基于SVR和地震属性的构造煤厚度定量预测

Quantitative prediction of tectonic coal seam thickness using support vector regression and seismic attributes

为了提高谱分解和甜面属性组合预测构造煤厚度的精度和可靠性,利用回归型支持向量机(SVR)的非线性处理能力,将SVR和地震属性相结合,研究采区构造煤厚度定量预测方法。在SVR预测模型建立时,以正演模型数据为基础,通过训练和测试获得SVR预测模型的主要参数;结合井旁道数据,建立了采区构造煤厚度SVR预测模型。通过输入实际谱分解属性和甜面属性,定量预测了构造煤厚度。相对于传统地震属性预测来说,本次所预测的构造煤厚度精度较高、误差较小。当核函数类型为径向基核函数、输入为谱分解属性和甜面属性时,预测模型的预测效果最好。由于模型建立时未考虑地震资料信噪比的影响,预测模型不能克服其造成的不确定性。

In order to improve the precision and reliability of thickness prediction,the authors combined the Support Vector Regression( SVR),which is good at dealing with nonlinear issues,and seismic attributes together to build a model for the prediction of tectonic coal seam thickness. During model building,the authors,first of all,optimized the key parameters of SVR model through training and testing forward model's data,then,combined those optimized parameters with in situ near-well traces to build a SVR predicting model. Through inputting the real attributes of spectral decomposition and sweetness into this model,the authors achieved a prediction of tectonic coal seam thickness in the study area. By comparison with true thickness at wells,the thickness prediction of the model has a higher precision and lower absolute error than that with the prediction using traditional seismic attributes. Setting RBF( Radial Basis Function) as kernel and spectral decomposition attributes and sweetness attribute as inputs,the model generates its best results. Since the influence of signal-to-noise ratio of seismic data was not considered during the model development,the model could not overcome its corresponding uncertainty.