Abstract:Under the environment of intelligent manufacturing, supplier efficiency evaluation is very important for the development of intelligent manufacturing enterprises. This study constructs the supplier efficiency evaluation index system of intelligent manufacturing enterprises according to the characteristics of suppliers, the construction principles of evaluation index system, and literature summary. The weights of each index are determined by AHP-entropy method, and the supplier is graded by using the enterprise data of subject cooperation and BP neural network. The improvement suggestions for suppliers are put forward and further cooperation and communication between enterprises and suppliers are promoted. The example results show that this method has strong practicability for supplier efficiency evaluation.

Abstract:As a national major scientific engineering project, LAMOST currently has the highest observation and acquisition rate of the spectrum in the world, and provides a large amount of data and information resources for the research and development of astronomy. According to the stellar spectral data file released by LAMOST, the data about the wavelength of the stellar spectrum is extracted, and the data is subjected to noise culling, data dimensionality reduction, data normalization, and data dimensionality reduction processing. The BP neural network algorithm is used to classify the data, and the pros and cons of the BP neural network model are judged according to the correct rate of the classification results. However, the BP neural network test results of the test set data do not mean that it has the same test effect on other data and is easy to produce over-fitting, so the method of cross-validation combined with BP neural network is adopted. The BP neural network algorithm can test multiple sets of different data, obtain multiple sets of test results and obtain the average value, and obtain the relatively stable test results of the BP neural network model and reduce the randomness of the results.

Abstract:Fine-grained analysis of emotion is a branch of sentiment analysis, with the expansion of social network, the division of simple positive or negative coarse-grained sentiment analysis cannot satisfy the need of practical application. Thus the fine-grained emotional analysis based on evaluation objects and their attributes has received attention in recent years. The successful application of deep learning in the field of natural language processing in recent years provides a new idea for the fine-grained analysis of emotion. Take NLPCC2013 task 2 Weibo data set as the research object, explore the classification results of microtext in different neural network structures and add word vectors for optimization. Finally, the influencing factors and development direction of finer-grained emotion analysis of neural network micro-blog essay are analyzed and summarized.

Abstract:Since the existing function secret sharing schemes require all participants to join in the reconstruction phase. Therefore, it cannot be flexibly applied to real-world scenarios. A function secret sharing scheme with thresholds is constructed in this study using polynomial techniques. According to the security model of function secret sharing, we proved that the proposed scheme has security in the sense of information theory. In addition, this study analyzes the function secret sharing scheme proposed by Yuan et al., and expounds the reason why their scheme does not satisfy the security of function secret sharing. Finally, a comprehensive comparison between the newly constructed scheme and the existing function secret sharing scheme is found. We note that the newly constructed scheme has higher level of security and higher efficiency through the comprehensive comparison.

Abstract:In order to improve the prediction accuracy and reliability of photo voltaic power prediction output, this study proposes a photo voltaic power prediction method based on Stacking model fusion. The historical measured data such as temperature, humidity, and irradiance of a PV power plant are selected as the research object. Based on the feature intersection of the photo voltaic power data and the pre-processing and feature selection based on the model-based recursive feature elimination method, XGBoost and LightGBM are used. The three machine learning algorithms of Random Forest are the first layer of base learning for Stacking integrated learning. Linear Regression is used as the second layer of element learner to construct a photo voltaic power prediction model with multiple stacking models embedded in machine learning algorithms. The prediction results show that the R² and MSE of the method reach 0.9891 and 0.1358, respectively, and the prediction accuracy is significantly improved compared with the single machine learning model.

Abstract:The five-character quatrain is the treasure of Chinese traditional literature, rendering people a unique language aesthetic and aesthetic experience. The process of machine-generated quatrain has a positive exploration of machine acquisition on human language. Inspired by the rhythm and antithesis features of poetry language itself, we trained language memory model on poetry datasets and couplet datasets. The model consists of a semantic model and a textual rule model. The semantic model uses a one-dimensional convolution network to extract the semantic features of poetry and learn the semantic information of the poetry. The word model uses an encoder-decoder model with attention mechanism to learn the isocolon features of poetry writing. The experimental results show the language memory model can generate poems that conform to the rules of poetry and our aesthetics.

