论文详情
微测井是地震勘探中常用的一种近地表纵波速度调查方法, 在场地条件和施工成本受限的情况下, 该方法得到的速度解释剖面常存在横向分辨率不足的问题。利用静力触探法布设方便、成本低廉的优势, 提出一种利用人工神经网络模型关联地层阻力和地层波速的方法, 以期通过少量实测微测井实现大范围纵波速度结构的有效预测。该方法的实施流程如下: ①两两配对静力触探和微测井数据以生成控制点位, 以岩性变化为网络分裂条件, 输入层神经元接收锥尖阻力、侧摩阻力和深度数据, 输出层神经元接收纵波速度, 在中间设置多个全连接隐藏层; ②通过前馈训练机制更新隐藏层神经元参数; ③将非控制点位的静力触探数据输入到训练好的神经网络模型以获取全区近地表纵波速度结构剖面。在苏北某场地进行方法测试和数据分析, 结果证实岩性分层的精细度和训练样本量是决定模型表现的两个关键因素。人工神经网络法预测浅层纵波速度的准确率超过90%, 在可靠性、分辨率以及鲁棒性方面都超越了现有的经验公式法, 可以辅助判断地下虚反射界面和低降速带分布范围, 是提高地震勘探浅层速度调查精度和效率的有益探索。
Micro-logging is widely employed in near-surface velocity investigations in seismic exploration, but its lateral resolution is low due to limited site conditions and operating cost.In this paper, we use an artificial neural network (ANN) to link cone penetration test (CPT) resistance with the P-wave velocity of near-surface layers to predict large-scale velocity distribution based on a small number of paired logging-CPT data.This method includes the following steps: ①using lithology as the separation condition, depth and cone resistances as inputs, and velocity as the target output; ②updating hidden layer neurons through a feedforward mechanism; ③obtaining near-surface velocity profiles by inputting CPT data into the trained ANN model.A case study in northern Jiangsu proves that the precision of lithologic division and the size of the training sample set determine our model performance.The ANN method is superior to empirical formula methods in reliability, resolution, and robustness, and the accuracy of shallow P-velocity prediction is over 90%.Using this method, it is easier to locate the ghosting interface and weathered layer close to the surface and perform near-surface velocity investigation more accurately and efficiently.