Abstract:近年来，随着智能无人技术的迅速发展，智能化无人机集群作战成为各国重点研究的新型作战样式，智能无人作战系统在实战中已崭露头角。人工智能技术和军事智能化在无人集群作战认知域、信息域、物理域的深化发展，必将颠覆未来作战理念、作战样式、作战管理与作战技术。智能无人集群作战技术是智能决策、群体智能、智能控制、通信网络等多种相关技术的综合运用。目前，要实现智能化无人集群作战，亟需解决全面战场环境感知、智能决策与自主攻击等难题，同时对集群系统的通信网络技术也提出了更高的要求。本期专题主要依托 “无人集群跟踪与编队重点自主协同关键技术研究”“航空集群空空导弹攻击区模型及智能预测方法研究”“融合眼动与事件相关电位的认知耦合目标识别方法研究”等国家自然科学基金项目，目的是汇聚智能无人作战的无人机自主控制、智能协作、通信网络和集群博弈对抗等方面理论与技术方法的研究成果，探索面向复杂环境的智能无人集群系统的运用模式和效能机理，搭建学术研究与技术运用的交流平台，为相关研究人员提供有益的参考，共同推动智能无人作战技术的发展和提高。本专题采用视频加载等增强出版形式，读者可扫描文中二维码链接来观看相关视频资料，以加深对所研究问题的认识。限于研究者水平，文中所述方法及结论可能存在一定局限性。

Abstract:In the process of visual navigationbased autonomous landing, aimed at the problems that being influenced by its mechanical vibration and compound wind field, rotor unmanned aerial vehicle (UAV) is low in landing accuracy, and slow at speed by means of which swarm recycling safety is affected, a swarm autonomous landing algorithm based on fuzzy control and visual navigation is proposed in this paper. When the UAV swarm is disposed to the landing area, all UAVs can find their corresponding landing marks through the object detection algorithm and calculate the actual horizontal distance to corresponding landing mark by making use of pixel distance, and then the control command of making UAVs into alignment with landing point can be acquired by fuzzification, fuzzy reasoning and defuzzification to realize the accurate landing of swarm. The results of stimulation experiment and actual flight experiment show that the algorithm is highly robust and can improve the speed of the swarm landing of UAVs.

Abstract:As an application platform in target detection, unmanned aerial vehicles play an incomparable advantage and characteristics in reconnaissance missions. However, the limited memory and computing power of the UAV platform are difficult in detection model deployment and slow at detection speed. To solve the above problems, an improved model based on YOLOv4 is proposed. Firstly, in order to reduce the memory usage of the model and save computing resources, this paper improves the prediction layer of the original YOLOv4 model according to the characteristics of the target size. Secondly, the improved model is trained, and then sparse training and channel pruning on the scaling factor of the BN layer are made to reduce the memory usage of the model again to improve the detection speed. The experimental results show that with the detection results being basically the same, the memory usage of the improved model is reduced by 54%, and the FPS is increased by 35% compared with the original model, reaching 58 frames per second respectively.

Abstract:In view of the problem of autonomous maneuvering counterpursuing in close air combat, a Markov decisionmaking process model for UAV counterpursuing is established, and for the abovementioned reasons, an autonomous maneuvering decisionmaking method for unmanned aerial vehicles (UAVs) based on deep reinforcement learning is proposed. The new method is based on the empirical replay area reconstruction, and improves the Twin Delayed Deep Deterministic policy gradient (TD3) algorithm, and generates the optimal strategy network by fitting the strategy function and the state action value function. The simulation experiments show that under condition of random initial position/attitude, being confronted with the drones adopted by the pure pursuit methods, the winning rate of intelligent drones trained by this method exceeds 93%. Compared with traditional TD3 and Deep Deterministic policy gradient (DDPG), this method is faster at convergence and higher in stability.

