Abstract:The accurate forecasting of daily natural gas load is pivotal to the reasonable supply and dispatch of energy in the city. This study proposes a multistage hybrid model based on Fuzzy Coding of Genetic Algorithms (FCGA) and the improved LSTM-BPNN residual correction model since gas load data is periodic but random and the single-stage and single-forecasting model has a limited role. In the first stage, the gas load is forecasted by the LSTM model to calculate its residual value. In the second stage, the residual value is predicted by the BPNN model, and then the learning rate of the LSTM-BPNN residual model is automatically adjusted with the Adam algorithm regarding the adaptive learning rate to accelerate fitting. Afterward, the initial weights and thresholds of the BPNN are optimized by the fuzzy coding of genetic algorithms to find the global optimal solution. Finally, the sum of the forecasting values in the two stages is taken as the final gas load. Comparative experiments prove that the model in this study ensures higher prediction accuracy than the single model and the original two-stage forecasting model.

Abstract:During the operation of the power grid, key devices such as converter valves continue to generate heat, affecting the stability and safety of the system. Then it is crucial to ensure the stable operation of those devices. As a major component in the cooling system, the valve cooling system releases the heat energy from the equipment with water of high thermal conductivity as the medium. The stable operation of the converter valve can be ensured by monitoring the temperature and pressure of the cooling water. Also, with inlet valve temperature in the valve cooling system as the main predictive index, the historical data of the system is fully mined for predicting the operating state of the power grid. An ARIMA-SVM hybrid model integrating the traditional time series model and machine learning is compared with the traditional ARIMA model, the SVM model and the GRU neural network model with regard to the time series analysis of the real valve cooling data from China Southern Power Grid. The comparative experimental results demonstrate that the above four models can all clearly indicate the trend of the inlet valve temperature. However, the ARIMA-SVM hybrid model behaves better in the evaluation indicators including the root mean square error, the mean square error and the mean absolute error than the other three, with a more accurate prediction of inlet valve temperature.

Abstract:In recent years, driven by the progress in artificial intelligence, deep learning models have been widely applied to ECG data analysis (especially the detection of atrial fibrillation). This study proposes an algorithm based on the multi-head attention mechanism to classify atrial fibrillation, which is trained and validated through the public data set of the PhysioNet 2017 Challenge. Firstly, the local features of the ECG signal are extracted through the deep residual network. Then, the bidirectional long short-term memory network is built to extract the global features on this basis. Finally, the multi-head attention mechanism layer is used to extract the key features, and cascade modules greatly improve the performance of the overall model. The experimental results show that the proposed heads-8 model can achieve precision of 0.861, recall of 0.862, F1 score of 0.861, and accuracy of 0.860, which is better than the latest methods based on ECG signals for classifying atrial fibrillation.

Abstract:The diverse and complex trend of public opinion has long made it difficult to manage. Negative public opinion will intensify contradictions, bringing adverse effects to social stability. Then a method of public opinion deduction based on the event knowledge graph is proposed. The causal logic of the event is mined through the neural network, and the event knowledge graph is drawn after causal events are connected. Vectorized event nodes can merge into similar nodes to reduce map redundancy while enhancing map generalization. Besides, the evolution of target public opinion events can be predicted based on the deductive logic indicated in the event knowledge graph. With a public opinion event related to a natural disaster as an example, the experimental results prove that the proposed method can reliably predict the trend of the event, supporting public opinion supervision.

Abstract:The abuse of Unmanned Aerial Vehicles (UAVs) brings great security risks to the low altitude area. Then the research on detection of UAVs’ illegal intrusion has become important for a low-altitude defense system. In this study, a multi-sensor information fusion technique based on radar and a RGB camera is designed to detect small objects in the low altitude range. After that, the Single Shot multibox Detector (SSD) for deep learning is introduced to train the UAV detection model and predict the category and location of objects captured by the RGB camera. An experimental platform is built to verify that the information fusion method can collect the location, speed, appearance of targets, and the deep learning model can determine the categories of suspicious targets.

