Search

In order to solve the over-fitting problem in Convolutional Neural Networks (CNN), a new method to improve the performance of CNN with noise layer on the basis of the previous studies has been proposed. This method improves the generalization performance of the CNN model by adding corresponding noise to the feature image obtained after convolution operation. The constructed noise layer can be flexibly embedded in a certain position of the CNN structure, and with each iteration of training, the added noise is also constantly changing, which makes the interference to the CNN model more profound, thus the more essential features of the input image are obtained. The experimental results show that the improved CNN model based on the noise layer has better recognition effect on some test images than the CNN model without any improvement; for different CNN models, the position of the noise layer which can improve the recognition accuracy is different; as the number of layers deepens, to improve the generalization performance of the CNN model, the position of the noise layer needs to be moved back. The improved CNN model based on the noise layer proposed in this paper solves the overfitting problem to a certain extent, and it has a certain reference significance for studying how to improve the generalization performance of CNN.

The dynamic simulation software of water flooding is of great significance for observing the exploitation of crude oil and understanding the distribution of microscopic remaining oil in the underground reservoirs. Based on Unified Modeling Language (UML), the design of the dynamic simulation software of water flooding is completed. First, based on the results of demand analysis, the function module diagram of the software is established; then the use case diagram is designed to clarify the goals that the software needs to achieve; finally, the class diagram and the sequence diagram of the software are designed, which provides the basis for the specific coding implementation. The design of the dynamic simulation software of water flooding based on UML enriches and develops the theory, technology and method of exploitation geology, and promotes the further application of advanced computer technology in the petroleum industry.

In order to realize the rapid and accurate recognition of fruit tree diseases in orchard environment, this paper puts forward a deep learning model based on Convolution Neural Network to identify fruit tree diseases. In this paper, the data set is processed by the Sobel operator and image enhancement respectively. Then, the network depth, convolution kernel, feature maps, and fully connected layer in the Convolution Neural Network structure use different parameters and softmax classifier. Differently composition networks are used to train processed dataset. Convolution Neural Network models are used to predict test sets, and the results show that deeper Convolution Neural Networks and mean pooling for tiny features in the dataset are more accurate. It can achieve the disease recognition, which includes cab disease, black rot, rust of apple leaves and bacterial spot disease of peach tree leaves. The model has a good recognition function for disease identification of fruit trees and can help real-time monitoring of orchard diseases.

Hydrate dissociation induced by artificial or environmental factors in submarine sediments may lead to drilling risks, submarine landslides, and the collapse of offshore platforms—all of which are associated with suddenly increased pore pressure. This study establishes a coupled model for the two-phase flow of fluids in porous media in which the pore pressure of water and gas are used as dependent variables and heat transfer and hydrate dissociation equations are combined. The simulation results of the model are consistent with Masuda’s experimental dataset. A quantifiable controlling hydrate dissociation specific reaction surface area (SRSA) model and an absolute permeability model are proposed to conduct numerical simulations of laboratory-scale hydrate cores, with the aim to investigate the mechanism of hydrate dissociation-induced geological hazard initiation and a law of their evolution in sediments. The results show that i) suddenly depressurized hydrate dissociation in porous media can be divided into three phases—pressure induction, transition, and temperature-controlled—and that the first two are much shorter than the third. Moreover, the lower initial permeability results in a shorter holding period for the first stage, and the transition phase becomes indistinguishable; conversely, the larger initial permeability makes the three stages more distinguishable; ii) hardly increasing the core permeability during the intermediate stage of hydrate dissociation makes the pore pressure rebound. However, the steep increase in SRSA does not result in a significant pore gas pressure rebound; conversely, a larger SRSA does facilitate pore pressure dissipation. This work provides theoretical and laboratory-scale model analyses for the safe development of natural gas hydrates bearing in marine sediments.

Natural gas hydrates may be the world's future resources, but extracting from the reservoirs involves accurately understanding the transport mechanism of the fluids caused by hydrate dissociation. In this paper, based on Darcy's unsaturated soil mechanics theory, we establish a coupling model of two-phase flow and heat transfer caused by hydrate dissociation, which considering methane gas compressibility and capillary force naturally. The new model is validated through Masuda's test dataset, and notably, the experimental phenomenon of the small peak in the gas pressure of the far-end boundary (Run 4) is reproduced for the first time. Additionally, an implicit function of pore pressure and time in the initial stage of hydrate dissociation is deduced to interpret the in-situ pore accumulation of the gas generated in dissociation, which leads to a more comprehensive description of the migration mechanism of hydrate reservoir fluids. Furthermore, the initial hydrate saturation of the Run 4 sample in Masuda's test is proved to be closer to 0.501 instead of 0.443 that the author reported, which can help Masuda's test dataset being a benchmark for the hydrate dissociation researching in future. Thus, this work provides further insight into the migration mechanism of the gases caused by hydrate dissociation and services practical support for efficiently and safely extracting methane gas from the hydrate reservoirs.

In compare with the traditional Artificial Neural Network, PSO-BP neutral network has fast convergence and is immune to local minimum. This paper presents an application of PSO-BP neural network for dynamic predicting small layer reservoir parameters of fault block E1f11-1 in well ZHuang 2. By defining input and output neuron number, our method firstly realizes quantization of input neuron. Then we choose proper samples for training neural network in order to build a dynamic prediction model of reservoir parameters. Such model has been successfully tested and the model itself is appropriate for predicting unknown reservoir parameters. Testing result indicates that PSO-BP neural network is superior to the genetic algorithm optimized BP neural network and the pure neural network. Finally, PSO-BP neural network gained certain achievements for dynamically predicting reservoir parameters according as dynamic production information.

This paper proposes an OpenGL approach with the rendering of material and texture for plotting reservoir profile. Such approach realizes the plotting along X or Y axis and arbitrary direction. At last, the application example shows correctness and intuitiveness of plotting reservoir profile by illustrating sectional drawing from subzone sedimentary facies between S1 and S2 in 106 well of one oil field.