Abstract:Amid the growing scale and complexity of software systems, their configuration has become an essential topic in the field of software engineering. Massive complicated configuration entries bring difficulties in correctly deploying and using software systems; for instance, misconfigurations will lead to performance degradation and significant losses. Researchers are devoted to handling software configuration, mainly for resolving software misconfiguration. This paper presents a systematic literature review on the work of software configuration, including the research status and main achievements. It first proposes a two-dimensional analysis framework for the research work from the perspectives of software lifecycle and techniques. Based on this framework, this paper analyzes and categorizes the state-of-the-art results. Finally, it summarizes the characteristics of the current work and envisions beneficial prospects of future work.

Abstract:Recently, sparse representation has made outstanding contributions to signal and image processing, target recognition and blind source separation. Firstly, the history and rationale of sparse representation are reviewed to summarize the existing sparse representation optimization algorithms. Secondly, those algorithms are divided into the greedy algorithm, the constraint algorithm and other algorithms. The basic principles and characteristics of the first two categories are elaborated respectively, and their representative algorithms and state-of-the-art applications are summarized as well. In addition, the five representative algorithms of the greedy algorithm are simply compared. Finally, the applications of sparse representation in various fields and its general outlook regarding existing problems are offered.

Abstract:Because the rules of the rule-based interpretability model may fail to reflect the exact decision-making situation of the model, the interpretability framework combining machine learning and knowledge reasoning is proposed. The framework evolves a target-feature result and a reasoning result, which implements interpretability when the two are the same and both reliable. The target-feature result is obtained directly by the machine learning model, while the reasoning result is acquired by sub-feature classification combined with rules for knowledge reasoning. Whether the two results are reliable is judged by calculating their credibility. A particular recognition case of cervical cancer cells for TCT image fusion learning and reasoning is used to verify the framework. Experiments demonstrate that the framework make model’s real decisions interpretable and improve classification accuracy during iteration. This helps people understand the logic of the system’s decision-making and the reason for its failure.

Abstract:Image caption represents a research hotspot in the field of image understanding. In view of the poor quality of sentences, we propose Chinese image caption combining dual attention and multi-label images. We extract visual features and multi-label text firstly, and then use multi-label text to enhance the correlation between the hidden state of the decoder and visual features. Next, we redistribute attention weights to the visual features according to the hidden state of the decoder and decode the weighted features into words. Finally, the words are output in a time sequence to obtain Chinese sentences. Experiments on Chinese image caption datasets, Flickr8k-CN and COCO-CN, reveal that the proposed method substantially improves the quality of sentences.

Abstract:The one-line diagram for a distribution network, as an electrical wiring diagram, uses the specified graphic symbols and equipment connections to visualize the distribution network, significantly improving the efficiency of distribution network management. Aiming at the actual needs for automatic generation of one-line diagrams for a distribution network, this paper provides an automatic generation method based on the trunk-branch line model. The method first creates the trunk line and branch lines and calculates the branch layout direction; then plans the initial branch layout and performs partial stretching; and finally enables branch partial contraction and automatic routing. It has been verified that this automatic generation method can automatically generate the one-line diagrams that meet the requirements of accuracy and aesthetics.

Abstract:In the field of Internet-based medical treatment, AI-based triage represents a key link, which allocates patients to departments according to conditions, disease attributes, medications, etc. We can adopt the BERT with a deep bi-directional Transformer structure for language model pre-training to enhance the word semantics; however, the text description of patients’ conditions offers sparse information, which is not conducive to the full learning of characteristics by BERT. This paper presents DNNBERT, a joint training model integrating knowledge. Combining the advantages of Deep Neural Network (DNN) and the Transformer model, DNNBERT can learn more semantics from text. The experimental results prove that the computing time of DNNBERT is 1.7 times shorter than that of BERT-large; the accuracy rate of DNNBERT is 0.12 higher than the F1 value of ALBERT and 0.17 higher than that of TextCNN. This paper will provide a new idea for sparse feature learning and the applications of deep Transformer-based models to production.