Abstract:The original U-Net integrates a jumping structure with high-level and low-level image information, which makes the U-Net model perform well in segmentation, but the results still present poor segmentation, over-segmentation, and under-segmentation at the edges of cervical nucleus. Then an improved U-Net network for image segmentation is proposed. First, the densely connected DenseNet is introduced into the encoder of U-Net to solve the problem that the encoder is too simple to extract abstract high-level semantic features. Then different weights are given to the cervical nucleus nuclei and background in the binary cross-entropy loss function, so that the network pays more attention to the learning of nuclear characteristics. Finally, during the pooling operation, reasonable weights are assigned to the pixel values in the pooling domain to avoid losing information in the pooling layer. Experimental results reveal that the improved U-Net network can behave better in cervical cell segmentation with a more robust model, and the proportions of over-segmentation and under-segmentation are also smaller.

Abstract:In light of scientific management and continuous innovation in “intelligent weather projects”, we design an intelligent mobile office system in line with meteorological observatories through the WeChat public platform of “Qingdao Meteorology” (青岛气象 in Chinese) and its background API for secondary development. It aims to strengthen the management and accelerate the response of the meteorological service system, improving the operational efficiency of the meteorological observatory. The system is designed with a logic architecture consisting of layers of storage, technology, applications and users. It is built in a B/S structure and relies on the ASP.NET MVC framework. Besides, it depends on C# language in the background and HTML, Javascript, and CSS languages in the foreground. The main functions include “fault repair order”, “real-time work”, “electronic scheduling”, and “event registration”, initially enabling intelligent meteorological service and office management. Therefore, the system, with sound scalability, is convenient, fast, and practical, deserving popularization.

Abstract:A co-simulation framework of a P2P power trading platform and a distribution network based on Blockchain technology is designed for the active cooperation between them to promote the application of Blockchain technology to power trading, protecting fair energy trading. According to the decentralization, strong security, and traceability of the Blockchain, a distributed double auction mechanism is proposed, which is combined with smart contracts to build an energy trading model for simulation analysis. Finally, the distribution network parameters under two scenarios are compared based with the established co-simulation model. Conclusions are drawn on the influence of the P2P power trading mechanism on the distribution network, providing a theoretical basis and a technical support for efficient energy utilization, safe and stable power trading and the application of block chain technology to power trading.

Abstract:Most of the traditional water data are subject to centralized management, but the rights and interests of the water data are related to water groups, equipment manufacturers, and insurance companies. Then whether the data is tampered during transmission and storage has become a challenge to stakeholders. In view of the above problem, we propose a method for the implementation of intelligent water Blockchain as a Service (BaaS) according to decentralized, unforgettable, and traceable Blockchain technology, avoiding unreliable water data. In addition, we expound the key technologies of BaaS, including the BaaS service architecture, the distributed ledger data model and the Merkle binary tree verification model of health data from intelligent water meters, as well as double buffer queue and Blockchain-based multiple access nodes.

Abstract:Amid the leap in procurement tasks in colleges, a large number of electronic documents have been accumulated. The original one-stop procurement management platform has been unable to manage such a great number of documents. In light of current demand for data management, the system function modules are divided according to the business process throughout the actual life cycle of documents. The object model of the system is built based on the idea of model-driven engineering, and class and sequence diagrams are drawn by rational modeling tools to describe the overall architecture and business logic of the system. Besides, the lightweight flask framework is adopted for research and development, and a document-oriented database (MongoDB) is relied on to tackle the problem of the large concurrent and release the pressure on the data server during reading and writing. For future big data analysis, a PyPDF method is proposed to facilitate the extraction of PDF metadata. As a result, information management is ensured for the final filing during electronic document circulation.

Abstract:In response to the data governance problems about naval aviation in the era of big data, a Spark-based aviation information service platform is designed, enabling data storage, analysis, and mining. The platform has a four-tier architecture, with the Hadoop Distributed File System (HDFS) and Hive for data storage and management. Finally, the performances of the aviation information service platform and the traditional aviation data warehouse are compared through simulation experiments regarding different data volumes. The naval aviation information service platform can serve as a strong data support for naval aviation training and assist users for decision-making.

Abstract:As required by intensive construction of government websites, an integrated information commutation and management solution is proposed, which is based on the micro-service architecture against continuous expansion of websites. It serves for the public base of the platform with unified organization, shared manuscripts, resource identifiers and system interfaces and supports the public data with a complete set of data resources. Thus the whole process management of information is realized, from producing, editing, and processing to network communication and performance statistics. Then an integrated government information commutation platform is formed, with an institutional framework as the basic management unit.

