Table of Content

15 May 2024 Volume 37 Issue 5

  

Lightweight Capsule Network Fusing Attention and Capsule Pooling

ZHU Zihao, SONG Yan

Electronic Science and Technology. 2024, 37(5):  1-8.  doi:10.16180/j.cnki.issn1007-7820.2024.05.001

Abstract ( 155 )   HTML ( 14 )   PDF (3050KB) ( 87 )  

Figures and Tables | References | Related Articles | Metrics

In view of the inefficiency of feature information propagation in capsule networks and the huge computational overhead in the routing process, a graph pooling capsule network that combines attention and capsule pooling is proposed. The network mainly has the following two advantages: 1) The capsule attention is proposed, and the attention is applied to the primary capsule layer, which enhances the attention to the important capsules, and improves the accuracy of the prediction of the lower capsules to the higher capsules; 2) A new capsule pooling is proposed. The capsule with the largest weight is screened out at the corresponding positions of all feature maps in the primary capsule layer, and the effective feature information is represented by a small number of important capsules while reducing the number of model parameters. Results on public data sets show that the proposed capsule network achieves the accuracy of 92.60% on CIFAR10 and has excellent robustness against white-box adversarial attacks on complex datasets. In addition, the proposed capsule network achieves 95.74% accuracy on the AffNIST data set with superior affine transformation robustness. The calculation efficiency results show that the amount of floating-point operations of the proposed capsule is reduced by 31.3% and the number of parameters is reduced by 41.9% when compared with traditional CapsNet.

A Q-Learning Differential Evolution Algorithm for Combined Heat and Power Dynamic Economic Emission Dispatch

FANG Shuai, CHEN Xu, LI Kangji

Electronic Science and Technology. 2024, 37(5):  9-17.  doi:10.16180/j.cnki.issn1007-7820.2024.05.002

Abstract ( 68 )   HTML ( 4 )   PDF (960KB) ( 45 )  

Figures and Tables | References | Related Articles | Metrics

The dynamic economic emission scheduling of cogeneration takes into account both fuel cost and pollution gas emission, and the thermoelectricity output in the next period is affected by the thermoelectricity output in the current period, which is an important problem in power system operation in recent years. In this study, a new QLMODE(Q-Learning Multi-Objective Differential Evolution) algorithm is proposed to solve the CHPDEED(Combined Heat and Power Dynamic Economic Emission Dispatch) problem. In QLMODE, the Q-learning technique is used to adjust the scale factor parameters of the algorithm, that is, in the iterative process, the action reward and punishment are determined by using the dominant relationship between the child solution and the parent solution, and the parameter values are adjusted by Q-learning to obtain the most suitable algorithm parameters for the environmental model. The proposed QLMODE is used to solve the CHPDEED with 11 units and 33 units. The simulation results show that compared with four mature multi-objective optimization algorithms, the QLMODE algorithm has the least fuel cost and the least pollution gas emission, the convergence and diversity index of QLMODE algorithm is better than the other four algorithms, and QLMODE has a better Pareto optimal frontier on both sets of problems.

Vehicle Detection and Analysis in Urban Waterlogging Area Based on Deep Learning

XIA Rongcheng, LIU Deer

Electronic Science and Technology. 2024, 37(5):  18-24.  doi:10.16180/j.cnki.issn1007-7820.2024.05.003

Abstract ( 137 )   HTML ( 7 )   PDF (4676KB) ( 55 )  

Figures and Tables | References | Related Articles | Metrics

In the urban waterlogging scene, many people and vehicles are trapped in the water, which brings adverse effects to the public life. With the rapid development of computer technology, deep learning is more and more widely used in solving practical problems. This study proposes a method to build a MaskR-CNN(Regions with Convolutional Neural Networks Features) model using TensorFlow deep learning framework, which has achieved good detection results in the detection of waterlogging areas in urban waterlogging scenes, with the mAP(mean Average Precision) value reaching 89%. Based on the YOLOv5(You Only Look Once version 5) model, the dense interframe difference operation is used to track people and vehicles in waterlogged areas, and the tracking accuracy reached about 90%. Moreover, ResNet(Residual Network) attached to YOLOv5 is used to analyze the risk of submersion of vehicles in waterlogging scenarios. The experimental results show that the vehicle risk detection effect of the proposed model is better than other models.

