基于BP神经网络的测井资料预测岩石热导率

Prediction of Thermal Conductivity of Rocks Through Geophysical Well Logs Based on BP Neural Network

为了获取无岩心深度段的岩石热导率,建立基于BP神经网络的热导率预测模型。根据声波、密度、中子、电阻率、自然伽马等5种测井响应预测岩石热导率,其模型计算所需时间较短,不需要岩性组分资料,比只考虑1种或其中几种物理参数影响的经验公式适用范围更广。对检验样本以及位于南海的1144A井、1146A井、1148A井等3口大洋科学钻探ODP(Ocean Drilling Program)钻孔的热导率预测结果表明,模型预测的热导率误差低于实验室岩石热导率测试的最大允许误差。该热导率预测模型为获取没有岩心的上述5种测井响应的深度段的岩石热导率提供了一种新途径。

In order to obtain thermal conductivity of rocks at the depth where no core is available, we build ua prediction model for thermal conductivity based on BP neural networks with sonic, density, neutron porosity, resistivity, gamma ray as input. The prediction model needs short es- timating time without any more lithological composition data, therefore it has more and wider ap- plications than the empirical formula only influenced by one or several physical parameters. The test results from the test samples and 1144A, 1146A, 1148A well logs show that the error given by our model is less than the maximum permissible error of thermal conductivity measurement under laboratory conditions. This model provides a new way for obtaining thermal conductivity of rocks at the depth where has no core but has the related geophysical well logs.