利用GRU神经网络预测横波速度

储层参数与横波速度之间存在一定的相关关系，但是这种复杂关系很难得到解析解。为此，构建了GRU（gated recurrent unit）神经网络方法，主要包括神经网络构建、数据预处理、样本训练和数据预测四个部分，通过训练神经网络逼近横波速度与储层参数之间的关系，利用纵波速度、密度和自然伽马等储层参数直接预测横波速度。采用D区的30口井的测井数据训练和测试神经网络，结果表明：①纵波速度、密度和电阻率对数与横波速度呈较好的正相关关系，自然伽马值、孔隙度与横波速度呈负相关关系。②对于多数井训练、少数井验证，训练数据预测的横波速度与真实值的相对误差和相关系数分别约为3.00%和0.9837，测试数据预测的横波速度与真实值的相对误差和相关系数分别约3.19%和0.9805；对于少数井训练、多数井验证，训练数据预测的横波速度与真实值的相对误差和相关系数分别约为2.49%和0.9867，测试数据预测的横波速度与真实值的相对误差和相关系数分别约3.92%和0.9686。因此所提方法具有较高预测精度和较强泛化能力。

Prediction of S-wave velocity based on GRU neural network

There is a relationship between reservoir parameters and shear wave velocity.But it is too complex to get analytic solutions.This paper proposes a GRU (gated recurrent unit) neural network which includes building a neural network,data preprocessing,samples training and data predication.After approximating the relationship between S-wave velocity and reservoir parameters by training a neural network,S-wave velocity can be predicted directly from P-wave velocity,density,gamma ray,porosity and logarithm of resistivity.The neural network was trained and tested by logging data from 30 wells in Block D.The results show that: ①The P-wave velocity,density and logarithm of resistivity are positively related to the S-wave velocity,while the gamma ray and porosity are negatively related to the S-wave velocity; ②In the case of being trained by more wells and tested by less wells,the relative error between the S-wave velocity and the real one is about 3.00%,and the correlation coefficient is 0.9837 for training data,while they are 3.19% and 0.9805 for tested data.In the case of being trained by less wells and tested by more wellsg,the relative error is 2.49% and the correlation coefficient is 0.9867 for training data,while they are 3.92% and 0.9686 for testing data.The new method has a high prediction accuracy and generalization ability.