Computer Engineering & Science

A cluster job execution time prediction model based on LSTM

ZHU Zheng-dong, WU Yin-chao, HU Ya-hong, JIANG Jia-qiang

2022, 44(08): 1331-1341. doi:

Abstract ( 213 )

PDF (1466KB) ( 208 ) 　　

To improve the quality of service (QoS), data centers need to ensure that user jobs can be completed within a specified deadline, so jobs must be efficiently scheduled based on real-time system resources. A job scheduling algorithm based on a LSTM (Long Short-Term Memory)-based job execution time prediction model is proposed to minimize the job completion time. The LSTM-based time prediction model predicts the execution time of user jobs according to the type of user jobs, the amount of jobs, the number of CPU cores and memory required by the jobs, and the ratio of the resources required by the jobs to the total system resources. The prediction results are used to judge whether the cluster is capable of completing user jobs on time, and provide a basis for rationally arranging the execution order of the jobs. The hyperparameters that affect the performance of the LSTM time prediction model, such as the number of iterations, the learning rate and the number of network layers, are determined through experiments. Experiments show that compared with the SVR model, ARIMA model and BP model, the job execution time prediction model based on LSTM improves the determination coefficient R2 by 297%, 2.34% and 5.66% respectively, and the average error of its prediction is only 0.78%.

An arctangent operator based on the Piecewise algorithm

LONG Ke-li, WANG Dong, CHEN Hu, LI Xu-jun

2022, 44(08): 1342-1348. doi:

Abstract ( 116 )

PDF (687KB) ( 138 ) 　　

The arctangent operation is one of the common operations in applications such as scientific computing, digital signal processing, and graph processing. Based on the piecewise algorithm, a single-precision floating-point arctangent arithmetic unit conforming to the IEEE-754 standard is designed for a certain type of DSP chip. To achieve the accuracy required by the design, all errors in the calculation process are analyzed and limited. To make the monotonicity of the output result consistent with the original function, a design method that can guarantee the monotonicity of the algorithm is proposed. To reduce hardware cost, a two-level hierarchical segmentation method is used, and the signal bit width design and optimization based on circuit static and dynamic analysis are realized. The FPGA-based simulation results show that the hardware cost of the arctangent operator is relatively small, the error between the output and the standard result is less than 1 ulp, and the operator has excellent output monotonicity.

Research on fault detection strategy of redundant flight control computer based on ARINC659 bus

DENG Ze-yong, CAO Dong, SHAO Hai-long

2022, 44(08): 1349-1356. doi:

Abstract ( 92 )

PDF (779KB) ( 116 ) 　　

This paper studies the fault detection strategy of the redundant flight control computer based on ARINC659 bus. The system architecture of the flight control computer is introduced, and the possible software and hardware fault types and their influences in the flight control computer system are analyzed. Aiming at the Three effective fault detection methods are proposed, and the specific implementation steps of each detection method are designed. In terms of detection reliability, a failure probability model is established to quantitatively analyze the detection accuracy of each failure mode after combining three failure detection methods. Finally, a semi-physical simulation platform is built to verify the rapidity and effectiveness of the detection method, which meet the reliability requirements.

Deep learning-based SAR target recognition on DSP

HE Tao, SHI Hui-li, LI Da-liang

2022, 44(08): 1357-1363. doi:

Abstract ( 93 )

PDF (1129KB) ( 107 ) 　　

SAR image target recognition mainly aims at strategic military targets such as bridges and airports, as well as tactical targets such as aircraft, tanks, and automobiles. Accurate identification, classification and positioning is an important part of SAR image interpretation. Firstly, the main processing layer of the convolutional neural network based on C6678 is constructed. Secondly, the processing and storage characteristics of C6678 are combined to optimize the design of the convolutional layer and network scheduling. Finally, a design and implementation method of the YOLOv3-TINY target recognition network on C6678 is completed. This method can reconstruct and modify common convolutional neural network models, and solves the problem of running deep learning networks on multi-core processing platforms such as C6678. The experimental results show that the method is consistent with GPU in detection performance. Considering the real-time image frame rate of airborne SAR, although the real-time performance of this method on C6678 is far from that of GPU, it can meet the real-time processing requirements of airborne SAR.