Abstract:Exploring human multi-channels perceptional characteristics is an important research field in physical psychology. This study innovatively designed a visual-tactile dot-matrix stimulator, as an experimental device for physical psychologist to analyze human perception. The device includes a host control platform, a sub-control module, a visual dot-matrix display module, a tactile dot-matrix display module, etc. The host control platform produces the visual-tactile instructions, and the sub-control module executes these instructions. The visual dot-matrix display module provides visual perception. It consists of 19 rows and 12 columns of 228 inch LED lamp arrays with 85 Hz refresh rate and 10 gray-value grades (7.6–75.8 lux). And the minimum package size of an LED lamp is 1.0×0.5 mm with alternative colors of red, white, yellow, and green. The tactile dot-matrix display module provides haptic perception by 6 rows of 4 columns of 24 contact arrays. Each contact has 1.5 mm diameter with 0.40 ±0.05 mm bump height and 10 Hz refresh rate. In the self-locking position case, the contact can endure 10 N force. The experimental results show that this stimulator can independently or combined control any LED lamps and contacts to provide the visual-tactile stimuli. This function shall satisfy for physical psychologist to acquire quantitative experimental data for analyzing human perception. This stimulator also has compact (70×66×80 mm volume), light (337 g), and mobile (15 m work distance) advantages.

Abstract:A multi-channel high speed data acquisition system is designed to collect acoustic emission signals of wood under long-term loading. Firstly, a four-channel acoustic emission signal hardware acquisition platform is built by using NI usb-6336 high-speed acquisition card, acoustic emission sensor and other hardware. Secondly, man-machine interface and software control system are designed based on Lab VIEW. Finally, the utility of the acquisition system is verified by wood three-point bending test. The experimental results show that the four-channel signal acquisition system can effectively collect and automatically store acoustic emission signals of wood during the damage process. As a kind of wood acoustic emission signal acquisition platform, this system can provide basic guarantee for the acquisition and analysis of wood acoustic emission signal.

Abstract:With the development of science and technology, the volume of hydrological information data has increased tremendously, how to make full use of these large-scale data to support decision-making is a big problem for scientists at present. Traditional water ecological carrying capacity analysis and calculation are complex and diverse, involving various types of data, with unsatisfied expansion, and focus on theoretical research and analysis. This work studies historical data, analyzes the factors affecting water ecological carrying capacity, divides the data into three layers, and proposes an analysis model of water Ecological Carrying Capacity based on Big Data (ECCBD). HDFS distributed file system of Hadoop cluster is used to implement the backup and storage of water ecological data, and MapReduce is used to implement the parallel computation of massive water ecological data. By comparing the output value with the water ecological carrying capacity, determining whether the water resources are surplus or deficit, the method and model proposed in this study can effectively analyze the current status of the aquatic environment from three different index layers: pressure, bearing capacity, and elasticity, it is of great significance to provide a basis for water ecological protection.

Abstract:Aiming at the high technical threshold of traditional Virtual Reality (VR) application development platform, which is not conducive to the popularization of VR, the draggable frame design based on state machine, extensible and low-coupling UI interface design and event system design in component form are adopted to optimize the common collision boxes and model details in VR. Hierarchical technical solutions, Solve the specific implementation of the data structure, multi-threaded implementation, design patterns and data storage and other issues. The final VR editing engine uses a fully visualized “drag-and-drop” design, and the released program can run on either PC or VR device. The engine is mainly used in the field of education and training. It is an experimental teaching assistant tool with easy operation, interesting and advanced. It can stimulate learning interest and improve learning effect. By dragging and dropping the mouse, the VR simulation software can be easily constructed in the “what you see is what you get” scene, which greatly reduces the development threshold and cost of VR simulation software.

Abstract:The college student’s practice innovation management platform aims to provide a network platform for teamwork and information exchange for students who are interested in participating in various discipline competitions, extracurricular practice, and innovative entrepreneurship activities. The management platform can also serve as an online platform for practical studios, engineering training centers, and other bases, mainly for management and publicity. The construction of practice innovation management website provides convenience for multi-disciplinary students to cooperate with each other and promote the integration of disciplines, which is conducive to the realization of the goal of training innovative talents. The platform is developed with Eclipse tools. Tomcat is used for server, MySQL is used for database, and Java is used for development language. JavaScript, CSS, Servlet, JSP, and other methods are also combined. The platform is simple and easy to use, and is very popular among students.