Abstract:According to the Moore’s Law, the scale of integrated circuits is getting bigger, and the integratable circuits within a single chip are increasingly complex. In a research and development cycle of an SoC chip, pre-layout verification becomes harder with more complex chip functions, taking uncontrollable time. How to reliably and efficiently verify complex chips within limited time represents a challenge to be addressed. In response to this problem, this paper customizes a UVM-based universal verification platform for AMBA bus interface. The platform is equipped with a scalable structure and random verification incentives, achieving dependable results. It can verify the modules to be tested in AMBA-APB, AMBA-AHB, and AMBA-AXI interfaces. In addition, the verification platform can be quickly set up for the target, and the preparation for pre-layout verification is simplified. The UVM-based platform produces random data with constraints, and verification results are converted into coverage reports, ensuring the efficiency and completeness.

Abstract:With the progress in artificial intelligence and other related technologies, the intelligent robot technology has been greatly advanced. Its increasingly mature software and hardware facilities put forward new requirements for the research and development of the robot operating system and architecture. In this paper, a traditional widely-used robot operating system, ROS, is introduced and analyzed, and the improved robot operating system, ROCOS, is compared with ROS to analyze the optimized system architecture. Additionally, the organizational frameworks of the both are detailed, and the simulation experiment is conducted to test the rationality and performance of the frameworks. The presented work can serve as a reference for the subsequent research and development of new robot operating systems.

Abstract:The whole lifecycle hosting platform of deep learning offers a web solution to experimental tasks and boosts the application of deep learning technology in production and life. To address the problem of training image recognition models by the platform, this study designs and implements a distributed task execution system for experimental tasks. The system is composed of modules for resource monitoring, task scheduling, task execution, and log management. It schedules tasks according to indicators, such as resource utilization, executes tasks in Docker containers and collects generated log data in real time. The test results demonstrate that the system fulfils the normal functional requirements, achieving the desired targets regarding reliability and stability while reducing about 20% of training time after being integrated into the deep learning platform.

Abstract:The rebar represents an essential material in civil engineering. In the rolling process, roll wear, billet quality, and other factors will cause surface defects. If they cannot be detected in time, a large number of waste products will be produced, seriously affecting the economic benefits of enterprises. In this study, a detection method of rebar defects based on deep learning is proposed. Images of rebars are collected by industrial cameras in the production site, and their surface defects are classified and labeled to establish sample datasets that are further enhanced by the Deep Convolutional Generative Adversarial Network (DCGAN). Faster RCNN is adopted to construct the detection model of rebar defects, which can identify the surface defects in a small sample size with migration learning. In addition, the detection model is optimized through the evaluation of the setting of loss function, optimization methods, learning rates and sliding average. The experiment reveals that the method can effectively solve the problems of low efficiency and high false detection rates caused by manual detection, with good stability and practicability.

Abstract:Objective: Establishing the Quality Assurance (QA) standardization and digital analysis system of linear accelerators (linacs) is an effective way to improve the level and quality of radiotherapy. Methods: Based on the Pylinac function library, a QA digital analysis system is built with the Django framework and the MySQL database structure. The stability and practicability of the system are evaluated through clinical tests. Results: The QA digital analysis system can substantially reduce the time of calculation and analysis in the QA process, while facilitating the monitoring and review of linac operation. Moreover, it can help new physicists in the radiotherapy department to master the QA process quickly. Conclusion: The QA digital analysis system not only simplifies the QA workflow and improves the working efficiency, but also promotes the QA standardization for the treatment system of radiotherapy linacs.

Abstract:The current print service is faced with many security challenges, such as network attack and data leakage, and its security level completely depends on the information security of the external environment. As such, a secure printing architecture based on private cloud and intelligent defense is introduced in this paper. The architecture with private cloud as the core provides a unified and transparent access interface for print service by virtual printing. It monitors and manages the printing business flow on the basis of authentication and printing security policies. Meanwhile, it securely isolates the printing output device from the client network and the data center network with an access control mechanism for the end point, realizing the on-demand access to print service and intelligent defense against network exceptions. The Jmeter-based stress testing and the hping3-based security testing demonstrate that the system with this architecture has good user experience and strong robustness. To be specific, it spends less than 2 s successfully handling 100 consecutive requests from 400 concurrent clients respectively for submitting and outputting print jobs when it is not attacked, and the exception rate of outputting print jobs for the same requests is only 3.62% when the system is attacked by 5000 SYN packets/s.

