Classification And Optimization Of Water Saturation Based On Convolutional Neural Network

This paper takes the water saturation involved in the process of petroleum exploitation as the research object.Water saturation can reflect the current stage of oil acquisition and development of the current reservoir.It is of great significance to determine water saturation for the formulation of injection-production strategy and summary analysis.The traditional classification method of water saturation is artificial classification,which is inefficient and has the problems of insignificant change of water saturation and difficult statistical analysis.Moreover,this study is the first attempt of convolutional neural network in the field of water saturation classification.Based on the data of a reservoir in Shengli Oilfield,this study aims to establish a classification model based on convolutional neural network in deep learning to achieve more efficient classification of water saturation and build an online classification system.The main research contents of this paper are as follows:1)Aiming at the problem of insignificant water saturation change and difficulty in statistical analysis,this paper proposes a CNN classification model integrating improved Elman neural network.Based on the water saturation data set,the structure change of convolutional neural network was used as the control group to verify the effectiveness of the proposed model.By evaluating the model through the confusion matrix,the accuracy of the proposed model reached 91.27%in the test set.After determining the availability of the model,the influence of random discarding of neurons and batch normalization on the performance of the network model was further investigated.It is found that 20%random discarding of training neurons and batch normalization can effectively improve the performance of the model.2)Considering the advantages and disadvantages of existing activation functions,a new activation function,Sunken Linear Unit(SnLU),was proposed to improve the performance of convolutional neural networks.The new function avoids output saturation,thus reducing the negative impact on training.Compared with the existing activation functions,the new function has smooth and non-monotone distribution characteristics,and the soft saturation in the negative domain can improve the ability of the model to process information.By studying the water saturation data set,MINIST data set and CIFAR-10 data set,it is verified that the activation function proposed in this paper can improve the accuracy of the model.3)An intelligent classification system for water saturation was developed in this study.In this study,Java development language is used to complete the intelligent classification system of water saturation in B/S architecture based on SpringBoot framework.The system realizes the visual online management of water saturation samples,and the classification of concentration grades one by one and in batches is of great significance to the analysis of water saturation in the production line.