储层参数预测和储层流体性质识别对储层测井精细描述具有十分重要意义。测井曲线与储层参数和流体类型之间有很强的非线性映射关系，用多元统计方法和传统BP网络方法反演，常会出现过拟合现象和模型推广能力差等问题。人工智能技术的迅速发展，为智能化测井解释提供了新的技术，在一定程度上解决了传统方法存在的问题。

       测井数据具有类时间序列特征，双向长短期记忆(BiLSTM)神经网络的双向记忆特性能同时利用测井数据序列的前后向信息进行测井储层解释。论文研究了改进的BiLSTM模型用以解决过拟合现象，并将改进后的BiLSTM网络模型分别对测井储层孔隙度和流体性质进行应用研究，具体研究工作如下：

       首先，研究钻井取芯分析所得储层参数值缺失问题的处理方法。论文采用线性插值法对取芯数据集进行插值，以确保每0.125m间隔深度有一取芯分析数据，使之与测井曲线数据相对应，这在一定程度上预防过拟合现象的发生。

       其次，研究智能化模型的过拟合现象。在训练BiLSTM网络模型获得最优超参数的基础上，本文将Dropout机制和最大范式约束(Max_Norm)结合，使模型从网络结构和权重参数限制两方面抑制过拟合现象的发生；进一步，本文采用小批量+自适应动量估计(Adam)算法对BiLSTM模型进行优化，控制模型的收敛程度和对学习率进行自动调优。

        最后，将改进的BiLSTM、误差反向传播(BP)神经网络、支持向量机(SVM)和循环神经网络(RNN)分别应用到储层孔隙度和储层流体性质问题进行对比研究。结果显示，改进的BiLSTM模型与BiLSTM模型相比，Dropout机制和最大范式约束融合对过拟合现象的发生起到一定的抑制作用；同时，改进的BiLSTM模型与其他三类机器学习方法相比，改进的BiLSTM模型与时间序列类问题具有较好的一致性。

      The prediction of reservoir parameters and the identification of reservoir fluid properties are of great significance for the accurate description of reservoir logging. Logging curves has a strong nonlinear mapping relationship between reservoir parameters and fluid types, It often leads to problems such as overfitting and poor model expansion ability by using multivariate statistical methods and traditional BP network method for inversion. The rapid development of artificial intelligence technology has provided new technologies for intelligent logging interpretation, and solved the problems existing in traditional methods to certain extent.

      The logging data has the characteristics of time series, and the bidirectional memory characteristic of Bi-directional Long Short-Term Memory (BiLSTM) neural network can used the forward and backward information of the logging data sequence for logging reservoir interpretation simultaneously. In this paper, the improved BiLSTM model was studied to solve the overfitting phenomenon, and the improved BiLSTM network model was applied to study the logging reservoir porosity and fluid properties respectively. The specific research work is as follows:

       Firstly, the treatment method of the scarcity of reservoir parameter values was studied, which was obtained by drilling and coring analysis. In this paper, the linear interpolation method was used to interpolate the coring data set to ensure that there is a coring analysis data corresponding to the logging curve data every 0.125m interval depth, which prevents the occurrence of overfitting to certain extent.

      Second, The overfitting phenomenon of intelligent models was investigated. this paper combined the Dropout mechanism with the Max_Norm constraints (Max_Norm) to make the model can suppress the occurrence of overfitting from two aspects of network structure and weight parameter constraints, which was obtained on the basis of training the BiLSTM network model to get the optimal hyperparameters; further, this paper used the mini-batch+Adaptive Momentum Estimation (Adam) algorithm to optimize the BiLSTM model to controls the degree of convergence of the model and tunes the learning rate automatically.

       Finally, the improved BiLSTM, Error Back Propagation (BP) Neural Network, Support Vector Machine (SVM) and Recurrent Neural Network (RNN) were respectively applied to the problems of reservoir porosity and reservoir fluid properties for comparative study. The results show that compared with the BiLSTM model and the improved BiLSTM model, the Dropout mechanism was combined with the Max_Norm constraints play a certain role in inhibiting the occurrence of overfitting; at the same time, the improved BiLSTM model was compared with other three types of machine learning methods, the improved BiLSTM model has better consistency with time series problems.

[1] 魏佳明. 机器学习在储层参数预测中的应用研究[D]. 西安: 西安石油大学, 2019.

[2] Prakash C, Kumar R, Mittal N. Recent developments in human gait research: parameters, approaches, applications, machine learning techniques, datasets and challenges[J]. Artificial Intelligence Review, 2018, 49(1): 1-40.