Abstract:The traditional circuit-switched telephone recording system has been unable to meet the needs of efficient and instant call recording in the new era administrative office due to its complicated structure, inconvenient storage, and non-real-time characteristics. Therefore, this paper presents a real-time recording system for power IMS telephone terminal to solve these problems. First of all, the article analyzes the business needs of real-time recording of telephone terminals in the power IMS exchange network. Secondly, the article introduces the implementation process of the system and clarifies the key technologies of the system: the system uses the recording server to parse the SIP message of its mirrored port for obtaining the media stream and decoding, and consistent Hash algorithm memory database is used as decoding data caching mechanism, and the message queue between the both is Ckafka. Finally, the performance of the recording server is analyzed based on the four indicators of response time, throughput, fault tolerance, and maximum delay. The analysis shows that the system has strong real-time performance, large throughput, certain fault tolerance, and load balancing.

Abstract:How to solve the problem of precise poverty alleviation and shaking off poverty in the whole country, ensure the precise target of poverty alleviation, the precise arrangement of projects and the precise use of funds, has become an important problem that the central government urgently needs to solve. Fujian Provincial Department of Finance takes the lead in researching and developing the Fujian Province online supervision system of poverty alleviation (benefiting people) funds by using advanced technologies such as big data, cloud computing, and microservice, so as to realize the accurate monitoring of the whole process of poverty alleviation (benefiting people) funds and ensure the accurate distribution of funds. In this study, the system architecture, key technology application, implementation effect and construction experience of the system are described in detail. The construction experience of this system can provide reference for other fields of national construction to use technologies such as big data, cloud computing, and microservices.

Abstract:The game digital industry plays an important role in the development of the digital economy. At present, there are many designs and developments for stand-alone games, but relatively few for multiplayer 3D online games. Combined with the principle of network communication, based on Unity3D, and using MVC architecture, this study designs and implements multiplayer online Action Role-Playing Game (ARPG). For the consistency of online player scenes, Communication protocol, such as position synchronization, weapon synchronization, damage synchronization, etc., has been proposed. Whether the game interface is friendly or not, would greatly affect the user experience of the player, and also affect the overall progress of the game. A general UI framework is designed for it. Due to the network delay, the system is difficult to achieve real-time and accuracy. The player’s position is synchronized, so this study uses the predictive position synchronization method, which effectively reduces the delay error during game running.

Abstract:Aiming at the problem of insufficient understanding of human body structure in the process of character animation and painting, and the advantages of virtual reality technology in education, a simulation system of artistic human anatomy drawing based on Forge cloud is proposed. According to the proportion of the human body structure, the system uses the block surface line model to complete the construction and visualization process of the human model. According to the law of animation motion, human motion simulation is realized by combining skeleton animation with three-dimensional motion capture. Human-computer interaction is completed through Forge cloud platform and Three.js. Finally, the manga module and Forge cloud module are communicated bidirectionally to complete the manga character posture simulation. Tests show that the system has a high degree of simulation and ease of use. It provides an environment for digital learning and mobile learning. It helps learners to understand the anatomical structure of human body deeply and grasp the modeling method of cartoon characters correctly.

Abstract:The mbedOS is a real-time operating system launched by ARM in 2014 for intelligent terminals and IoT nodes. It is mainly used in embedded systems with the high real-time response time. The study analyzes the common problems of transplantation and gives specific migration steps based on the in-depth analysis of the basic functions of mbedOS, scheduling mechanism, delay function mechanism, and communication mechanism between tasks. This work is based on the portable mbedOS engineering framework. On the basis, the mbedOS is implemented in different cores of ARM Cortex-M series and different MCU transplantation. The analysis of the common problems of transplantation between different development environments are given and the basis for the application research of mbedOS are provided. It effectively reduces the difficulty of mbedOS transplantation and can also provide reference for other RTOS transplantation.