Abstract:In view of the problems in traditional drilling information management systems, we design a drilling information system based on Baidu map to take full advantage of drilling data for research and applications and constantly improve the management of them. The architecture of the system is given, with the corresponding database, and the key processes and algorithms are detailed. The proposed system provides a more real-time, efficient and transparent display of drilling data, making the management process of information more standardized and scientific. The system can effectively realizes the efficient management of drilling data and information.

Abstract:The traditional search engine cannot match the actual information needed by the candidates with searching results when they fill the list of preference in college entrance application, consuming extra energy of them to filter the data, which undoubtedly increase the time cost. We design an intelligent question answering system for academic planning of examinees with the knowledge graph of the college entrance examination, a model for Chinese word segmentation and the Bayesian classification algorithm. Unlike traditional search engines, the artificial intelligence-based question answering system can accurately match the candidates’ questions with search results, reducing the number of repeated searches and data filtering. The test results demonstrate that the system can offer accurate and targeted answers to most of the questions involved in the academic planning.

Abstract:Aiming at the engineering problems such as the blind spots of large and medium vehicles during driving and parking and the difficulty of multi-channel real-time video stitching, we design a safe driving assistance system based on multi-channel cameras as well as FPGA and GPU platforms. FPGA is responsible for preprocessing, including image data acquisition and parameter transfer, and parallel acceleration of stitching algorithms is enabled on GPU. Besides, automatic calibration of multiple cameras is achieved with the optimized algorithm, and a fusion parameter table was generated to acquire accurate image registration and fusion. The experimental results prove that the system can adapt to the real-time stitching of multi-channel fish-eye cameras, and the stable stitching speed on TX2 can reach 33 fps.

Abstract:Partial Transmission Sequence (PTS) is one of the effective methods to suppress the Peak to Average Power Ratio (PAPR) in Orthogonal Frequency Division Multiplexing (OFDM) systems. However, the algorithm needs a full traversal search for the best phase factor, resulting in great computational complexity. Therefore, this study proposes a PAPR suppression algorithm based on Discrete Particle Swarm Optimization (DPSO). Firstly, a new method for determining inertia weight is defined. Then the mutation operator is introduced to improve the original velocity update formula, which enhance the traditional DPSO algorithm that is easy to be premature and difficult to converge to the global optimal. The simulation results reveal that the PAPR performance of the proposed algorithm is better than that of the traditional DPSO algorithm by about 0.3 dB, and the computational complexity is lower than that of the traditional PTS algorithm.

Abstract:Estimation of facial pain expressions is effective for pain assessment. In this study, facial pain is recognized through a feature extraction method integrating block weighted Local Binary Pattern (LBP) and multi-scale partition. First, the pre-processed image is weighted after the histogram is extracted in blocks. Then statistical features of histograms are extracted in multi-scale partitions to concatenate them with different sizes of blocks and cascade the block weighted histograms into the feature vector of the entire image. Finally, the Principal Component Analysis (PCA) is relied on to reduce the dimensionality of the feature vector, and the Support Vector Machine (SVM) is used for classification and recognition. The experiments on a self-built database of pain expression images prove that the proposed method, compared with traditional feature extraction methods and those before fusion, greatly improves the recognition rate of pain expressions. Then it can serve as an effective way for studying and recognizing pain expressions.

Abstract:Second-order optimization can accelerate the training of deep neural networks, but its huge computational cost hinders it from applications. Therefore, many algorithms have been proposed to approximate second-order optimization in recent studies. The K-FAC algorithm can approximate natural gradient, based on which an improved K-FAC algorithm is proposed according to the quasi-Newton method. The K-FAC algorithm is applied to the first few iterations. Then, a rank-one matrix is built, and its inverse matrix is computed by the Sherman-Morrison formula, greatly reducing computational complexity. The experimental results prove that the improved K-FAC algorithm has similar or even better performance than the original K-FAC, especially with much less training time. It also has the advantage over first-order optimization in regard to training time.

Abstract:This study analyzes the solution to the max-cut problem of different vertices according to quantum adiabatic approximation. In this algorithm, the vertices of an undirected graph are equivalent to qubits, the edge between vertices to the coupling between two qubits, and the weight value of an edge to the coupling strength. The algorithm is written in the Python programming language, and the solution to the max-cut problem of a completely undirected graph with 6–13 vertices is simulated. Experimental results demonstrate that when the completely undirected graph has 8, 12, and 13 vertices and coupling strength is 1.0, the expected value of Hamiltonian in the max-cut problem does not converge. Then the coupling strength between qubits is adjusted to observe the changes in the expected value. Experiments reveal that for a completely undirected graph with 12 vertices, the expected value converges when coupling strength is 0.95. For completely undirected graphs with 8 and 13 vertices, it converges with time when coupling strength is 0.75. Accordingly, it is inferred that the coupling strength between qubits can be normalized to about 0.75 when the quantum adiabatic algorithm is used to solve the max-cut problem for a completely undirected graph with more than 13 vertices, so that the expected value can eventually converge.