Abstract:The book localization system based on RFID technology produces few desired results in universities; the system needs to be reconstructed because finding target books is still challenging. The library’s localization system of China University of Petroleum based on RFID technology is adopted in this study. Combined with the practical procedures of circulation and shelf arrangement, the frameworks of Python and Django and the Oracle database are used to develop the backstage management module of the system. In addition, the localization data collection, database structure, and the calculation of localization are upgraded. The improved book localization system based on RFID technology is easy to manage and maintain, effectively solving the problem of finding target books.

Abstract:Sleep quality, which influences human health, can be greatly enhanced by a pillow with proper height. Substantial studies have revealed that the pillow’s height for persons lying on their sides should be greater than that for them lying on their backs. This paper introduces an adaptive pillow based on deep learning, which can recognize human sleeping positions to adjust the pillow’s height. The paper also presents hardware platform design and the construction and transplantation of neural network models. First, the pressure sensor and the air pressure sensor embedded in the pillow respectively collect the pressure of the head on the pillow and the air pressure in the pillow airbag to generate a time-series data frame. Then, the 1DCNN-GRU network model identifies and classifies the sleeping positions. Finally, the pillow’s height is adjusted according to classifications.

Abstract:In this study, we propose an intelligent and green-based urban garbage collection and transportation system by combining the common software development methods and database technology. This aims to improve the urban appearance, reduce the empty-loaded rate of vehicles, and realize source traceability and transparent management of the whole process. The system is designed with Beidou positioning data as a basis and the Gaode open platform as assistance. The development of the system relies on the ThinkPHP framework based on the PHP language and the design mode of B/S architecture, and the client adopts Ajax and Web technologies for data transmission and display. The goal of this system is to provide users with real-time location and historical trajectory queries of vehicles, user management, statistical analysis, and other services. For the entire trajectory data during vehicle transportation, the system applies the Beidou series to data collection, which more effectively maintains the safety of information and technology. In the later stage, it can also be combined with machine learning and other techniques to analyze the driving behavior of drivers and the health of vehicles. In conclusion, the system combining Beidou positioning data with Internet techniques facilitates efficient management of vehicles.

Abstract:Since medical data are professional, complex, and diverse, sample data are lacking during the development of artificial intelligence in the medical field. In this study, we design and develop a multidimensional medical image data management system for algorithm researchers. The system includes data modules of preprocessing, management, and visualization. It can semi-automatically generate medical image data with classification labels, efficiently manage them, and display their distribution characteristics by graphs or tables.

Abstract:The server management controller, as a key component of cloud computing equipment, is mainly based on ARM architecture at present. However, ARM’s high licensing fees push up its design cost, which is not conducive to the iteration and upgrade of SoC-related products. RISC-V, which is an open-source processor architecture proposed recently, belongs to the reduced instruction set as ARM. It has many features such as modularity and scalability. In this study, we combine the BOOM core of a RISC-V open-source processor to design and implement an FPGA prototype system of a server management controller based on a RISC-V processor. In the system, we build a prototype based on Xilinx’s Virtex Ultra Scale 440 FPGA and complete the functional and CoreMark tests in the actual application scenarios. The results show that the processor performance is improved by 26%, which is superior to that of products of the same level that use ARM as the core, and the system’s functions are in line with design expectations. In addition, based on OpenBMC, this system implements special management control protocols such as IPMI, and its basic functions are verified, which proves the feasibility of optimizing SoC architecture by replacing ARM with RISC-V.

Abstract:Since skin melanoma images are featured by large intraclass differences and small sample datasets, the deep residual network can effectively solve the problem of over-fitting during training and improve the recognition accuracy. However, the network model has many training parameters and high time complexity. To improve the training efficiency and the recognition accuracy, we theoretically analyze its structure. By modifying the network structure, we replace the convolutional and pooling layers in the residual network with the Inception structure to lower the number of training parameters and the time complexity of the model. On this basis, we propose an Inception Deep Residual Network (IDRN) based classification and recognition algorithm for skin melanoma, where the Inception structure and the SeLU activation function respectively replace the convolutional and pooling layers and the traditional ReLU function. Subsequently, experimental validation is carried out on the published ISIC2017 dataset of dermoscopic images of melanoma. The theoretical and experimental results show that compared with the traditional convolutional neural network ResNet50, the proposed algorithm reduces time complexity and improves recognition accuracy.