Abstract:In order to realize the management and real-time visual display of the stored air quality data in the cloud, a cloud data monitoring and visualization system based on Spring Boot was established. The system was a B/S architecture, the Spring Boot framework was used to build a back-end micro-service instance to make the configuration and monitoring become simple, the Vue.js framework implemented front-end page development. The Axios plug-in was used to implement data interaction. It reduced the server overhead and response while implementing front-end connection and logical interaction. The system could query 11 kinds of air component information through the database, including: PM_1.0, PM_2.5, PM₁₀, CO, CO₂, NO, NO₂, O₃, SO₂, formaldehyde, TVOC, in addition to monitoring temperature, humidity, wind speed, coordinates, time and other related attributes.The system realized multiple functions such as data monitoring download, alarm management, home web front-end visualization, Baidu map visualization, etc. The program was deployed in the Alibaba Cloud, which was convenient for users to access web projects remotely.The combination of Spring Boot framework and Vue realized separation of front and back ends, which made the system to have sound stability, real-time and high efficiency.

Abstract:The automatic mapping algorithm for the transmission line network wiring diagram is a very complex global optimization problem. It involves two aspects: the automatic layout of the plant site and the automatic planning of the transmission line. In this study, a specific idea and algorithm for solving this problem are given. The issue is divided into three parts: the first part uses the force-oriented algorithm to make the initial layout of the plant station position, and uses the simulated annealing algorithm to perform iterative calculation, which is realized by concurrent technology. The gravitational and repulsion coefficients are selected to obtain the initial layout of the initial plant with the least cost. In the second part, the A* algorithm is used to plan the transmission line, and a cost model of the line direction is constructed. The cost model is used to standardize the line and obtain a beautiful line layout. In the third part, the layout results are evaluated and feedbacked, and the common layout defects are eliminated through the program, which reduces manual intervention. At the same time, the study also processed the historical line and the newly added line, so that the algorithm can realize the layout planning of the newly added station line without changing the layout of the historical plant station. The experimental results show that the graphical results obtained by the method satisfy the advantages of beautiful line planning, reasonable layout, less crossover, and less corners.

Abstract:The end-customer recommendation system is an effective tool for large-scale manufacturer terminal marketing. How to design a search method for finding the best target customer by collecting global market data in the Internet+ environment has become a challenge. To solve this problem, This study proposes a terminal customer recommendation method based on global market data perception (GMF). That is to use the idea of global analysis to preprocess the customer data nationwide, establish a comprehensive, multi-angle evaluation index, and obtain the target customer value. Then, through the method of domain subspace decomposition, the data is decomposed and analyzed in the domain subspace, and the customer evaluation criteria in a certain region are obtained. The analysis results of the two are effectively merged, and the similarity of the coupled objects is calculated, and the most similar TopN data is used as the best target customer result set. The experimental results on the data set generated by the large-scale manufacturer marketing activities show that the proposed algorithm is significantly better than the current mainstream collaborative filtering algorithm.

Abstract:Word segmentation is one of the most important steps in Chinese natural language processing, it is the basis for keyword extraction, automatic text summarization, and text clustering, the quality of the word segmentation directly affects the accuracy of further text processing. In recent years, with the rise of free public opinion platforms such as Microblog, live broadcast platform, and WeChat Moments, a large number of new words have brought great challenges to word segmentation methods. To solve the problem such as the small overall amount of new words, the flexible usage of new words, and excessive merging of words leads to the formation of phrase blocks in the process of new words discovering. This study proposed a new word detection algorithm combining correlation confidence and Jieba word segmentation. The algorithm is based on the preliminary word segmentation results by Jiaba library in Python, then calculates the correlation confidence between adjacent words to merge incorrectly split words into candidate new words, and by splitting the conjunctions to prevent multiple words from being connected into phrases. Compared with other confidence-based word segmentation methods, the proposed algorithm can greatly improve the accuracy of discovering new words, especially named entities and network terms, and reduce the length of new words while ensuring the integrity of new words. In the context of a small amount of test corpus, the proposed algorithm still has the ability to recognize low frequency new words.

Abstract:Aiming at the scientific, accurate, and operable regional independent innovation capability evaluation classification, a Decision Tree Genetic Algorithm and Back Propagation neural network (DTGA-BP) is proposed. The characteristics of the evaluation index are selected and the structure of the neural network is improved by optimizing the number of neurons in the hidden layer. The genetic operation of the nonlinear crossover probability value is combined with a new selection operator to optimize the initial weight and threshold of the BP neural network. The experimental results show that the evaluation results of the combined model are more scientific and accurate than the traditional subjective valuation method. Compared with the single BP neural network model and the GA-BP model, the classification accuracy is improved by 41% and 20%, respectively.

