In view of the operational efficiency,safety,service and development of the metro system and other application research needs,a multi-level and multi-resolution simulation model system for subway lines is construc-ted,and the application of modeling methods based on building information model(BIM),cloud data and mathemat-ical analysis,geographic information system(GIS)and hybrid simulation are analyzed.This paper discusses the key problems and measures such as multi-resolution modeling,resolution switching rules for space-time coupling and model aggregation algorithm based on dynamic weight that need to be solved in the application of model system.Ac-cording to the industry case study,the effectiveness and feasibility of the methods proposed in this paper in practical applications are verified.Finally,a summary and outlook are made.The research results have important theoretical significance and practical value for metro line planning,operation management,and development.
Breast cancer is the most dangerous killer for women. Accurate early diagnosis is the first step of treatment. This paper proposes a breast cancer detection model CBAMDNet based on breast pathological sections and deep learning. We adopt a pre-trained DenseNet121 embedded with spatial attention mechanism as the backbone model. Three random network models are trained in CBAMDNet to predict, and fused through majority voting to obtain more accurate results. We used a breast pathological section image data set to evaluate the generalization ability of our CBAMDNet, using 4 times cross validation. Simulation experiments show that CBAMDNet can produce higher classification results than the four existing breast cancer classification methods. Therefore, our CBAMDNet is an accurate tool to detect breast cancer and can be used for clinical diagnosis.
It is one of the key directions of the new infrastructure to establish an experimental teaching platform for synergy between science and education, thus promoting the construction of intelligent laboratories. As a new engineering major supporting the development of national strategic emerging industries, the experimental teaching of Internet of Things (IoT) engineering is crucial in the process of talent training. However, the existing experimental teaching system is constrained by many factors including funding, location, faculty strength, and network conditions, which make it difficult to implement large-scale and creative IoT experiments, especially process-integrated experiments. This paper focuses on how to build an experimental teaching process and experimental teaching system that meets the needs of the IoT industry and talent training under the new infrastructure perspective. By analysing the IoT curriculum structure and talent demand, the deficiencies of the existing experimental teaching platform are discussed in combination with the school's talent training positioning and the current IoT industry chain situation. Simultaneously, the technical system structure of IoT is set as the main theme to sort out the knowledge points of IoT experimental teaching, and then a virtual simulation platform of IoT experimental teaching is proposed. The practical application on National Cyberspace proves that the proposed experimental platform can meet the needs of basic experiments, engineering practical training, innovation, scientific research assistance, and industrial applications.
This study aims at applying the improved You Only Look Once V5s model for the assessment of regional poverty using remote sens- ing image target detection. The model was improved from structure, algorithm, and components. Objects in the remote sensing images were used to identify poverty, and the poverty alleviation situation could be predicted according to the existing detection results. The results showed that the values of Precision, Recall, mean Average Precision (mAP)@0.5, and mAP@0.5:0.95 of the model increased 7.3%, 0.7%, 1%, and 7.2%, respectively on the Common Objects in Context data set in the detection stage; the four values increased 3.1%, 2.2%, 1.3%, and 5.7%, respectively on the custom remote sensing image data set in the verification stage. The loss values decreased 2.6% and 37.4%, respectively, on the two data sets. Hence, the ap- plication of the improved model led to the more accurate detection of the targets. Compared with the other papers, the improved model in this paper proved to be better. Artificial poverty alleviation can be replaced by remote sensing image processing because it is inex- pensive, efficient, accurate, objective, does not require data, and has the same evaluation effect. The proposed model can be considered as a promising approach in the assessment of regional poverty.
Due to unreasonable distribution of urban vertical green space,this paper puts forward a method for ex-tracting vertical distribution features of high-density urban greening system.Firstly,an acquisition platform of remote sensing images based on UAV was used to collect greening system images.Secondly,a convolutional neural network architecture was constructed,and then the noise was removed from the initial image through residual learning.The tas-seled hat transformation algorithm was adopted to rotate the image axis,thus enhancing the image,and highlighting the details.According to the similarity and discontinuity of the spectrum,a data-driven algorithm was adopted to deter-mine a homogeneity standard.After that,the image was divided into different areas to ensure that the same target was in the same area.Finally,the vertical distribution features of the greening system were extracted from three aspects:spectrum,shape and texture.Experimental results prove that the vertical distribution feature of the commercial area is similar with that of the residential area,but the green area of the residential area is larger,and the plant species in commercial area are more abundant.In addition,the extracted features can provide reasonable plan.
