基于交叉验证SVM储层预测方法研究及应用

东营凹陷深部储层埋深大，构造及相带变化复杂，钻遇目标层的井少，储层预测有很大的困难。本文以东营凹陷东部孔一段为例，将适合较小样本预测的支持向量机方法（Support Vector Machine, SVM）应用到储层预测中。为了提高预测精度，惩罚因子选取和核函数参数训练过程中引入了交叉验证。输入样本为井点处的地震属性和储层厚度，属性通过井震关系优先，选取的是带宽、能量半时、最大振幅、均方根振幅、过零点个数和弧长等6种属性。预测结果表明，本文方法较常规的多元线性回归、不加交叉验证的SVM方法，有更高的预测精度，在深层勘探中有推广价值。

Research and Application of SVM Reservoir Prediction Method Based on

The deep reservoirs in the Dongying sag have large buried depths, complex changes in structure and facies, and few wells in the target layer. Reservoir prediction is very difficult. This paper takes the eastern section of the Dongying sag as an example, and applies the Support Vector Machine (SVM) method suitable for small sample prediction to reservoir prediction. In order to improve the prediction accuracy, cross-validation is introduced in the penalty factor selection and kernel function parameter training. The input sample is the seismic attribute and reservoir thickness at the well point. The attributes are optimized by the well-seismic relationship. The six properties of bandwidth, energy half-time, maximum amplitude, root mean square amplitude, zero-crossing point and arc length are selected. The prediction results show that the proposed method has higher prediction accuracy than conventional multivariate linear regression and SVM without cross-validation, and has popularization value in deep exploration.