[3] Cutler M, Walsh T J, How J P. Real-World Reinforcement Learning via Multifidelity Simulators[J]. IEEE Transactions on Robotics, 2017, 31(3): 655-671.

[4] 石双虎, 魏铁, 张翊孟, 等. 页岩气地震勘探资料采集方案[J]. 非常规油气, 2016, 3(01): 1-6.

[5] 邓述全, 洪月英, 王永涛, 等. 电测井和电阻率法在表层调查中的应用[J]. 勘探地球物理进展, 2004(01): 66-71.

[6] 陈国华, 杜育红. 地质录井信息采集管理系统[J]. 录井工程, 2006(S1): 44-46+77.

[7] 罗帅. 华北地区SY1井钻井技术难点及对策[J]. 石油地质与工程, 2019, 33(03): 86-89.

[8] 桑凯恒, 张繁昌. 基于模糊粗糙集的机器学习储层参数预测[J]. CT理论与应用研究, 2018, 27(04): 455-464.

[9] Minar, M, R, J.Naher. Recent Advances in Deep Learning: An Overview[J], 2018, DOI: 10.13140/RG2.2.24831.10403.

[10] 白云程, 何川. 人工智能专家系统在测井综合解释中的应用[J]. 测井技术, 1984(04): 16-32.

[11] 夏林. 基于全噪声自动编码器的深度神经网络优化算法[D]. 武汉: 武汉科技大学, 2016.

[12] Schapire R E, Freund Y. A decision-theoretic generalization of on-line learning and an application to boosting[J]. Journal of computer and system sciences, 1997, 55(1): 119-139.

[13] Vapnik V, Cortes C. Support-vector networks[J]. Machine Learning, 1995. 20(3): 273-297.

[14] Breiman L. Random forests[J]. Machine learning, 2001, 45(1): 5-32.

[15] Rumelhart D E, Hinton G E, Willians R J. Learning representations by back-propagating errors[J]. Nature, 1986, 323(6088): 533-536.

[16] Elman L. Finding structure in time[J]. Cognitive Science, 1990, 14(2): 179-211.

[17] Schuster M, Paliwal K K. Bidirectional Recurrent Neural Networks[J]. IEEE Transactions on Signal Processing: A publication of the IEEE Signal Processing Society, 1997, 45(11): 2673-2681.

[18] Bengio Y, Simard P, Frasconi P. Learning long-term dependencies with gradient descent is difficult[J]. IEEE transactions on neural networks, 1994, 5(2): 157-166.

[19] Hochreiter S, Schmidhuber J. Long short-term memory[J]. Neural computation, 1997, 9(8): 1735-1780.

[20] Graves A, Schmidhuber J. Fram ewise phoneme classification with bidirectional LSTM and other neural network architectures[J]. Neural networks, 2005, 18(5-6): 602-610.

[21] Glorot X, Bordes A, Bengio Y. Deep sparse rectifier networks[C]. JMLP Workshop and Conference Proceeding Volume 15: AISTATS 2011. Brookline, MA: Micotome Publishing, 2011: 315-323.

[22] Pascanu R, Mikolov T, Bengio Y. On the difficulty of training recurrent neural networks[C]. International conference on machine learning. 2013: 1310-1318.

[23] Srivastava N. Improving neural networks with dropout[J]. University of Toronto, 2013, 182(566): 7.

[24] MacQueen J. Some methods for classification and analysis of multivariate observations[J]. Berkeley Symposium on Mathematical Statistics and Probability, 1967, 1(14): 281-297.

[25] Bernhard Schö, lkopf, Alexander Smola, Klaus-Robert Mü,ller. Nonlinear Component Analysis as a Kernel Eigenvalue Problem[J]. Neural Computation, 1998, 10(5): 1299-1319.

[26] Aïfa T, Baouche R, Baddari K. Neuro-fuzzy system to predict permeability and porosity from well log data: A case study of Hassi R׳Mel gas field, Algeria[J]. Journal of Petroleum Science and Engineering, 2014, P: 217-229.

[27] 王娜娜, 张国英, 王明君. 改进的BP神经网络在石油测井解释中的应用[J]. 北京石油化工学院学报, 2008(01): 17-20.

[28] Yuan, Z, Huang H, Jiang Y and Tang J. Multi-attribute reservoir parameter estimation based on a machine learning technique[C]. 88th Annual International Meeting, SEG, Expanded Abstracts, 2018, 2266-2270.

[29] 杨柳青, 陈伟, 查蓓. 利用卷积神经网络对储层孔隙度的预测研究与应用[J]. 地球物理学进展, 2019, 34(04): 1548-1555.