Abstract:Monarch Butterfly Optimization (MBO) is a meta-heuristic algorithm proposed in 2015 to simulate the migration behavior of monarch butterflies. There are two problems in the MBO algorithm when dealing with high-dimensional problems: It is easy to fall into local optimum, and the offspring generated by migration operators are greatly influenced by their parents. For these reasons, we propose a new algorithm, Monarch Butterfly Optimization with a Logistic Chaotic Map (LCMMBO). It uses a Logistic chaotic map to disturb the optimal solution and optimizes the offspring transfer mode in the migration operators. These operations respectively aim to enhance the ability to jump out of the local optimum and the ability of global search. The simulation results show that, in the case of handling high-dimensional optimization problems, the new algorithm enjoys excellent robustness and a strong ability to jump out of the local optimum.

Abstract:The research on 5G communication technology and the construction of new infrastructure has witnessed the rapid development of the smart grid. In the era of big data, the Internet of everything leads to the access of massive equipment to the power network, which also brings a great burden to the smart grid, and the stability problem of the power network is urgent to be solved. Therefore, we propose a prediction algorithm for smart grid stability based on Convolutional Neural Network (CNN). It collects the data generated by the power network, processes them in the CNN model, and finally outputs the judgment results of smart grid stability. The simulation results show that the algorithm has higher accuracy than SVM, AdaBoost, and random forest. Furthermore, four different optimization algorithms are used to improve the CNN model. SGD algorithm with momentum can achieve a prediction accuracy of 98.13%. The proposed model can effectively help the power system to pre-warn the unknown problems, reducing the security risks and power accidents.

Abstract:A basic mathematic library is one of the most basic and core underlying software in high-performance computers. Its performance directly determines the efficiency of the upper computing program. The use of rfpcr and wfpcr instructions in some functions of the current domestic SW basic mathematic library leads to the interruption of the pipeline, which reduces the performance of the functions. To solve this problem, we propose an equivalent substitution method for functions in the instruction segments by combining the effects and instruction characteristics of the functions. The experimental results show that this method can improve the performance of the functions by 27.83% on average.

Abstract:Particle Swarm Optimization (PSO) can easily fall into the local extremum and has slow convergence and low precision in the late evolution. For these reasons, we propose an Improved Particle Swarm Optimization (IPSO) algorithm that integrates multiple strategies. It includes the following four improvements. Firstly, the grouping strategy is adopted. According to the fitness values, the population is divided into an optimal particle group and an inferior particle group, which are subject to crossover and mutation operations, respectively. Secondly, the elite strategy is used to update the population. The first 50% particles are selected from the population after crossover and mutation operations and the initial population according to fitness values and taken as a new population. Thirdly, the particle learning mode is improved to make full use of the population information. The particle best is replaced with the mean of the optimal particle group. Fourthly, probability control is introduced to control the probability of the algorithm’s entering crossover and mutation operations. The simulation results of the test functions show that compared with the standard PSO and its improved variants, the IPSO algorithm can effectively take into account the global exploration and local mining capabilities, and has the advantages of fast convergence, high accuracy, and avoidance from the local optimal solution.

Abstract:With the vigorous development of the electric vehicle industry in China, the problem of charging electric vehicles has gradually emerged and become a bottleneck restricting the development of this industry. The long charging time of electric vehicles, the insufficient number of urban charging piles and the uneven distribution of urban charging stations are the direct reasons for the charging problem. We propose a charging scheduling algorithm for electric vehicles based on traffic conditions. It schedules the charging vehicles in a unified manner by comprehensively considering the actual situation in the road network. In addition, a vehicle scheduling model is established and simulation experiments are carried out. The results show that the algorithm can reduce the charging time of electric vehicles, balance the loads of charging stations in a region, and improve the overall charging efficiency.

Abstract:The existing 3D face reconstruction models have the problems of high complexity and poor reconstruction accuracy of multiple face poses. For these reasons, we propose a convolutional neural network that can effectively achieve face alignment and reconstruct a 3D face from a single face picture in the case of a variety of face poses. First, we design an encoder-decoder network composed of a DenseNet module and a deconvolution module. The evaluation of image Structural SIMilarity (SSIM) is introduced into the loss function to construct a new loss function. Then, we train the neural network to get a model, which implements face alignment and 3D face reconstruction tasks. Experiments on the ALFW2000-3D dataset show that the proposed network effectively improves the accuracy of face alignment and reconstruction.