Abstract:Highway OD data is a kind of important data for highway operation management and condition analysis. How to use massive toll data to quickly generate and effectively manage highway OD data is an important problem in the current highway intelligent construction. Aiming at the problems of various types and long periods of highway OD data, a storage model of highway OD matrix based on Hadoop and corresponding calculation method are proposed. Two kinds of OD matrices are established as storage models, i.e. statistics of highway vehicle travel time and statistics of highway traffic flow. The comparison between the experiment based on massive real highway toll data and the traditional storage of highway toll data shows that the storage method proposed has better storage efficiency and saves storage space compared with the traditional relational data storage.

Abstract:Aiming at the load balancing problem caused by the imbalance of resource utilization of heterogeneous nodes in data center, this study proposes a virtual machine dynamic migration selection strategy based on dynamic threshold-based migration timing decision algorithm and load type perception-based selection algorithm. This strategy first dynamically adjusts the state threshold by monitoring the global load and the proportion of the high and low load nodes, and combines this threshold and load evaluation value to determine the migration timing. Then this strategy analyzes the virtual machine load type, selects the VM to be migrated based on the dependency of the virtual machine and the node resources, the current memory bandwidth ratio of the virtual machine, and the contribution of the virtual machine, and selects the destination node according to the resource matching degree and the migration cost of the virtual machine and the destination node. Thereby this strategy implements dynamic adjustment of virtual machines for high-load and low-load nodes to optimize node resource allocation. The experimental results show that this strategy can effectively reduce the number of virtual machine migrations and ensure the quality of data center services, and ultimately improve the load balancing ability of the data center.

Abstract:In view of the Non-Line-Of-Sight (NLOS) error of Ultra-WideBand (UWB) signal propagation in complex indoor environments, an adaptive UWB/DR co-localization approach based on Unscented Kalman Filter (UKF) is proposed. It combines the positioning information of UWB and Dead Reckoning (DR) by establishing an adaptive UKF filtering model. In this process, the principle of innovation and Gaussian distribution is used to detect whether the UWB positioning result contains NLOS error, and then the environmental adaptation coefficient, which is constructed by real-time estimation covariance and theoretical covariance of the innovation, dynamically correct the observed noise of UWB and make it adaptive to the real environment to reduce the impact of NLOS error on the positioning result to a greater extent. The experimental results show that the proposed approach can effectively reduce the NLOS error of UWB positioning, and because of the innovative introduction of environmental adaptation coefficient, it has higher positioning accuracy and stronger anti-NLOS error performance than UKF positioning and Particle Filtering (PF) positioning.

Abstract:The early detection of diabetes is of great significance for successful control of diabetes, prevention of complications, and reduction of prevalence. Existing diabetes diagnosis models based on machine learning have weak precision due to insufficient generalization ability. Therefore, this study proposes a multi-layer perceptron model combined with batch normalization to ensure the consistency of data distribution in the model. The proposed model is based on the PIMA training set for training evaluation. The experimental results show that the model has sound generalization ability in early recognition of diabetes, fast convergence, and high accuracy.

Abstract:Aiming at the problems of low recognition accuracy and poor anti-interference ability of traditional machine vision gesture recognition methods, a static gesture recognition method based on Support Vector Machine (SVM) gesture segmentation and transfer learning is proposed. This study uses SVM and transfer learning method to build a new gesture recognition model, uses SVM to segment the sample gesture, uses the Inception-v3 model as the basis of Convolutional Neural Network (CNN) model, carries out fine tuning on the network parameters, imports the sample processed by gesture segmentation into the model training, adjusts the super parameters using fine-tuning to get the new optimal gesture recognition model. The test results, obtained in disturbed environment, show that the recognition accuracy and real-time feedback efficiency of this method are higher than those of traditional methods, which can effectively recognize gesture and meet the practical application requirements.

Abstract:Traditional image fusion algorithms have many shortcomings, such as high computational complexity and inability to effectively extract image texture. To compensate these shortcomings of above traditional algorithms, an image fusion method is proposed based on the Siamese Convolutional Neural Network (Siamese CNN). First, we use the Siamese CNN to generate a weight graph, which contains all pixel information from the two images to be fused. Then, the image pyramid is fused in a multi-scale way, and the local similarity strategy is adopted to adjust the decomposition coefficient adaptively. Finally, several existing image fusion methods are compared. Experimental results show that the proposed method has sound fusion effect and is practical to some extent.