Abstract:Aimed at the problems that UAV is high mobility at nodes, flying ad hoc networks are drastic changes in network topology and frequent link disconnection. Reactivegreedyreactive (RGR) routing protocol is proposed with a good network performance in high dynamic environment. In view of high network overhead and network congestion, an improved RGR routing protocol (LBHGSRGR) based on load balancing and high greedy geographical forwarding success probability is proposed. Based on the RGR protocol, three key improvement measures are proposed, which are scoped flooding mechanism based on node load status and geographic location information, path selection strategy with high packet success probability in GGF mode, and packet forwarding strategy based on node load prediction and motion characteristics. The simulation results show that compared with AODV, RGR and their improved protocols, this protocol improves packet delivery rate, reduces network control overhead and average endtoend delay, improves the ability of network to cope with highly dynamic topology changes, and effectively improves network performance.

Abstract:Awareness of existing unmanned aerial vehicle identification method being visual detection, and easily affected by weather changes and many other factors such as visible detection range, and the surrounding buildings shade, etc., a convolution of the neural network based on depth unmanned aerial vehicle link perceptual recognition algorithm is proposed, giving a multimode multitype uavs RF signal database build steps, and the proposed convolution neural network is designed and optimization method is made in detail. The measured results show that the depth algorithm proposed in this paper can not only realize multibatch and multitarget UAV intrusion identification, but also further distinguish its model from flight mode. Under condition of low signaltonoise ratio as low as -20 dB, the uav batch identification rate is 96.8% (6 categories), and the flight mode identification rate is 94.4% (12 categories). This method is prosperous in a strong application.

Abstract:Aimed at the problems that the stability region of the system corresponding to parameters such as DC side filter is difficult to determine and select the main factors of parameters in the multielectric aircraft power system, and the main parameter participating factors and the long stability time are also difficult to select, a method based on eigenvalue participating factors is proposed. dq transformation method is used to carry out the equivalent transformation of the aircraft power system, and its mathematical model is established. Taylor series is used to linearize the small signal, and a linear small signal analysis model of its dynamic characteristics is obtained. And then, the eigenvalue method is used to analyze the stability of the system, and the change trend of the stability when the parameters change is obtained through its participation and participation factors, determining the stability boundary of the system and predict the position of the unstable point. Through Matlab/Simulink, the system simulation model is established and the simulation analysis is carried out. The results show that the stability analysis of the power system by using the participation factor could significantly shorten the time to select the dominant parameters, clearly and 〖JP2〗quickly display the stability domain and instability point of the system under different dominant parameters.

Abstract:The establishment of a wing structural load model with flight parameters as variables is an important technical basis for flight safety monitoring and aircraft fatigue life estimation. Firstly, the influence of wing fuel quality on its structural load is separated. On the basis of this and in accordance with the correlation between aircraft structural load and flight parameters, and the lowdimensional and the uncorrelated modeling parameters determined through correlation analysis combined with principal component analysis, a model is established by the BP neural network method pretrained by GaussBernoulli restricted Boltzmann machine. Taking the aircrafts transonic pitch maneuver for example, a shear force model of a certain measured load profile of the wing is established. The model verification results show that the pretraining can effectively reduce the initial error of model, and improve the modeling efficiency and accuracy.

Abstract:In view of the problem of infrared nondestructive testing for unbounded defects of glass fiber reinforced plastic laminates, a kind of artificial unbounded defect sample is prepared firstly. The unbounded defects are detected by infrared pulse thermal wave imaging technology, and the transient response process of surface thermal signals in unbounded area and non debonding area of laminates is analyzed. The image signalnoise ratio and normalized contrast are used as evaluation indexes to quantitatively analyze the effects of four thermal image reconstruction algorithms, including thermography signal reconstruction, complex modulation ZoomFFT, improved independent component analysis (ICA) and principal component analysis (PCA), and the function of the unbounded defect recognition. On this basis, the PCA algorithm based on thermal signal reconstruction enhancement is proposed, and the effect of the algorithm in unbounded defect recognition is verified. The results show that the four thermal image reconstruction algorithms can improve the quantitative identification ability of unbounded defects, in which the thermal signal reconstruction is the most significant to improve the contrast between the defect area and the non defect area, and the principal component analysis has the strongest ability to suppress the thermal image noise; and the principal component analysis based on the enhancement of thermal signal reconstruction can significantly improve the quantitative identification ability of unbounded defects with the depth of 0.5 mm, 1.0 mm and 1.5 mm.