The “dual carbon” goal has become an important universal value of the international community. Market measures to tackle excess CO2 emissions are effective. Accurate prediction of carbon price lays the foundation for the implementation of civil aviation carbon reduction responsibilities. In this paper, a carbon price prediction model based on the bi -directional long and short term memory network based on multi-frequency feature engineering is proposed to solve the problem of significant fluctuations of carbon price due to external uncertainties, and help the civil aviation industry achieve the goal of “dual carbon”. Considering that carbon price data are complex nonlinear time series, complete ensemble empirical mode decomposition with adaptive noise, CEEMDAN method to obtain data feature characterization. Then, White Gaussian noise (WNG) was superimposed into the simple sequence components of sample entropy to characterize the strong volatility of the original sequence, and then the final feature project was obtained. Finally, it predicts carbon price through Gate Recurrent Unit (GRU). The empirical analysis results show that the prediction model is effective and can provide some method support for airlines to fulfill their carbon reduction responsibilities
Due to the small size of the fire target in the early stage of the forest fire, the model has the problem of missing detection of the small target forest fire. This paper proposes a forest fire recognition method based on YOLOv5 small target detection. In order to obtain flame space information at different scales and obtain an output containing abstract feature information, replace the SPP module in the YOLOv5 backbone network with the WASP module; in order to enhance the expression of flame features, add the Triplet attention module to the C3 module; In order to improve the detection effect of the detection algorithm on small flames, a small target detection layer is added. In this paper, a set of ablation experiments and a set of comparative experiments are designed to verify the effectiveness of the proposed algorithm. The experimental results show that compared with the original YOLOv5 algorithm, the algorithm in this paper has improved mAP@50 by 3.49%, and can accurately detect small target forest fire targets, effectively improving the ability to prevent forest fires.
The stock market is affected by many variables and factors, and the current forecasting models for time series are often difficult to capture the complex laws among multiple factors. Aiming at this problem, a stock price prediction model based on dual attention mechanism and temporal convolutional network is proposed. First, a convolution network more suitable for time series is used as the feature extraction layer. Feature attention is introduced to dynamically mine the potential correlation between the input factor features and closing prices. Second, based on Gated Recurrent Unit, on the other hand, a temporal attention mechanism is introduced to improve the model's ability to learn important time points and obtain importance measures from a temporal perspective. The experimental results show that the proposed model performs better than the traditional prediction model in the error index of stock price prediction and realizes the interpretability of the model in terms of index characteristics and time.
To solve the problem that the torque pulsation is serious due to the large current error in the period when the duty cycle is calculated by the deadbeat method in the duty cycle dual-vector model predictive current control method, a vector switching method is proposed. The control objective of the proposed method is not to equate the predicted torque current with the reference value under the dead-beat idea, but to reduce the torque current control error by changing the switching time of the voltage vector and ensuring the symmetry of the positive and negative current errors. By deducing the duty cycle, the maximum current errors in a single control cycle before and after vector switching is compared and analyzed. The vector switching method is verified on the Simulink simulation platform and compared with the duty cycle double vector method. The results show that the vector switching method can effectively reduce the intra-cycle torque current tracking error and realize the optimal process control in each cycle.
According to the problem of transmitter and receiver node arrangement of multistatic buoy in underwater area surveillance, this paper presents an array optimization method based on GAPSO (Genetic Algorithm - Particle Swarm Optimization). Firstly, the performance evaluation model of multistatic buoy array is established by probability fusion based on the bistatic sonars equation. We take the effective coverage rate of the model as the objective function, then Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) are used to optimize the array. Secondly, PSO and GA are combined to improve the global searching ability and convergence speed. The simulation results show that the coverage ability of the optimized multistatic buoy array is significantly improved compared with the traditional array scheme. Compared with PSO and GA, fused GAPSO has obvious improvement in search ability and convergence speed, which achieves the purpose of optimal defense deployment.