Abstract:In the process of privacy preserving high-dimensional data publishing, the size of the differential privacy budget directly affects the addition of noise. The privacy budget cannot be allocated reasonably for independent low-dimensional attribute sets, compromising the security and restricting availability of composite data sets. Then a Personalized Privacy Budget Allocation (PPBA) algorithm is proposed. The maximum support tree and weight values of attribute nodes are introduced to reduce the candidate space of attribute relationship pairs selected by the differential privacy index mechanism and enhance the accuracy of the Bayesian network. The dynamic weight values of nodes in the Bayesian network are set to rank the sensitivity of low-dimensional attribute sets. According to the personalized requirements for security and availability of published data sets, the constant allocation ratio q of differential privacy budgets is customized for the personalized allocation of Laplace noise to the low-dimensional attribute sets sorted by sensitivity. Theoretical analysis and experimental results reveal that the PPBA algorithm can meet the personalized requirements for security and availability of high-dimensional data publishing, with lower time complexity.

Abstract:This study integrates a knowledge graph into a model for learning resource recommendation considering the logical relation between knowledge points, aiming to address the “cognitive overload” and “learning trek” in online learning and meet the users’ personalized learning needs. Firstly, a knowledge graph, a learning resource model, and a user-oriented mathematical model are developed. Then, we establish a multi-objective optimization model by taking into account the user preference and the correlation between the users’ knowledge base and the knowledge points covered by the learning resources. After that, this model is solved by the Adaptive Multi-Objective Particle Swarm Optimization (AMOPSO). Furthermore, we reduce the size of the external population through sorting the individual crowding distance in a descending order, thus obtaining the two-object Pareto frontier with optimal distribution and the recommended resource sequence. The proposed algorithm is also compared with the standard multi-objective particle swarm optimization and evaluated by HV and IGD, demonstrating its robust diversity, stability, global optimization, and convergence. Finally, five-fold cross-validation verifies the recommendation from the proposed algorithm.

Abstract:Since the features obtained from a single action mode fail to accurately express complex human actions, this study proposes a recognition algorithm for human actions based on multi-modal feature learning. First, two channels extract the RGB and 3D skeletal features from the action video. The first channel, i.e., the C3DP-LA network, consists of an improved 3D CNN with Spatial Temporal Pyramid Pooling (STPP) and LSTM based on spatial-temporal attention. The second channel is the Spatial-Temporal Graph Convolutional Network (ST-GCN). Then the two extracted features are fused and classified by Softmax. Furthermore, the proposed algorithm is verified on the public data sets UCF101 and NTU RGB+D. The results show that this algorithm has higher recognition accuracy than its counterparts.

Abstract:This study proposes an Improved Particle Swarm Optimization with Genetic OPerators (IPSO-GOP) to determine the optimal trajectory of mobile robots in a complex environment. Firstly, we improve the Particle Swarm Optimization (PSO) and adaptively adjust the inertia weight during the algorithm operation to facilitate the particle search. Besides, we disturb the particles with the chaotic variables to increase the convergence speed. Secondly, we introduce the Genetic OPerators (GOP), i.e., multi-crossover and mutation inherited by the genetic algorithm, to optimize the improved PSO (IPSO), thus getting rid of the local minimum and promoting the population diversity. Finally, the shortest continuous geometric path without collisions is obtained after cubic spline interpolation smooths the path generated by the proposed algorithm. In addition, the proposed algorithm in a multi-obstacle environment circumvents the local optimum and accelerates the convergence. Compared with the PSO, it has significant optimization and advantages in path planning.

Abstract:This study proposes an improved collaborative filtering recommendation algorithm integrating expert trust aiming at the data sparsity and cold start in the current algorithms. This algorithm divides users into different community clusters based on the optimization of initial clustering centers in DBSCAN. Considering the influence of user activity on similarity calculation, we introduce the penalty weight of user activity to improve the similarity calculation. After expert selection, the balance factors in projects are introduced, since the expert trust for different projects varies. Thus, each project evaluated has an independent expert trust. Experimental results on the MovieLens data set show that the proposed algorithm can effectively alleviate data sparsity and cold start, increasing the recommendation accuracy.