[30] 安鹏, 曹丹平, 赵宝银, 等. 基于LSTM循环神经网络的储层物性参数预测方法研究[J]. 地球物理学进展, 2019, 34(05): 1849-1858.

[31] Xun Z F, Yu J F. The application of cluster and discriminant analysis in logging lithology recognition[J]. Journal of Shandong University of Science and Technology, 2008, 27(5): 10-13.

[32] 江凯, 王守东, 胡永静, 等. 基于 Boosting Tree 算法的测井岩性识别模型[J]. 测井技术, 2018.8, 42(4): 395-401.

[33] Yu S, Zhu K J, Diao F Q. A dynamic all parameters adaptive BP neural networks model and its application on oil reservoir prediction[J]. Applied Mathematics and Computation, 2007, 195(1): 66-75.

[34] 刘丹, 潘保芝, 周玉凤, 等. 基于高分辨率陈列感应测井的GA-SVM流体识别方法[J].地球物力学进展, 2017, 32(5): 2051-2056.

[35] Dunham, M. W., A. Malcolm, J.K. Improved well log classification using semi-supervised algorithms[C]. 89th Annual International Meeting, SEG, Expanded Abstracts, 2019, 2398-2402.

[36] Hochreiter S, Schmidhuber J. Long Short-term Memory [J]. Neural Computation, 1997, 9(8): 1735-1780.

[37] 周杰. 基于LSTM神经网络的股票价格趋势预测的研究[D]. 云南: 云南大学, 2019.

[38] GAL Y, GHAHRAMANI Z. Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning[J]. 2014, 15(1): 1929-1958.

[39] 周建科, 印兴耀, 曹丹平. 井震资料尺度匹配过程中声测井数据的精细分层方法研究[J]. 物探化探计算技术, 2015, 37(02): 242-248.

[40] Nitish Srivastava, Geoffrey E. Hinton, Alex Krizhevsky, Ilya Sutskever, Ruslan Salakhutdinov. Dropout: a simple way to prevent neural networks from overfitting[J]. Journal of Machine Learning Research, 2014, 15(1): 1929-1958.

[41] DAHL G E, SAINATH T N, Hinton G E. Improving deep neural networks for LVCSR using rectified linear units and dropout[J]. 2013, 26: 8609-8613.

[42] 刘昌伟. 基于深度的双向LSTM-RNN模型的股票预测研究[D]. 黑龙江: 黑龙江大学, 2020.

[43] Semeniuta S, Severyn A, Barth E. Recurrent Dropout without Memory Loss[J]. CoRR, 2016, abs/1603. 05118:051189.

[44] SRIVASTAVAN, HINTON G, KRIZHEVSKY A, et al. Dropout a simple way to prevent neural networks from overfitting[J]. The journal of machine learning research, 2014, 15(1): 1929-1958.

[45] Saito H, Kato M. 2018. Machine Learning Technique to Find Quantum Many-Body Ground States of Bosons on a Lattice [J]. Journal of the Physical Society of Japan, 87(1): 014001.

[46] Arcos-García A, Álvarez-García J A, Soria-Morillo L M. Deep neural network for traffic sign recognition systems: An analysis of spatial transformers and stochastic optimisation methods [J]. Neural Networks, 2018, 99: 158-165.

[47] Weiling Chen, Chai Kiat Yeo, Chiew Tong Lau, Bu Sung Lee. Leveraging social media news to predict stock index movement using RNN-boost [J]. Data & Knowledge Engineering, 2018, 118: 14-24.

[48] Liu M L, Grana D. Accelerating geostatistical seismic inversion using TensorFlow: A heterogeneous distributed deep learning framework [J]. Computers and Geosciences, 2019, 124: 37-45.

[49] 李军, 武清钊, 路菁, 等. 页岩气储层总孔隙度与有效孔隙度测量及测井评价——以四川盆地龙马溪组页岩气储层为例[J]. 石油与天然气地质, 2017, 38(03): 602-609.

[50] 赵存秀. 基于混淆矩阵的分类器性能评价指标比较[J]. 电子技术与软件工程, 2020(13): 146-147.

[51] Deng X, Liu Q, Deng Y, et al. An improved method to construct basic probability assignment based on the confusion matrix for classification problem[J]. Information Sciences, 2016, 340: 250-261.

[52] Li Y, Chen W, Liu D, et al. IFFLC: An Integrated Framework of Feature Learning and Classification for Multiple Diagnosis Codes Assignment[J]. IEEE Access, 2019, 7: 36810-36818.