很难直接找到孔隙度与测井信息间的映射关系。而从数据挖掘的角度来看,储层孔隙度的预测问题是典型的回归问题,根据测井信息建立适当的模型,通过参入已知测井信息和孔隙度资料的训练,就可以建立储层孔隙度与测井信息之间的非线性关系,进而对储层孔隙度进行预测。本文在采用补偿中子(CNL)、自然伽马(GR)、自然电位(SP)、声波时差(AC)、感应电导率(COND)等测井数据来建立孔隙度预测模型的基础上,使用基于局部加权的决策树算法来对储层孔隙度进行预测,预测的实验结果表明:采用的算法能够利用测井资料准确地对储层孔隙度进行预测;此外,与常规决策树算法相比,应用局部加权的决策树算法进行孔隙度预测时,预测的结果具有更高的精度。

Recently, the application of regression algorithm of data mining in geosciences has drawn increasingly attention, especially in the area of reservoir porosity prediction. Po- rosity is a particularly important parameter in describing features and estimation of reser- voirs. Log data can he seen as influence factors for porosity, but it is difficult to find a di rect relationship between porosity and log data. From the standpoint of data mining, the problem of reservoir porosity prediction is a typical regression problem. Use the training data of log data and porosity to build an appropriate model. Then it can get the non--linear relationship between porosity and log information, and thus to predict porosity. Selecting the locally weighted decision tree algorithm to build the model, the log data of compensated neutron, natural gamma, spontaneous potential, acoustic and conductivity can be seen as the attributes to build the model for porosity prediction. Experimental results show that：compared to the decision tree which is not improved, the decision tree applied locally weigh- ted improves the accuracy of porosity prediction.