Research on Feeder Automation Test Based on RTLAB Virtual Reality Combination

MA Xiaoyu, CHEN Zhuo, YANG Chao, LI Qingsheng, HAO Zhenghang

Electronic Science and Technology. 2024, 37(5):  25-31.  doi:10.16180/j.cnki.issn1007-7820.2024.05.004

Abstract ( 79 )   HTML ( 1 )   PDF (1019KB) ( 40 )  

Figures and Tables | References | Related Articles | Metrics

In view of the function test of feeder automation, in order to avoid the adverse impact on the safe operation of power system caused by the test experiment of distribution terminal connecting to the actual distribution network, and to improve the flexibility, efficiency and correctness of the test, a virtual and real feeder automation test method is proposed. The RTLAB(Real Time Laboratory) real-time full digital simulator is used to build the distribution network and simulate the fault operation. A simulated virtual distribution terminal with protection, recloser and in-place FA (Feeder Automation) functions is designed to carry out the RTLAB real-time simulator and physical model hardware simulation research. An interface implementation scheme based on high performance linear power amplifier is designed to form a closed-loop access test system for the measured distribution terminal, and an automatic simulation test environment combining virtual and real is constructed. The accuracy of the interface device and the effectiveness of the test platform are verified through the hardware-in-the-loop simulation experiment of the basic fault handling capability test and fault tolerance capability test.

Multiple Embedded AI Processors Low Latency Data Exchange Technology Based on PCIE Interface

WEI Xuan, WEN Kailin, LI Bin, LIU Shutao, CHU Jie, CAI Jueping

Electronic Science and Technology. 2024, 37(5):  32-37.  doi:10.16180/j.cnki.issn1007-7820.2024.05.005

Abstract ( 68 )   HTML ( 2 )   PDF (1212KB) ( 48 )  

Figures and Tables | References | Related Articles | Metrics

In view of the conflict of task scheduling and data exchange between multiple embedded AI(Artificial Intelligence) processors and improving the reliability and efficiency of stack expansion of multiple AI processors, a high speed data exchange technology and data frame structure of multiple embedded AI processors with wormhole switching structure are proposed in this study. Based on the PCIE(PCI Express) high-speed data interface, the data is transmitted in the form of data unit, and the multi-weight decision algorithm is designed to avoid the conflict in data transmission and realize the concurrent multi-threading of the task.The FPGA(Field Programmable Gate Array) platform is designed and tested. The results show that the transmission bandwidth utilization efficiency of PCIE reaches more than 85%, the data exchange delay is less than 20 μs, and the average maximum delay time of interrupt task response is 8.775 μs.The technology is suitable for high-speed switching circuits with multi-processor collaboration and can be extended to hybrid PCIE and RapidIO switching circuit architectures.

Research on Automatic Current Equalization Control Strategy of Redundant Power Supply

CHI Kangwei, XIE Suxia, HUANG Song, WANG Jinhao

Electronic Science and Technology. 2024, 37(5):  38-46.  doi:10.16180/j.cnki.issn1007-7820.2024.05.006

Abstract ( 78 )   HTML ( 4 )   PDF (3260KB) ( 50 )  

Figures and Tables | References | Related Articles | Metrics

Redundant power supplies are commonly used in systems that require long-term uninterrupted operation and high reliability. When redundant power supplies are used as N+1 backup power supplies to supply power to devices at the same time, parallel current sharing must be considered.In order to solve the problem of large loss of passive redundant power supply and inability to achieve current equalization, this study adopts the design scheme of active redundancy and designs a 1+1 redundant power supply system with two parallel DC voltage inputs.The system uses GD32E103 as the main control chip. By changing the on-resistance of MOSFET(Metal-Oxide Semiconductor Field Effect Transistor), PI(Proportional Integral) control is adopted to realize the automatic balance of the output current of two redundant power supplies.The simulation and experimental results show that the active redundancy design scheme can achieve the purpose of automatic current equalization.In normal operation, the maximum error of the two current channels is less than 0.1 A, and the accuracy of current sharing meets the design requirements.