Abstract:Multiple technology projects in a group are evaluated by several experts. Each project covers certain fields and each expert has his/her own advantageous fields. Thus, it is a significant challenge to scientifically and automatically group experts in suitable fields of relevant projects from a large number of candidates. This study proposes a multi-match model GIS of evaluation experts for technology projects based on the greedy algorithm. This model is applied to the two associated “project-field” and “expert-field” correlation matrices. Specifically, it separately calculates the discrete distribution of projects and experts in each field. Then it uses a proper evaluation function to measure the project-expert match and finally obtains the optimal team. The experiments based on the data sets in the power industry show that this model can match the experts with the technology projects and thus has high rationality and accuracy. It avoids the careless mistakes and unfairness in traditional expert selection while reducing human costs.

Abstract:Similarity matching is crucial for natural language processing and also for extracting answers from the question answering system. This study proposes a model of text similarity matching based on positive and negative samples and Bi-LSTM. Firstly, this model constructs question answering pairs for positive and negative samples in model training, improving the similarity between a question and its correct answer. Secondly, it applies the dual-layer word vector embedding for pre-training to solve the experimental error caused by segmentation mistakes. Thirdly, it adopts the internal attention mechanism before feature extraction to solve the backward offset of the characteristic vectors caused by the attention mechanism. Then this model trains the data on the Bi-LSTM neural network to retain important temporal characteristics. Finally, it puts forward a similarity calculation function including semantic information to calculate similarity at the semantic level. The model proposed in this study is simulated on the public data set DuReader and compared with other models. The experimental results show that the proposed model has high accuracy and good robustness, and the accuracy of top-1 reaches 78.34%.

Abstract:This study proposes a tool for analyzing the tsunami outbreaks in coastal cities based on a 3D particle method of Smoothed Particle Hydrodynamics (SPH). Firstly, we obtain the 3D location information (SHP) and Digital Elevation Model (DEM) through the Geographic Information System (GIS), thus display the topography, altitude, and the outside appearance of the buildings. Then, the STL data defined by the surface profile are converted to particle data. Finally, the incompressible condition in the SPH method is alleviated, and the boundary problem is solved by virtual marking. The 3D simulations show that the tsunami particles do not penetrate into the buildings, and there is no gap between the buildings and the ground, which verifies the good results of the proposed method.

Abstract:This study mainly optimizes the structure and parameters of the LSTM model, so that the accuracy of predicting the trend in stock prices is significantly improved. Besides, we investigate the weekly and daily data on US stock in terms of predicting the LSTM neural network. On one hand, we compare the difference to verify the impact of different data sets on the forecast. On the other hand, we provide selection suggestions for data sets so as to increase the accuracy of stock forecast. This study uses the multi-time-series stock forecast in the improved LSTM model to predict the trend in stock prices. The results demonstrate that the weekly data perform better in forecast than daily data. To be specific, the average accuracy of daily data and weekly data is 52.8% and 58%, respectively. In summary, the application of weekly data to training the LSTM model yields higher accuracy in stock forecast.

Abstract:As a substrate protocol for real-time multimedia application, DCCP is featured by congestion control and unreliable transmission. However, the congestion control algorithm CCID2 in DCCP is still based on AIMD, which will cause bufferbloat, longer network delay, and jitters. Hence, such a Loss-Base model is no longer suitable for the high-BDP environment. In contrast, the BBR algorithm can effectively control the network delay, minimize the network queuing, and maintain high bandwidth utilization and low link delay even at a high packet loss rate. Therefore, it is suitable for the real-time multimedia applications with DCCP. This study adds a detection model for packet loss rates to the BBR algorithm after its introduction to DCCP and applies the congestion control algorithm in the model of delay and bandwidth product to addressing the above-mentioned problems. Compared with CCID2, the proposed algorithm reduces the average delay by 20% under heavy loads and can produce a large throughput at a high packet loss rate.

Abstract:Aiming at the problems faced by college student management in the context of educational big data, this study proposes an academic early warning algorithm for college students. It mines potential education data with the results of digital campus construction in colleges and universities. Eight characteristic data with higher correlation coefficients selected by the Kendall correlation analysis are taken as the input for the BP neural network, and the relevant results are applied to improving the GA-BP algorithm. Thus, the academic situation is predicted by taking into account various factors. The tests demonstrate that the prediction accuracy of the proposed algorithm can reach more than 90%.