Abstract:As an important part of overhead transmission line, the safety of transmission tower will affect the operation of the whole power system. The construction of bird's nest is one of the important factors affecting the normal operation of transmission line, which needs to be monitored. Nevertheless, the existing monitoring methods not only are inefficient, but also require a lot of manpower and material resources. To cope with this phenomenon, this study puts forward a real-time detection method based on the algorithm of SSD. In addition, lead network VGGNet is replaced by ResNet-101 based on the network structure of SSD, so as to improve their ability of feature extraction. The Focal loss instead of Softmax loss improve SSD sample imbalances in the algorithm. And the data augmentation is used to increase diversity, in order to improve the robustness of the model. Experimental results show that the detection accuracy of the method proposed in this study is improved by 3.17% and 6.35% respectively in terms of accuracy and recall rate compared with the original SSD algorithm.

Abstract:In November 2013, General Secretary Xi Jinping first proposed the important idea of “precise poverty alleviation” when he visited West Hunan. In order to achieve the “precision” requirements, it is necessary to accurately identify poor households. For the convenience of the government to the precise poverty alleviation work effectively, this study analyzes the collected family information data and comprehensively considers that the information data based on multidimensional poverty contains discrete and continuous numerical values. And the characteristic data of the series has hierarchical characteristics. A model based on the new feature selection algorithm of REAHCOR and GBDT classification algorithm is constructed. The model is applied to the poverty rating evaluation system and has achieved sound results.

Abstract:In this work, the problem of collision detection of bloody particles and soft tissue organs in virtual surgery system was studied. The problem of collision detection between bloody blood and soft tissue in virtual surgery is different from that of traditional rigid body or software collision detection. The topological structure of bloody model changes greatly. The traditional method of collision detection by updating topology cannot ensure real-time and accuracy. A collision detection algorithm for bloody particles and software based on space partitioning is proposed, which can handle collision detection between software based on Smoothed Particle Hydrodynamics (SPH) simulation and software simulated by any dynamic model. At the same time, the uniform space grid established in the nearest neighboring particle search of SPH algorithm is proposed to be reused. The space grid is used for the space division of collision detection and the localization of fluid particles, thus reducing the time and space resources repeated consumption. Experimental results show that the algorithm can meet the accuracy and real-time requirements of collision detection between bloody particles and software in virtual surgery.

Abstract:Aiming at the limitation of the clustering effect caused by the preference and damping factors in Affinity Propagation (AP), a Teaching and Learning Based Optimization (TLBO) algorithm is proposed. First, the search space of parameter p is determined, and then the TLBO algorithm is used to find the optimal parameter value in the search space. At the same time, the damping factor is automatically adjusted to prevent numerical oscillations during the clustering process, so as to improve the clustering quality of AP algorithm. The experimental results show that the algorithm can effectively solve the problem caused by preference and damping factors, improve the contour coefficient of clustering, and reduce the clustering error rate.

Abstract:Aiming at the problem that the room name and room number (ID) cannot be displayed in the 3D view of Revit model, a middleware is designed to automatically label room information in Revit 3D model. This method uses object-oriented C# programming language and Revit API extension method. Firstly, the building components of each floor plan of Revit model are identified and the room coordinate data information is extracted. Combining SQL server, BIM database (Epplus library) is established to store the extracted data information. Finally, using Visual Studio and Revit platform, coordinate data matching method is adopted to realize automatic annotation of room information of Revit 3D model. The example simulation proves that the method can automatically and accurately locate and identify rooms in Revit model, complete labeling, and improve the work efficiency of designers and drawing examiners.

Abstract:There is an important application value for the research of vehicle active safety technology through the recognition of drivers’ emotional state. In this study, seventeen subjects’ frontal dual-channel EEG signals were collected by emotional video induction method, and EEG characteristics of different emotions were extracted. After dimensionality reduction, the data were classified by multiple classifiers. The results show that compared with single-core classifier and ensemble learning classifier, Gradient Boosting Decision Tree (GBDT) algorithm has the highest recognition accuracy of happiness and sadness. This study provides a new method for real-time monitoring and recognition of drivers’ emotional state, and provides a theoretical guarantee for improving driving safety.