A military training recommendation model based on Cross-DeepFM

GAO Yong-qiang, ZHANG Zhi-ming, WANG Yu-tao

2022, 44(08): 1364-1371. doi:

Abstract ( 105 )

PDF (714KB) ( 86 ) 　　

In order to apply the recommendation system to the field of military training and give full play to the value of military training big data in personalized training, a hybrid recommendation model called Cross-DeepFM is proposed. Firstly, the real military training data are collected and preprocessed, and the custom military training dataset is constructed. Then, the structure of the Cross-DeepFM model is designed by combining the deep residual neural network, deep cross network and factorization machine, and the details of the model are analyzed. Finally, the comparison and analysis are carried out on the custom military training data set. The experimental results show that the proposed model is more accurate than the mainstream recommendation model and can effectively complete the personalized recommendation task of military training.

A security index system of security risk assessment behavior based on STAMP model

WANG Ke-ke, GUO Li-li, LANG Jing-hong

2022, 44(08): 1372-1381. doi:

Abstract ( 135 )

PDF (1440KB) ( 114 ) 　　

The existing security risk assessment methods and models do not fully consider the impact of the risk assessment behavior itself on the assessment results, which is a big lack of understanding that the behavior of risk assessment may introduce security risk. In response to this problem, this paper first establishes a complete STAMP model of risk assessment behavior. On this basis, the STPA analysis method is used to conduct security analysis on risk assessment behavior, the STAMP theory is used to construct a risk assessment behavior security index system, and the improved AHP method is used to screen important index factors in the security index system. The proposed security index system focuses on the emergence of the system as a whole rather than the reliability of individual components. According to the reasons for the occurrence or danger of system safety accidents, it provides a more effective way of constructing a safety index system.

A threshold multi-secret sharing scheme with identity lock

CUI Chen-yu, ZHANG Li-na,

2022, 44(08): 1382-1391. doi:

Abstract ( 101 )

PDF (670KB) ( 121 ) 　　

In order to avoid the disadvantage that different secrets has the same access control structure in the existing secret sharing schemes, a threshold multi-secret sharing scheme based on identity lock is proposed, which determines the authorized subset of the secret. Only the user in the authorized subset can recover the secret. There are different identity locks for different secrets.H3： preset that the system has a secure channel to transmit secret shares,Under the premise of keeping the sub-secrets reusable and detectable for deception, it does not increase the information interaction of any participants, and effectively solves the problem that the access control structure of different secrets is difficult to change. At the same time, based on the session key negotiation algorithm, the scheme does not need to use a secure channel to transmit the secret share in advance, so it has better security and practicability. The scheme is suitable for the scenarios of multi-secret threshold sharing based on identity access control, such as video conference and file distribution.

An active magnetron memristor hyperchaotic circuit and image encryption

ZHANG Jie, XU Long-hao, YIN Bao-quan,

2022, 44(08): 1392-1401. doi:

Abstract ( 98 )

PDF (3608KB) ( 117 ) 　　

On the basis of a generalized third-order Rucklidge system, a fourth-order hyperchaotic circuit system based on an quadratic nonlinear active magnetron memristor is proposed. The dynamics characteristics of the system, such as system phase diagram, Lyapunov exponential spectrum, bifurcation diagram, and coexisting attractor are studied by numerical simulation. The analysis results show that the system has rich dynamics characteristics and new topological structure. The system's circuit is designed and implemented by using Multisim circuit simulation software and FPGA digital hardware circuit. The experimental consequences are very consistent with the numerical simulation results. Finally, an image encryption algorithm is designed by combining scrambling-diffusion algorithm with DNA encryption. The new system chaotic sequence is used to encrypt image, and the encryption histogram, correlation between adjacent pixels, sensitivity, robustness and information entropy are analyzed. The results show that the new system is very sensitive to both chaotic key and plaintext, and the key space is large. The proposed chaotic system has high security performance when applied to image encryption.

A Stacking ensemble clustering algorithm based on differential privacy protection

LI Shuai, CHANG Jin-cai, LI-L Mu-zhi, CAI Kun-jie,

2022, 44(08): 1402-1408. doi:

Abstract ( 94 )

PDF (870KB) ( 125 ) 　　

Aiming at the problem that the accuracy and security of the single clustering algorithm under differential privacy protection are insufficient, a stacking ensemble clustering algorithm based on differential privacy protection is proposed. Stacking is used to integrate a variety of heterogeneous clustering algorithms. K-means clustering, birch hierarchical clustering, spectral clustering and gaussian mixture clustering are used as primary clustering algorithms. By combining the contour coefficient, the clustering results generated by the primary clustering algorithms are weighted into the original data. K-means algorithm is used as the secondary clustering algorithm to cluster the expanded data set. According to the clustering results of the original data and the primary clustering algorithms, adaptive ε functions are proposed to determine the privacy budget, and different degrees of Laplace noise are allocated to the data with different sensitivities. Theoretical analysis and experimental results show that, compared with the single clustering algorithm, the proposed algorithm can effectively improve the clustering accuracy while satisfying the ε-differential privacy protection, and achieve a high balance between privacy protection and data availability.