Abstract:This study proposes a strategy based on an improved harmony search algorithm to improve the success rate and quality of intelligent test paper generation in English tests. The objective optimization function of intelligent test paper generation is established and then solved by the harmony search algorithm. After that, the harmony search algorithm is improved. Finally, an example for intelligent test paper generation is analyzed. The results indicate that the improved algorithm achieves a high success rate and good quality of intelligent test paper generation. Additionally, the proposed strategy outperforms its counterparts.

Abstract:This study optimizes the multi-pattern matching algorithm for the intrusion detection system in the computer network, so that the system can be in operation in the high-speed environment. First, we comprehensively analyze the algorithm and principle for network intrusion detection. Second, we elaborate the idea of optimizing the multi-pattern matching algorithm for its implementation, so that the algorithm efficiency is increased and then the detection system is improved. In summary, the optimized algorithm can enhance the network detection system.

Abstract:This study constructs a Distributed Denial-of-Service (DDoS) attack detection model based on Particle Swarm Optimization-Convolutional Neural Network (PSO-CNN). First, it uses the weight sharing and maximum pooling of CNN to automatically mine the features of data streams. Then, it applies PSO to the convolution kernel, thus increasing the training efficiency and enhancing the global optimization. In conclusion, the model proposed in this study has high detection accuracy for DDoS attacks.

Abstract:Loan risk analysis is a common test faced by global financial institutions. In the context of big data, it is of practical significance to prevent loan risks through machine learning algorithms. Aiming at the imbalance in loan data and high noise, this study uses the Boruta feature selection algorithm to sort the importance of loan data. In addition, it proposes the CatBoost integrated learning algorithm based on Comprehensive Learning Particle Swarm Optimization (CLPSO-CatBoost) for loan risk prediction. This algorithm improves the global search and avoids the local optimum. Compared with the traditional credit evaluation models, CLPSO-CatBoost has high accuracy.

Abstract:Face biometrics are unique, natural, and invariant throughout one’s life cycle. Therefore, face recognition, as an identity authentication method, is much more convenient than traditional techniques. However, it is susceptible to forgery attacks such as pictures, videos, and masks. The leakage or tampering of invariant biometrics will induce irreversible losses. The solution to secure face recognition proposed in this study uses a special module of secure face acquisition for liveness detection. A special secret key in the module is applied to encrypting every face image and signing the sensitive information. Additionally, multiple steps in the recognition process guarantees the security of face biometrics.

Abstract:With the fast development of information technology and the consequent surge in the unstructured text and audio data, traditional manual ways of processing the cases are not suitable for practical applications, which has posed great challenges to the public security organs in case investigation. Thus, this study devises the artificial intelligence-based natural language processing technology to extract and analyze the characteristic information such as reports to the police, brief cases, and records of inquiries from the information system of cases of encroachment, telecom fraud, and gang. In this way, unstructured texts can be mined and analyzed, further supporting the judgment by investigation departments and intelligence departments. Moreover, spatio-temporal information, trajectories of the crime, and the characteristics of tools and means are compared. In this way, the high-risk suspects can be found and actively recommended, greatly reducing the scope of investigation and improving the efficiency of detection.

Abstract:Aiming at the low efficiency in repairing failed nodes of fractional repetition codes, this study proposed a construction algorithm of Fractional Repetition codes based on Spanning trees of Harary graph (FRSH). As a result, FRSH codes have lower computational overhead in repairing bandwidth overhead and locality than RS codes and SRC. Besides, FRSH codes are more efficient and spend less time in repairing failed nodes, compared with the other two codes.

Abstract:The deep neural network can better express features but results in difficult optimization, high training cost, and vanishing gradient. The surge in quantity of parameters leads to a too bloated model to be deployed on the platform with weak computing power and small storage, such as mobile terminal and industrial control equipment. Aiming at these problems, we construct a lightweight neural network combining atrous convolutions and multi-scale sparse structures to extract the features of images, and realize the end-to-end recognition for the captcha images with color pattern noise and seriously touched and distorted characters. The dataset containing one million images was divided into training sets, validation sets, and test sets in the ratio of 98:1:1 and trained in batches. Consequently, the lightweight neural network has a recognition rate of 98.9% on test sets with much fewer parameters.