Currently, the development of the power of the Internet of Things brings some challenges to all aspects of power system, one of which is the power supply problem of electric transmission line condition monitoring equipment. Energy-taking current transformer get the favor of the researchers at home and abroad because of its high practical value Aiming at the existing problems of large starting current, easy saturation of iron core, and inability to adapt to wider electric transmission line current of energy-taking current transformer, many of them pay more attention to the research of post-stage circuit, but neglect the design and improvement of iron core structure. To solve the above problems, this paper mainly work as follows: on the basis of the initial permeability, saturation magnetic flux density and iron core loss and cost to determine the iron core material; Ansys software is used to analyze the iron core air gap shape design and simulation, through the iron core under different shape of air gap magnetic leak situation draw rectangular half opening effect optimal; Control variable method is used to analyze the air gap size simulation experiment, through the iron core of the saturated magnetic induction intensity and size of secondary side power output for the quantitative analysis of the air gap size. For the energy-taking current transformer's iron core structure design and parameters of the actual selection provides a good engineering value.
Intelligent and connected vehicles (ICVs) is developed based on Internet of Vehicles (IOV) and intelligent vehicles, has become an effective solution to optimize vehicle traffic and alleviate traffic congestion based on the collaborative strategy of vehicle - road - environment. This paper proposes a lane-changing strategy of vehicles at the exit of roundabout based on the Vehicle Profile(VP) which achieved by the driving characteristics of vehicles in traffic roundabout. Initially, Vehicle Profile(VP) is defined due to the hybrid problem of multi-source heterogeneous data existing in ICVs, then the multi-source heterogeneous data is represented by VP. Moreover, the random forest algorithm is used to dynamically obtain and update the label weight of Vehicle Profile because of the problem of label weight existing in the practical application. In addition, the dynamic weight of the Vehicle Profile is introduced into the design of the vehicle payoff function involved in lane change decision, which solves the problem that the weight of each part of the payoff function is used as a parameter. Finally, the performance of the algorithm is tested and verified by SUMO in roundabout general traffic conditions, congested traffic conditions and sparse traffic conditions. The experimental results show that this algorithm can improve the Efficiency and comfort of vehicle driving in general traffic environment more greatly compared with the situation of traffic congestion and sparse traffic flow. Meanwhile, this algorithm verifies the optimization effect of introducing Vehicle Profile on roundabout traffic strategy.
The key to underwater target recognition is to extract the effective features of underwater target radiation noise. This paper presents an effective method for underwater target recognition and classification by extracting Mel-Frequency Cepstral Coefficients (MFCCs) features of underwater target radiation noise. Compared with traditional spectral analysis methods, MFCC makes full use of the non-linear auditory effect of the human ear with different perception capabilities for sounds of different frequencies. In this paper, the classification experiment of the radiated noise of the three types of measured underwater targets is done, where the MFCC feature vectors of the three types of targets are extracted, and the K-Nearest Neighbor (K-NN) algorithm is used to classify and identify them. Finally, the experimental results show that the method is effective.
The accurate segmentation of retinal blood vessels in fundus is of great practical significance to help doctors diagnose fundus diseases. Aiming to solve the problems of serious segmentation errors and low accuracy in traditional retinal segmentation, a scheme based on the combination of U-Net and Dense-Net was proposed. Firstly, the vascular feature information was enhanced by fusion limited contrast histogram equalization, median filtering, data normalization and multi-scale morphological transformation, and the artifact was corrected by adaptive gamma correction. Secondly, the randomly extracted image blocks are used as training data to increase the data and improve the generalization ability. Thirdly, stochastic gradient descent was used to optimize the Dice loss function to improve the segmentation accuracy. Finally, the Dense-U-net model was used for segmentation. The specificity, accuracy, sensitivity and AUC of this algorithm are 0.9896, 0.9698, 0.7931, 0.8946 and 0.9738, respectively. The proposed method improves the segmentation accuracy of vessels and the segmentation of small vessels.
