摘要
针对目前油气柱高度预测技术局限于传统的地质方法且预测效果不太理想的现状,展开一种基于改进残差神经网络的油气柱高度预测的研究.该模型从断层解释和油藏解剖提取的圈闭结构化特征数据中提取特征信息,以估计油气柱高度.模型将原始残差块中的串行连接网络变成多个并行连接的网络,可以在多个尺度上同时进行卷积再聚合,能提取到不同尺度的特征,使其变成一个稀疏性、高计算性能的网络结构;同时保留了网络中跳跃连接的结构,缓解了在深度神经网络中增加深度带来了梯度消失和网络退化的问题,通过直接将输入信息绕道传到输出,保护信息的完整性;并在模型的首层和尾层增加注意力模块,来捕获集中于某个局部信息,使模型其能更快地收敛.此外对机器学习中常用的RF和BP神经网络以及深度学习中CNN、GoogleNet、ResNet和ResNet+Atten在圈闭数据上的应用进行了比较和分析.实验结果表明,改进的ResNet对油气柱高度预测有更加准确的结果 .
Aimed at the current situation that compared to the traditional geological methods,the oil and gas col⁃umn height prediction technology is limited,and the prediction effect is not ideal,the study on the oil and gas column height prediction based on improved residual neural network was carried out.The model extracts the fea⁃ture information from the structural feature data of traps extracted from the fault interpretation and reservoir dis⁃section to estimate the oil and gas column height.The model turns the serial connection network in the original residual block into multiple parallel connection networks,which can be simultaneously convolved and re aggre⁃gated on multiple scales,and can extract features of different scales,and which becomes a sparse,highperformance network structure;at the same time,the skip connection structure in the network is retained,which alleviates the problem of gradient disappearance and network degradation caused by increasing the depth in the deep neural network.The integrity of information is protected by directly bypassing the input information to the output;additionally,the attention modules are added to the first and last layers of the model to capture some local information,so that the model can converge faster.The commonly used RF and BP neural networks in machine learning and the applications of CNN,GoogleNet,ResNet,and ResNet+Atten in deep learning in trap data are compared and analyzed.The experimental results show that the improved ResNet has more accu⁃rate results in predicting the height of oil and gas column.
作者
杜睿山
程永昌
孟令东
Du Ruishan;Cheng Yongchang;Meng Lingdong(School of Computer and Information Technology,Northeast Petroleum University,Daqing 163318,China;Key Laboratory of Oil and Gas Reservoir and Underground Gas Storage Integrity Evaluation,Daqing 163318,China)
出处
《海南大学学报（自然科学版）》
CAS
2024年第1期19-29,共11页
Natural Science Journal of Hainan University
基金
国家自然科学基金(41702156)
东北石油大学引导性创新基金(2020YDL-04)。