A new 3D chaotic circuit design and its synchronous control

YAN Shao-hui, SHI Wan-lin, SONG Zhen-long, WAN Er-tong, SUN Xi, HUANG Yi-bo,

2022, 44(08): 1409-1417. doi:

Abstract ( 131 )

PDF (1464KB) ( 105 ) 　　

A new three-dimensional chaotic system is constructed on the basis of Bao system. Through theoretical analysis and numerical simulation, the dynamic characteristics of the chaotic system are studied, such as system dissipation and equilibrium stability, Lyapunov exponent spectrum and bifurcation diagram, Poincare cross section and 0-1 test motion trajectory, etc. The results show that it has the abundant chaotic dynamics characteristics. Through spectral entropy complexity (SE) and C0 complexity, the complexity of the system under different parameters is analyzed to find the parameter value range with the highest complexity. Finally, the non-linear feedback synchronization method and the linear feedback control method are used to realize the synchronization of the chaotic system. To determine the parameter range of linear synchronization control according to the maximum Lis exponent of the system, and the Multisim software is used for circuit simulation. The simulation results are completely consistent with the numerical analysis, which verifies the feasibility of the new three-dimensional chaotic circuit synchronization control, and provides an experimental basis for the application of the chaotic system in the field of synchronous secure communication.

A lightweight instrument dial detection algorithm based on SSD algorithm

ZHANG Jian-wei, ZHOU Ya-tong, SHI Bao-jun, HE Hao, WANG Wen

2022, 44(08): 1418-1425. doi:

Abstract ( 65 )

PDF (1310KB) ( 83 ) 　　

There are problems such as cumbersome process, long processing time, and poor detection effect when the traditional image recognition algorithm is used to recognize the numbers in the instrument dial. Aiming at the above problems, a lightweight instrument dial detection method based on deep learning is proposed. Based on a single-shot multi-scale detection method (SSD), the proposal uses deep separable convolution instead of standard convolution to design feature extraction networks to improve feature expression capabilities and lightweight performance. At the same time, an anchor boxes construction process based on the distribution of ground truth boxes is proposed. In the process, an index that can quantify the matching degree of anchor frames (matching rate) is designed, and an anchor box scheme with a higher matching rate and a smaller number is constructed. Experimental results show that the proposed algorithm has less model parameters and computation, higher detection accuracy, and can obtain real-time processing speed in CPU environment.

Facial expression recognition of different age groups based on face sub-region weighting

YU Su-xin, HE Jun-ji

2022, 44(08): 1426-1432. doi:

Abstract ( 62 )

PDF (788KB) ( 91 ) 　　

Different sub-regions in a face image contribute differently to human face expression recognition, and meanwhile one sub-region contributes differently to expression recognition for people of different ages, such as the old and the middle-aged, the young and children. Therefore, the best recognition rate may not be achieved if a fixed sub-region weighting mode is used for facial expression recognition. To improve the recognition rate, an expression recognition method with variable weight is proposed. Firstly, expression databases for old and middle-aged people, young people and children are established respectively. Secondly, pure face region is segmented from the face image. The regions of eyes and mouth are picked up further. The features of these regions are extracted, weighted and fused. By setting different weights, their effect on ex-pression recognition of different types of people is studied. The experimental results show that the facial expression recognition method using variable weighting value has significantly higher recognition rate than the method using fixed weighting value. For images of the middle-aged and old, the young, and the children, the expression recognition rate is improved by 8.6%, 4.8%, and 1.4%, respectively.