在电厂灵活运行期间,转子内部因温度梯度较大而产生热应力,导致转子疲劳损伤.而传统有限元分析热应力的方法无法满足实时监测的需求.研究了一种基于数据驱动的LSTM神经网络模型.模型具有从历史序列数据中学习深度信息的能力.通过多组超参数对比实验,发现在神经元数量6,单元节点28,学习率0.005,Dropout比例0.5时网络预测效果较好;在冷态启动过程下使用LSTM神经网络模型的热应力预测数据与有限元样本数据相比,RMSE为7.8740MPa,最大热应力误差9.7480MPa.结果表明,上述模型相比传统有限元计算时间大大缩短,在保证较高精度同时,也能够满足未来实时在线监测的需要.
针对大规模混流制造车间生产过程的调度问题,本文提出一种基于数字孪生系统的架构体系,结合复合调度模型,对大规模、复杂约束、不确定性的调度优化提供了较好的解决思路.明确了物理车间进行数据采集、清洗和融合的层次结构;提出了多层次多维度虚拟车间的构建机理,对多维度融合的基础单元模型构建进行了阐述;对复杂环境下的生产过程调度,提出了孪生数字驱动的传统模型、预测模型和知识模型的复合调度模型;最后对某大型发动机制造企业的复杂混流生产调度进行了实例验证.
由于在高灵敏紫外成像仪中紫外光与可见光成像原理存在一定差异,现有的方法配准精度不高,融合效果不理想.针对以上问题,提出了一种基于AlexNet网络与果蝇优化算法(Fruit fly optimization algorithm,FOA)小波融合的紫外光与可见光的图像的配准融合算法.首先,对预训练的AlexNet网络的权值阈值进行微调,利用AlexNet网络对图像进行各层次的特征提取,配合空间变换矩阵,实现紫外光与可见光的高精度配准.其次,将配准后的紫外光与可见光图像输入到FOA优化后的小波算法模型中实现融合.仿真结果显示,所提方法配准精度远远高于工业要求且收敛速度快,融合图像的信息熵较高.
生态景观是城市建设中不可或缺的一部分,如果规划不合理会导致土地面积利用率不高.为此提出基于CLUE-S模型的生态景观空间规划优化方法.通过土地利用转移网络分析景观土地转移方向,并通过景观土地动态度分析模型描述景观土地在规划过程中的变化速度.分析生态景观空间内各个景观规划要素对应的属性及其属性的空间相关特征,从斑块识别和廊道规划角度出发规划生态景观空间,运用CLUE-S模型模拟空间土地布局,求解空间规划中土地的面积,实现对生态景观空间规划结果的优化.实验结果表明,所提方法的Kappa指数高,说明其规划结果和实际景观之间具有一致性,上述方法的规划效果较好.
为改进车载数据采集及分析主机中GPU的散热器结构设计,提高系统主机在高工作频率下的散热效果,采用机器学习进行仿真结果预测,得到优选的结构参数组,加快模型仿真效率,探究不同参数对散热效果的影响.首先在Ansys软件中导入散热器的几何模型,根据流体动力学建立热仿真模型,对模型进行热仿真,确定各部件的温度分布情况得到温度仿真数据.然后建立以支持向量机为基础框架结构,采用了灰狼优化算法更新种群策略的一种高效的模型,通过与实际仿真结果对比验证了仿真预测模型的准确性,同时能使仿真过程达到79.59%的约简.对散热器进行结构参数优化,如散热器位置、长度、宽度和高度以及翅片片数、厚度和高度,获得了降低GPU温度的较优结构参数取值.
为实现无需协议的先验知识进行车载网络安全漏洞检测,提出一种基于深度卷积生成对抗网络的测试用例生成方法.通过引入卷积注意力机制来改进DCGAN,从空间和通道子模块出发,提高模型的特征表现力和生成CAN报文的格式正确性.搭建合适的生成对抗网络模型,将预处理后的CAN报文输入到模型中生成测试用例,对目标协议进行模糊测试.实验结果表明,所提方法能够有效发现总线协议的安全问题,且改进后的模型具有更高的测试用例通过率(87.7%)和漏洞发现能力(4.81%),测试效率显著提高.