Abstract:Data collection in modern industrial production processes requires high sampling rates and high real-time transmission, while existing OPC clients of data acquisition cannot meet the requirements. In order to solve this problem, by analyzing the OPC standard, accessing the server data interface and component object model, combined with the characteristics of data acquisition in current industrial control, the OPC client based on subscription data collection is designed and implemented. This solution has high application value in solving the problems encountered in current industrial data collection. And in the actual production environment, it verifies the stability and real-time performance of the data transmission with the standard OPC server, which provides a reliable data foundation for production process control.

Abstract:Image captioning is an important task, which connects computer vision and natural language processing, two major artificial intelligence fields. In recent years, encoder-decoder frameworks integrated with attention mechanism have made significant process in captioning. However, many attention-based methods only use spatial attention mechanism. In this study, we propose a novel dual refined attention model for image captioning. In the proposed model, we use not only spatial attention but also channel-wise attention and then use a refine module to refine the image features. By using the refine module, the proposed model can filter the redundant and irrelevant features in the attended image features. We validate the proposed model on MSCOCO dataset via various evaluation metrics, and the results show the effectiveness of the proposed model.

Abstract:Video scene recognition has attracted much attention in the field of machine learning and computer vision. It is not only an important practical application, but also a challenge for image understanding in the field of computer vision. Nevertheless, current exploration of video scene recognition has not been unable to meet the needs of production environment. And most proposed models only use video-level feature information, while ignore association of multi-granularity video feature. In this study, we propose an architecture of attention mechanism with multi-granularity video features, which can make use of the rich semantic association among the various dimensions of video information dynamically and efficiently, and improve the performance of the model. The experiments are conducted on the latest VideoNet dataset released by CCF China MM 2019. The result shows that the proposed model based on attention mechanism model with multi-granularity video features outperforms the previous methods.

Abstract:Due to the large amount and variety of LTE network data, the manual drive test analysis has been unable to meet the current requirements for poor quality cell detection based on drive test data. In order to improve the efficiency and accuracy of the poor quality cell detection, machine learning is gradually applied in the detection of poor quality cell. In this study, a poor quality cell detection method based on four-dimensional feature of distance is proposed for the small number of road survey data. This method uses clustering algorithm and artificial judgment to calibrate road test data. And it compares the extraction effect of the distance based four-dimensional features and the traditional two-dimensional features. The featuresare classified by logistic regression classifier, decision tree classifier, support vector machine classifier and k-nearest neighbor classifier. The experimental results show that the distance-based four-dimensional features are more beneficial to the detection of quality difference cells than the traditional two-dimensional features. Support vector machine classifier works best when four-dimensional features are used for classification.

Abstract:Motion simulation on ramp is an important part of the whole motion simulation of high-speed railway. In order to study the motion simulation of high-speed railway on ramp, this study puts forward a simulation method based on virtual reality. First, the high-speed railway model and terrain are constructed. Then, high-speed railway is abstracted into rope model, and the force and motion state of high-speed railway on ramp are analyzed from a dynamic point of view. Finally, related blueprint nodes are connected in virtual reality engine Unreal Engine, and it is applied to the high-speed railway simulation using the complete suite and real-time high-fidelity rendering effects provided by it. It provides a satisfied means and method for studying the design of high-speed railway line selection, simulated operation, and the adaptability of parameters such as train speed and terrain, so that the simulation effect can be matched with the actual operation status of the train to the maximum extent, and also to provide reference for other related studies.

Abstract:Code change which introduces risk in software quality is associated with regression test case prioritization. It is an important topic that evaluates the code change impact on regression test case prioritization, which plays a significant role in software quality assurance. This study analyzes the relationship between regression test case and code change from a test coverage and coupling perspective based on the test case prioritization evaluating model, and presents a new code change impact metric model which introduces both the dominant and recessive impact level. Experiments indicate that, the quantitative metric results of code change impact to regression test cases prioritization computed by this model are comprehensive and objective, and could provide effective support for regression test case prioritization evaluation.

Abstract:In the text classification methods, the text representation based on the Word2Vec ignores the weight of words in distinguishing text. The method of combining Word2Vec weighted by TF-IDF and CNN is designed. In news text classification, the importance of news title is always neglected. Therefore, this study proposes an improved TF-IDF method, which takes both news title and body into account. Experiments show that the news text classification method based on weighted word vector and CNN has a greater improvement than the logistic regression classification. And its effect increases by 2 or 3 percentage points than the un-weighted method.