Research on Transformer Fault Diagnosis Based on Improved Bayesian Network

TONG Zhaojing, LAN Mengyue, JING Lifei

Electronic Science and Technology. 2024, 37(5):  47-53.  doi:10.16180/j.cnki.issn1007-7820.2024.05.007

Abstract ( 87 )   HTML ( 6 )   PDF (1826KB) ( 47 )  

Figures and Tables | References | Related Articles | Metrics

In view of the low accuracy of transformer fault diagnosis, a transformer fault diagnosis method based on ISMA(Improved Slime Mold optimization Algorithm) and optimized BN(Bayesian Network) is proposed. The hill-climbing algorithm searches the oriented maximum support tree to obtain the initial structure of the Bayesian network, that is, the initial population. The reverse learning strategy and SCA(Sine Cosine Algorithm) are introduced into the improved slime mold optimization algorithm to increase population diversity, update population location, and avoid the population falling into local optimal. The characteristics of transformer fault state are selected by the improved code-free ratio method, and the structure of Bayesian network is optimized by the improved slime mold optimization algorithm to improve the accuracy of transformer fault diagnosis based on Bayesian network. Different kinds of test functions are used to verify that the improved slime mold optimization algorithm has the excellent performance of fast convergence speed and high convergence accuracy. The simulation results show that the accuracy of the training set and the test set of the ISMA-BN diagnostic model is up to 98.2% and 97.14%, respectively, which has certain research value.

Research on S-Type Speed Control Strategy of Vending Machine Lifting System

FENG Gaoming, QIU Wenxian, JIN Minglei

Electronic Science and Technology. 2024, 37(5):  54-61.  doi:10.16180/j.cnki.issn1007-7820.2024.05.008

Abstract ( 50 )   HTML ( 1 )   PDF (1240KB) ( 42 )  

Figures and Tables | References | Related Articles | Metrics

Under the traditional three-closed loop control mode, the vending machine lifting and receiving system is prone to impact due to the sudden acceleration during the start and stop process, which leads to the unstable operation of the system. To solve this problem, the traditional acceleration and deceleration control strategy of lifting system is improved, and a S-type velocity curve control algorithm is proposed by adding feed forward and differential negative feedback. According to the discriminant conditions of path planning, S-type velocity curve is divided into three types: seven-stage, six-stage and four-stage. The parameter solving method and the specific execution process under each path constraint condition are given. The algorithm is applied to the lifting control system for simulation and actual working condition test. The experimental results show that compared with the traditional three-closed-loop control, the provided control strategy can improve the stability of the lifting system, reduce the impact, make the speed curve change more gentle, and maintain good tracking performance.

Surface Defect Detection of Transparent Objects Based on Phase Measuring Deflectometry

DING Yujie, ZHOU Zhifeng, WANG Yong, WANG Lirui

Electronic Science and Technology. 2024, 37(5):  62-70.  doi:10.16180/j.cnki.issn1007-7820.2024.05.009

Abstract ( 180 )   HTML ( 3 )   PDF (3719KB) ( 92 )  

Figures and Tables | References | Related Articles | Metrics

At present,the traditional method is still used to control the quality of glass,lens and other transparent objects in China,and the traditional visual evaluation method is inefficient.In order to realize automatic detection of surface defects of transparent objects,a transmission detection method based on PMD(Phase Measuring Deflectometry) is proposed.The structure light fringe pattern is generated by PMD algorithm combined with the new phase shift pattern generation formula,and the fringe pattern is projected to the surface of the measured object using the transmission system. The distorted fringe image after refraction of the measured object is collected by the camera,then the absolute phase diagrams are generated, and the defects are extracted by identifying the local distortion in the absolute phase diagram.Through analyzing the reason of false detection caused by periodic misalignment,a method of correcting absolute phase periodic misalignment is proposed. The new formula of phase shift pattern generation can also correct periodic misalignment in advance,the combination of the two methods can improve the precision of phase unwrapping. A concave and convex lens with a focal length of 300 mm is taken as an example, the experimental results show that the proposed method can accurately extract the local distortion in the absolute phase diagram caused by surface defects with an accuracy of 0.1 mm.