Abstract:As a kind of cognitive radio network based on cooperative communication technology, cognitive cooperative network not only solves the problem of uneven distribution of spectrum resources in different networks in specific time and space, but also improves network performance through cooperation between primary and secondary users. However, with the increasing variety of wireless technologies and mobile applications, due to the inherent characteristics of wireless channels can lead to the loss of and fluctuation in performance of service quality, due to safety problems. To deal with the descending transmission of the cognitive cooperative network attacked by malicious users, we put forward a Blockchain-based safe transmission strategy. First, Blockchain is used for authentication, avoiding the loss of and fluctuation in transmission performance caused by data corruption. Second, Reed-Solomon (RS) codes are adopted for better error correction, further enhancing the stability. Eventually, the simulation result shows that the proposed strategy is more stable than existing transmission modes of the cognitive cooperative network.

Abstract:An adaptive learning model of English vocabulary is developed, which contains a machine learning algorithm. The model records learners’ self-selection of what they learn to reflect individual differences. The key parameter of such a learning tool of dynamic modeling is conditional probability that measures the adaptive relationship between a cognitive feature and certain learning content. Therefore, this parameter is called adaptability. It is updated every time a learner self-selects the learning contents about a word, which is regarded as a time of training. The adaptability is constantly adjusted to modify and maintain the model through training. The model abstracts the problem to be solved, according to the adaptive test process based on the item response theory, into mathematical formulas with our reference to those in the AdaBoost algorithm. This model can continue to iterate the adaptability until it is stable and recommends proper learning contents for users. This paper first reviews relevant literature and talks about the value of this topic, then expounds on the theoretical basis, and focuses on the construction of the model with case study at last.

Abstract:The learning ability of traditional models is reduced by copious constrained samples, so an Internet user classification model based on improved Support Vector Machine (SVM) is designed, which simulates the browsing trajectories of Internet users by constructing sample data. A brand-new user classification strategy according to user preferences is formulated. Then, Internet users are classified based on improved SVM. According to the three performance tests, the model has satisfying classification ability because its average accuracy is 98.56%, higher than the expected value. Seen from the comparative tests with two traditional user classification models, this model can maintain a high level of learning ability in the face of increasing sample data.

Abstract:The accuracy of crowd density estimation is low in complex backgrounds and the scenario with dense and mutually occluded crowds. To solve this, we propose a method based on YOLOv3 enhanced model fusion to estimate crowd density. The heads and bodies in the data set are labeled to generate head and body sets, which can then help train the two YOLOv3 enhanced models: YOLO-body and YOLO-head. Finally, the two models are reasoned on the same test data set, and their outputs are fused to the maximum value. Consequently, the method based on YOLOv3 enhanced model fusion has great robustness because its accuracy is 4% higher than that of original target detection and density map regression.

Abstract:The bolt-fastening detection of transmission towers is critical to the safety of high-voltage power grids. Traditional detection methods are often risky it needs manual detection high on transmission towers. What’s more, UAV detection fails to live up to our expectation affected by multiple external factors. Therefore, this study proposes a bolt-fastening detection method for transmission towers based on Gated Recurrent Unit (GRU) networks. Specifically, the vibration sensor and sensor analyzer are used to construct a work flow for collecting acoustic wave data of transmission towers, and then the Linear Predictive Cepstral Coefficients (LPCCs) of acoustic wave data in training samples are extracted to form feature vectors. The classification model of GRU networks is trained to predict unknown fastened acoustic wave samples. As a result, this method is practical. The application of this algorithm can avoid the much manpower of traditional ones and is superior to them in bolt-fastening detection of transmission towers.

Abstract:With the advancement in modern power grid dispatching and control systems, many human–machine cloud terminals need to be equipped with applications, but the present demand is hardly met because the application management is facing such problems as low automation in the release, upgrade, and maintenance of applications, the lack of information feedback loops after software deployment, and high cost of later maintenance. In addition, cross-regional distribution is another problem in transmission due to the partition of the network in the power grid system. User experience is largely worsened by the tardy uploading and downloading of large files. To solve this problem, this study proposes a design scheme for the application store of power grid dispatching and control systems, which is mainly used for more automatic release, upgrade, and maintenance of applications as well as faster and smoother network transmission in application distribution. In this way, the quality and usability of applications are improved. Besides, information such as user behavior can be analyzed to make recommendations. The application store is capable of both life cycle management and tracking analysis of applications.