Abstract:In this paper, we propose an algorithm based on YOLOv4 to solve the problem that manual inspection and traditional video monitoring methods cannot identify the external hidden dangers of transmission lines in time. In this algorithm, cluster analysis is performed with the improved K-means algorithm on the size of the targets in the image sample set to select the anchor frames that conform to the characteristics of detection targets. After that, the CSPDarknet-53 residual network is used to extract the deep-seated network feature data of the images, and the feature map is processed by the SPP algorithm to increase the receptive field and extract higher-level semantic features. Finally, in combination with the monitoring pictures of transmission lines, the test results show that the proposed algorithm can detect external hidden dangers timely and accurately.

Abstract:With the rapid development of the computer field, the increase in software size and complexity leads to more exposed software crises. As such, improving the efficiency of software production has become an urgent task for the software-related industry. Component-based software integration is the main way to solve the related problems. On the VxWorks system, we can adopt a component-based software integration method defining standardized operation interfaces for software components in functional forms. This method is mature in engineering practice but cannot be reused in cross-heterogeneous environments. Therefore, to be reusable in the heterogeneous environments, a component-based software integration method based on class reflective mechanism is proposed, which combines the factory mode with the callback mechanism to generate the basic elements of the reflective class. On this basis, it standardizes the description and implementation of software component classes and manages the integration of software components in the whole life cycle. In this method, the whole process is implemented by the cross-platform unified code language program, which enables it to be reused more conveniently in business information systems. The designed software-component base class and integrated management class, with complete structures and clear descriptions, are standard and user-friendly and can well support the functional adaptation and update of components and the assembly and evolution of software systems. The experiments show that this method is suitable not only for VxWorks, Android, and Windows systems but also for domestic system environments such as ReWorks, AOS, and Kylin.

Abstract:The neural network ranking model has been widely used in the ranking task of the information retrieval field. It requires extremely high data quality; however, the information retrieval datasets usually contain a lot of noise, and documents irrelevant to the query cannot be accurately obtained. High-quality negative samples are essential to training a high-performance neural network ranking model. Inspired by the existing doc2query method, we propose a deep and end-to-end model AQGM. This model increases the diversity of queries and enhances the quality of negative samples by learning mismatched query document pairs and generating adversarial queries irrelevant to the documents and similar to the original query. Then, we train a deep ranking model based on BERT with the real samples and the samples generated by the AQGM model. Compared with the baseline model BERT-base, our model improves the MRR index by 0.3% and 3.2% on the MSMARCO and TrecQA datasets, respectively.

Abstract:Human centrifuges are getting more attention for their ability to provide continuous high G-load simulation. We analyze the application demand and research status of fidelity evaluation of human centrifuges based on motion perception in this paper. The evaluation model considers the characteristics and influence of the human perception threshold. The membership function of motion perception evaluation is established by fuzzy evaluation. Regarding the characteristics of motion perception, we focus on analyzing the influence of the direction error of G-load simulation on the fidelity of perception simulation. To improve the evaluation rationality in the case of close memberships, we establish the corresponding rules with strict evaluation principles, enabling more reasonable fidelity evaluation. Moreover, two groups of flight data are evaluated and compared.

Abstract:To solve the problems of network information security caused by the rapid development of the Internet, predecessors proposed an image information hiding technology based on the least significant bit algorithm. This method cleverly uses the redundant attributes of media information to hide secret information in other media information, obtaining a carrier embedded with secrets that can be publicly disseminated. The information can be stored and transmitted because lawless people cannot detect the hidden data. Previously, the concealed information is limited as only the lowest digit of secret information is hidden. To solve this, we embed every two consecutive digits of the secret information in two of the least four digits of every byte in the carrier image on the principle of “matching first and making replacement if it fails”. In this way, the hidden amount of the traditional least significant bit algorithm is increased. The results show that the algorithm can meet the needs of users for information hiding due to its large hidden amount, little reduction in image quality after hiding, high safety, and ability to correctly extract secret information.

