Research And Application Of Recognition Algorithm Of Interlayer In Reservoir Based On Convolution Neural Network

The existing interlayer identification methods in oil fields have strong subjectivity,poor adaptability and low data utilization.Therefore,based on the existing geological data and production data of the oilfield,this paper proposes a recognition algorithm of interlayer in reservoir based on convolution neural network.The main work is as follows:(1)The image of the interlayer logging curve is special,and its characteristics are mostly concentrated in the change of the curve.Therefore,the parameter selection of the convolutional neural network is very important in the establishment of the interlayer recognition model.Firstly,the network based on AlexNet is built,and then the appropriate network structure and parameters are determined by comparing the effects of different convolution kernel size,strides,number of convolution kernels and batch normalization layer on the classification accuracy of sandwich recognition model.(2)Aiming at the problems of gradient disappearance and excessive network parameters in deep network structure,residual module is introduced.And because logging curve images pay much attention to curve features,adding Squeeze-and-Excitation helps CNN to extract interlayer features more accurately.The experimental results show that the proposed Jc Net training time is640.30 s,the model size is 11.2 M and the accuracy is 95.15 %,which are better than other contrast group experiments.(3)Artificial fish swarm algorithm,particle swarm algorithm and genetic algorithm are introduced to optimize the network structure and some parameters in network training in order to solve the problem of waste of resources in the establishment of CNN sandwich recognition model.After many times of training,the accuracy of the optimized CNN to establish the interlayer identification model was improved by 0.09 % – 0.19 %,the model parameters were reduced by6.83 % – 18.85 %,and the training time was reduced by 31.1 %.The comprehensive experimental results show that compared with other methods,the proposed reservoir interlayer recognition algorithm has a certain reduction in time loss and resource utilization in the training process,and has a great improvement in the accuracy of interlayer recognition and classification.This algorithm has important theoretical and practical significance for further improving oil recovery and promoting the deep application of deep learning method in petroleum industry.