Development of a CFETR neutral beam injection experimental data plotting and analyzing system

YANG Zi-yan, HU Chun-dong, ZHAO Yuan-zhe

2022, 44(08): 1433-1439. doi:

Abstract ( 42 )

PDF (973KB) ( 63 ) 　　

During the experiment of the neutral beam injection of the RF negative ion source, megabytes of data such as experimental signals and log files are generated. The new system is designed in order to solve the problems existing in the current system, such as low storage efficiency of experimental data, no support for real-time drawing, and drawing of experi-mental data of 3600-second long pulse in the future, a new system is developed. And it uses MDSplus, which is commonly used in the field of international fusion, as a data management method to achieve segment-ed management and efficient storage. It uses the Winform framework to realize the design of the drawing soft-ware interface, and cooperates with socket, multi-threading and other technologies to realize various drawing functions. After extensive testing, the compression rate of MDSplus data files can reach 16%. The plotting soft-ware runs stably on the experimental platform and has comprehensive functions. The average time required for real-time plotting, resampling plotting, and contrast plotting is within 1000 milliseconds. The results show that the system can realize real-time plotting and analysis of a large amount of experimental data, and is suitable for most application scenarios that require analysis of a large amount of floating-point data.

A mask detection method based on multi-scale optimized awareness network

GOU Song, ZHAO Xu-yan, HOU Song, LI Wei

2022, 44(08): 1440-1448. doi:

Abstract ( 94 )

PDF (3101KB) ( 120 ) 　　

Wearing a mask is recognized by global medical experts as one of the most effective ways to prevent COVID-19 infection. The vision-based intelligent mask detection plays an important role in urging people to wear masks in public. However, compared with general object detection, there are currently few studies focusing on mask detection. To solve the problem, an optimized multi-scale awareness network, called PyramidMask, is proposed for mask detection. Firstly, PyramidMask obtains the multi-layer features of the image from different scales of the backbone. Secondly, the scale-awareness branches are designed to perform independent predictions of different layers of high-density candidate boxes. Finally, the multi-scale faces with masks in an image is accurately detected in an end-to-end manner. In addition, in order to improve the robustness of PyramidMask under complex scenes, the training samples are augmented by image stitching in the training stage. The experimental results show that PyramidMask outperforms the state-of-the-art methods on the public mask detection dataset. Compared with the benchmark, PyramidMask improves 5.4% and 12.5% in the recall of detection with and without masks, and 6.0% and 4.1% in the precision of detection with and without masks.

A complex pedestrian detection model based on improved YOLOv4 algorithm

LI Lan, LIU Jie, ZHANG Jie

2022, 44(08): 1449-1456. doi:

Abstract ( 165 )

PDF (1046KB) ( 157 ) 　　

When pedestrian detection is carried out in image and video sequences, there are some problems, such as diverse pedestrian posture and scale as well as pedestrian occlusion, which leads to inaccurate detection of some pedestrians by YOLOv4 algorithm and false detection and missed detection. In order to solve this problem, a complex pedestrian detection model based on improved YOLOv4 algorithm is proposed. Firstly, the ground truth size of pedestrian dataset is analyzed by the improved k-means clustering algorithm, and the size of anchor box is determined according to the clustering results. Secondly, PANet is used for multi-scale feature fusion to make it more sensitive to multi-attitude and multi-scale pedestrian targets and improve the detection effect. Finally, for pedestrian occlusion problem, the repulsion loss function is proposed to make the predicted box as close to the correct target as possible. The experimental results show that the new de-tection model has better detection effect than YOLOv4 and other pedestrian detection model.

Echo state networks with improved particle swarm optimization algorithm for electricity demand forecasting

WANG Lin, WANG Yan-li, AN Ze-yuan

2022, 44(08): 1457-1466. doi:

Abstract ( 107 )

PDF (1339KB) ( 108 ) 　　

Firstly, an adaptive operator is introduced to improve the inertial weights and learning factors of the standard particle swarm optimization algorithm (PSO) to improve the balance between the exploration of the current space and the unknown space. At the same time, a nonlinear function is used to construct the nonlinear relationship between the internal states of the Echo State Networks (ESN). Then the improved PSO (APSO) is used to optimize the key parameters of Nonlinear ESN (NESN) to propose an assembled prediction model named APSO-NESN. Finally, the model is used to solve electricity demand forecasting problems. The experimental results show that the APSO-NESN model has higher prediction accuracy than the ARIMA, MLR, standard ESN, and other models.

Text feature selection based on sine and cosine algorithm

WEN Wu, WAN Yu-hui, WEN Zhi-yun,

2022, 44(08): 1467-1473. doi:

Abstract ( 69 )

PDF (638KB) ( 92 ) 　　

In order to obtain a better feature subset in the text and eliminate interference and redundant features, a hybrid feature optimization algorithm combining filtering and swarm intelligence algorithm is proposed. Firstly, the information gain value of each feature word is calculated, the better feature is selected as the preselected feature set, and then the sine cosine algorithm is used to optimize the preselected feature to obtain the selected feature set. In order to better balance the global search and local development capabilities in the sine-cosine algorithm, adaptive inertia weights are added. To more accurately evaluate feature subsets, a fitness function weighted by the number of features and accuracy is introduced, and a new location update mechanism is proposed. Experiment results on KNN and Bayesian classifier show that this feature selection model improves the classification accuracy, compared with other feature selection methods and the model before improvement.