Abstract:Currently, the widely used first-order deep learning optimizers include non-adaptive learning rate optimizers such as SGDM and adaptive learning rate optimizers like Adam, both of which estimate the overall gradient through exponential moving average. However, such a method is biased and hysteretic. In this study, we propose a rectified SGDM algorithm based on difference, i.e. RSGDM. Our contributions are as follows: 1) We analyze the bias and hysteresis triggered by exponential moving average in the SGDM algorithm. 2) We use the difference estimation term to correct the bias and hysteresis in the SGDM algorithm, and propose the RSGDM algorithm. 3) The experiments on CIFAR-10 and CIFAR-100 datasets proves that our RSGDM algorithm is higher than the SGDM algorithm in convergence accuracy.

Abstract:In the cloud manufacturing environment, manufacturing resources and capabilities are encapsulated in the form of services. Different tasks are collected to the cloud platform via the cloud and corresponding services are assigned to each task through appropriate scheduling. Due to the uncertainty in task execution, emergency can take place, forcing the remaining tasks to be rescheduled. In addition, the complexity and diversity of tasks in the cloud manufacturing environment will lead to difficulty in finding the optimal solution within a reasonable time period. With the maximum completion time of all the tasks as the optimization goal, a metaheuristic algorithm that combines an improved genetic algorithm and the neighborhood search technique is proposed to tackle the rescheduling caused by the uncertainty of tasks and resource services in the cloud manufacturing environment. The experimental results show that the proposed algorithm can deal with the rescheduling during dynamic scheduling and obtain the optimal solution quickly.

Abstract:Recently, attention mechanisms have been widely used in computer vision in such aspects as the common encoder/decoder framework for image captioning. However, the current decoding framework does not clearly analyze the correlation between image features and the hidden states of the Long Short-Term Memory (LSTM) network, leading to cumulative errors. In this study, we propose a Similar Temporal Attention Network (STAN) that extends conventional attention mechanisms to strengthen the correlation between attention results and hidden states at different moments. STAN first applies attention to the hidden state and feature vector at the current moment, and then introduces the attention result of two adjacent LSTM segments into the recurrent LSTM network at the next moment through an Attention Fusion Slot (AFS) to enhance the correlation between attention results and hidden states. Also, we design a Hidden State Switch (HSS) to guide the generation of words, which is combined with the AFS to reduce cumulative errors. According to the extensive experiments on the public benchmark dataset Microsoft COCO and various evaluation mechanisms, our algorithm is superior to the baseline model and can get more competitive attention results.

Abstract:Intelligent customer service accurately answers users’ inquiries with artificial intelligence technology, and a good sentence similarity algorithm can improve the accuracy of questions and answers. In this study, we focus on the customer service in financial securities and propose a multi-feature fusion-based sentence similarity algorithm model, which improves the intelligence of customer service. Matrix splicing is adopted to fuse the morphological and semantic features of user question sentences and knowledge base sentences. The morphological features include N-gram similarity, edit distance, and Jaccard similarity. To extract semantic features, we propose a multi-head attention based neural network model named LBMA. With these fused features, a machine learning classifier is used to determine whether two sentences are similar, and the classification result of the classifier is regarded as the calculation result of the multi-feature fusion model. On the premise of not changing the semantic information as much as possible, the data set is expanded through data augmentation to improve the generalization of the model. Experimental results show that the proposed model can calculate the similarity of customer service data sets at an accuracy of 94.69%, higher than that of existing methods.

Abstract:The Basic Linear Algebra Subprogram (BLAS) is a mathematical function standard for basic linear algebra operations. The library function is divided into three levels in which basic operations between vector and vector (level 1), vector and matrix (level 2), and vector and vector (level 3) are offered. In this paper, we study the optimization scheme of BLAS level1 functions on SW1621 processor. With the function AXPY as an example, the architectural characteristics of the platform are fully used to optimize its performance, and an automatic thread allocation scheme is designed. The experimental results show that compared with the reference implementation version of GotoBLAS, the optimized BLAS level1 function, AXPY, has a high single-core acceleration ratio of 4.36 and a multi-core one of 9.50 respectively. Every optimization scheme can improve the performance.