Multimodal Android Malware Detection Method Based on Behavioral and Semantic Characteristics

ZHU Jinkai, FANG Lanting, JI Xiaowen, HUANG Jie

Electronic Science and Technology. 2024, 37(5):  71-78.  doi:10.16180/j.cnki.issn1007-7820.2024.05.010

Abstract ( 103 )   HTML ( 2 )   PDF (998KB) ( 45 )  

Figures and Tables | References | Related Articles | Metrics

Existing methods for detecting Android malware only consider a single kind of features, which do not fully describe the features of Android software. In order to solve the above problems, this study presents a multimodal Android malware detection method based on the permissions, byte code probability matrix and function call graph. At the same time, in order to solve the problem of feature representation of function nodes, a new node feature generation method is presented in this study in the generation of function call graph. In order to enrich the semantic information of opcode, a byte probability matrix generation method based on 2-gram is presented. The experiment proves that the method described the characteristics of Android software more comprehensively than other methods, and the detection accuracy rate reached 95.2%. Compared with the existing methods, the accuracy of this method has been improved by 22% on average, effectively improving the detection ability of Android malware.

Event-Triggered Fault Detection for Delta Operator Network Control Systems

GE Xiaowei, ZHOU Ying

Electronic Science and Technology. 2024, 37(5):  79-87.  doi:10.16180/j.cnki.issn1007-7820.2024.05.011

Abstract ( 90 )   HTML ( 5 )   PDF (1536KB) ( 42 )  

Figures and Tables | References | Related Articles | Metrics

A fault detection method based on a combination of event-triggered mechanism and Delta operator is used for the fault problems caused by time delay and packet loss in the network due to bandwidth limitation of the communication network in the case of high-speed sampling of the network control system. In this study, while co-designing the fault detection filter and the controller, the construction vector is added to the Delta operator filter model to avoid the generation of bilinear terms in the parameter design. Constructing Lyapunov-Krasovskii generalized connotations of the δ-domain for the stability and H_∞ performance of the residual system. The numerical simulation example verifies the effectiveness of the proposed event triggering mechanism and demonstrates its high sensitivity to faults by the residual evaluation function.

Research on Robot Global Path Planning Based on Improved Ant Colony Algorithm

WANG Yanchun, GUO Yongfeng, XIA Ying, WANG Yangyang

Electronic Science and Technology. 2024, 37(5):  88-94.  doi:10.16180/j.cnki.issn1007-7820.2024.05.012

Abstract ( 152 )   HTML ( 6 )   PDF (892KB) ( 100 )  

Figures and Tables | References | Related Articles | Metrics

In response to the problems of traditional ant colony algorithm such as lack of initial pheromone, slow convergence speed and inability to effectively avoid obstacles, this study proposes a global path planning based on improved ant colony algorithm.The introduction of a normal distribution function improves the traditional heuristic function, greatly improving the efficiency of the algorithm and shortening the time required for convergence; adaptively adjusting the pheromone volatility coefficient to limit the pheromone range and avoid premature convergence; and smoothing the algorithm path to shorten the path length, thus realising global path planning for the robot.Simulation results show that under a 20×20 environment, the average number of iterations of the proposed algorithm is 28 generations less than that of the traditional ant colony algorithm, resulting in faster convergence, the average number of inflection points is reduced by 33.3%, making the path smoother, overcoming the lack of initial pheromone, speeding up convergence, reducing the number of inflection points, and enabling effective avoidance of obstacles in the environment, demonstrating the feasibility of the algorithm.