A two-way homologous coherent microwave optical fiber stable phase transmission system

JI Xian, LIU Peng, ZUO Peng-sha, QI Shuai, KANG Hai-long

2022, 44(08): 1474-1480. doi:

Abstract ( 77 )

PDF (915KB) ( 85 ) 　　

In many fields, such as electronic countermeasure system, radar system, satellite navigation and deep space exploration, microwave signal is modulated by laser and loaded on the optical signal. After transmitted by optical fiber, its phase is affected by the external environment, so it can not achieve stable phase transmission. In this paper, a two-way homologous coherent microwave optical fiber stable phase transmission system is proposed, which can realize microwave signal's stable phase transmission in optical fiber. In this technique, the constant temperature crystal oscillator is used to generate multi-channel reference signals with stable phase. One signal is used as the phase change identification signal for optical fiber transmission. The signal is modulated into an optical signal by the laser, and then the multi-channel optical reference signal is separated by the optical splitter and the signal to be stabilized is transmitted with the optical fiber. In addition, the same reference signal is used as the reference signal to identify the phase with the phase identification signal transmitted through the optical fiber, and the reference signal and the reference signal are coherent. Two kinds of signal coherence mechanisms are constructed to realize the phase identification system with the phase change as the parameter. The phase change of the reference signal is used to identify the phase change of the transmitted broadband RF microwave signal. In the realization process of phase detection, a double balanced mixer is used to complete the real-time phase detection of the reference signal phase change. The MCU samples the real-time change of the reference signal phase, and controls the adjustable electric delay line (VODL) in real time to track and compensate the optical path difference of the optical transmission link, so as to complete the active phase compensation of the required stable phase signal and the reference signal. It can realize the stable phase transmission of wideband RF microwave signal through optical fiber.

Neural machine translation based on dictionary model fusion

WANG Xu, JIA Hao, JI Bai-jun, DUAN Xiang-yu

2022, 44(08): 1481-1487. doi:

Abstract ( 83 )

PDF (643KB) ( 85 ) 　　

Unsupervised neural machine translation can train models using only a large amount of monolingual data without the need of parallel data, but it is difficult to establish the connection between two linguistically distant languages. To address this problem, this paper proposes a new neural machine translation training method without parallel sentence pairs. A bilingual dictionary is used to replace words in monolingual data, so as to establish the connection between the two languages. Meanwhile, word embedding fusion initialization and dual-encoder fusion training are used to enhance the alignment of the two languages in the same semantic space, in order to improve the performance of the machine translation system. Experiments show that, compared with other unsupervised models, our method can improve the BLEU values by 2.39 and 1.29 over the baseline system on the Chinese-English and English-Chinese translation tasks, and also achieve good results on the English-Russian and English- Arabic translation tasks with monolingual data.

Water quality prediction based on fuzzy time series model of dynamic membership degree

ZHAO Chun-lan, LI Yi, HE Ting, WU Gang, WANG Bing

2022, 44(08): 1488-1496. doi:

Abstract ( 76 )

PDF (998KB) ( 98 ) 　　

Scientific and effective water quality prediction is especially important for water resources management and water pollution early warning. The accuracy of the traditional prediction model is somewhat limited by the existence of nonlinearity, non-smoothness, fuzziness, seasonality and other characteristics of water quality index series. This paper combines the characteristics of autoregressive integrated moving average ARIMA model and classical fuzzy time series model, and proposes a new model of fuzzy time series water quality prediction based on dynamic affiliation. First, the use of fuzzy C-mean clustering from the original data to build the affiliation series; second, the use of the classical time series model to predict different sub-subordination series to get the dynamic affiliation; finally, defuzzification to get the predicted value of water quality indicators. The new proposed model in this paper was applied to short-term prediction of water quality indexes at a section of Minjiang River, and compared with the classical fuzzy time series model, ARIMA multiplicative seasonal model. The experimental results show that the RMSE, MAPE and MAE of the new model are better than the classical fuzzy time series model and ARIMA multiplicative seasonal model, so the new prediction model greatly improves the prediction accuracy and can provide valuable reference for water pollution prevention and control.

Current Issue

Author center

Review center

Online journal