Abstract:The prediction accuracy of the Remaining Useful Life (RUL) is of vital importance to rejuvenation decision of Web-based software systems, so we propose a real-time prediction method for the remaining useful life of Web-based software systems based on the Long Short-Term Memory (LSTM) network. Firstly, an accelerated life test platform is built to collect the performance indicators of the aging of Web-based software systems. Then, according to the time-series characteristics of indicator data, an LSTM-based real-time prediction model for the remaining useful life of Web-based software systems is constructed and trained. The experimental results show that the model can effectively predict the remaining useful life of Web-based software systems in real time with higher accuracy and stronger applicability. This method provides technical support for optimizing system’s rejuvenation decision.

Abstract:Webshell is a highly concealed tool for Web attack, which is used to obtain the operating authority of servers. When writing Webshell, the attacker uses a series of anti-virus techniques to bypass the firewall, which leads to ineffective Webshell detection by existing methods. In response to this situation, we propose a Bi-GRU-based Webshell detection method from the perspective of text classification. Firstly, this method compiles webpage script files to obtain the opcode instructions. Secondly, the instructions are converted to feature vectors by the Word2Vec algorithm. Finally, a variety of deep learning models are used for training with accuracy, false positive rate, and false negative rate as evaluation criteria. The experimental results confirm the feasibility of the Bi-GRU-based Webshell detection since it is better than other algorithm models.

Abstract:On the basis of analyzing metadata standard specifications of major open data, we integrate the open data described by various metadata standard specifications and construct the field mapping table of core metadata of open data, providing a solution for the cross-platform sharing of open data. The index fields needed for metadata mapping are extracted, and the mapping data are in the formats of JSON and XML, respectively. Through the metadata mapping, the open data described by different metadata standard specifications are incorporated into the same framework. The mapping mechanism in this paper can effectively realize the interconnection of open data resources on different platforms and enhance the sharing of open data.

Abstract:This study aims to deal with the blur and color distortion of images collected by imaging equipment in hazy weather, enhance the detail information of foggy images, and improve the contrast and clarity of original images. The RGB channels of color images are processed by corresponding image enhancement algorithms. The global histogram equalization makes the whole gray histogram evenly distributed; the wavelet transform algorithm decomposes color images at multiple levels, and the multi-scale Retinex algorithm is adopted for the multi-scale transformation of images through the Gaussian convolution operation. The experimental results show that the global histogram equalization, the wavelet transform, and the multi-scale Retinex algorithm can enhance the hazy scenes in original images with their own advantages and disadvantages. Among the three algorithms, the multi-scale Retinex algorithm achieves the best dehazing and enhancement because of its enhanced brightness, prominent detail information, and slight distortion.

Abstract:Multi-core processors point out the mainstream direction of processor development, but there are many challenges in hard real-time assurance. By analyzing the real-time requirements and the current application of multi-core processors, we propose an asymmetric multiprocessing program based on general-purpose processors. In this study, we focus on the overall software design, the management of shared resources, and the designs of image loading and starting of slave processors and inter-core communication in the asymmetric multiprocessing mode. The low-voltage protection device for measuring and control is developed by ourprogram, and its on-site operation shows that the program meets the real-time performance requirements of secondary power equipment.

Abstract:The Software Defined Networking (SDN) technology, which has been booming in recent years, solves the prominent problems of IP networks such as layout difficulty and complex updates. In response to SDN-based traffic prediction, the chaos theory is used to reconstruct the phase space of the time series sample group. Then, the Least Squares Support Vector Machine (LSSVM) is introduced to build the SDN-based traffic prediction model, and the key parameters are optimized by the improved Particle Swarm Optimization (PSO) algorithm. The experimental results show that the model effectively improves the accuracy and error control level of SDN-based traffic prediction and is valuable in practical application.

Abstract:How to build a complete early warning rule repository is a key issue in the research on the early warning of university student achievements. In this study, after cleaning and discretizing the data on university student achievements, we use the Apriori algorithm to mine the correlation between failed courses and construct the basic early warning rule repository. On this basis, the influence of courses with “pass” and “good” grades are explored to further expand the early warning rule repository. In the case of copious redundant rules, strong association rules are filtered out by lift and interest in the traditional support-confidence framework to improve the accuracy of the repository and specifically analyze the mined rules. Our methods and conclusions can support the decision-making of teaching management.