Abstract:In order to realize the intelligent identification of internal lamination defects of aviation glass fiber composites, a spectroscopy system with multidegree of freedom fiber coupling terahertz time domain is built. The samples with simulated internal lamination defects are detected, and the detection results are filtered, enhanced and marked, and the data sets for target detection are constructed. At the same time, a modified YOLOv4 algorithm is proposed to improve the accuracy of intelligent defect recognition. The experimental results show that the improved YOLOv4 algorithm achieves 91.05% accuracy and 92.02% recall rate in the test set, which is 5.73% and 8.51% higher than the original YOLOv4 algorithm, respectively. This algorithm is characterized by a stronger feature extraction capability and good robustness, and obviously eliminates the error detection and omissions of the original YOLOv4 algorithm.

Abstract:Aimed at the problems that recognition rate is low, and stability of anti jamming and anti noise is poor, with the result that individual classification of communication emitter is poor and interference ability of fingerprint feature extraction algorithm of communication emitter is poor, a method based on empirical mode decomposition and singular value decomposition is proposed. The effect of noise on fingerprint feature extraction is overcome with signal being subjected to the Empirical Mode Decomposition, the fingerprint feature extraction of signal source is realized by the HilbertHuang Transform and Singular Value Decomposition in combination with Support Vector Machine (SVM) algorithm to complete individual identification of communication source, thus improving the accuracy of the classification and recognition. The experimental verification of the four types of emitter signals show that the ascension of recognition effect is obvious.

Abstract:Aimed at the problem that the traditional sparse array is hard to realize direction finding error caused by the synchronization optimization of aperture and mutual coupling, an interelement spacing restriction array with high degree of freedom and low mutual coupling is designed. The array is composed of four uniform linear arrays connected from beginning to end at a certain interval, and the restriction at which the array element spacing of each uniform linear array and the space between each uniform linear array are as large as possible is done, thus forming three sparse uniform linear arrays and one dense linear array, and effectively reducing the mutual coupling effect among array elements. Based on this array, the closed solutions of physical array element position and difference joint array and the closed solutions of degree of freedom are derived. Compared with the traditional and improved sparse arrays with the same number of array elements, the designed interelement spacing restriction array has larger aperture, lower mutual coupling and more continuous virtual array elements. The advantages of interelement spacing restriction array are verified by experimental simulation.

Abstract:Active neutral point clamped (ANPC) inverters are characterized by low output waveform distortion rate and high transmission efficiency, and have been wide use, but a large numbers of switching devices reduce the reliability of the inverter. The current path of a threelevel ANPC inverter under different opencircuit faults of switching devices is analyzed, and the output level and output voltage space vector changes under different faults are obtained. Combining the relationship between the midpoint current and the output current under the action of each vector, a fault diagnosis method based on the midpoint current is proposed. According to the characteristics of the ANPC inverter circuit, different faulttolerant modes are proposed for its bridge arm devices and clamp devices, and a faulttolerant circuit structure with nonderating fault tolerance is proposed. The proposed fault diagnosis method and faulttolerant strategy are verified by building a simulation model and experimental platform. The simulation and experimental results show that the fault diagnosis method and faulttolerant strategy are feasible.

Abstract:Aimed at the problems that at present most of the current researches on multicontroller deployment in softwaredefined networks mainly focus on the optimization of the control network delay, reliability and load balancing, but less on the overall consideration of the above factors, this paper firstly analyzes the impact of controller deployment on network delay, reliability and load balancing. Secondly, the paper proposes an optimal evaluation model of controller deployment, which takes the average delay, control path reliability and load balance of the whole network as parameters while the comprehensive performance of the network as the objective. Finally, based on the simulated annealing genetic algorithm, the paper proposes a delay and reliability aware controller deployment method in consideration of the overall performance of the network, enhancement of the global search ability of the solution space, obtaining the non inferior optimal solution set of controller deployment from global. The simulation results show that on the premise of ensuring load balancing, the proposed deployment strategy improves the reliability of the control network, reduces the network delay and improves the overall performance of the network.

Abstract:In view of the problem of intelligent jamming decisionmaking in battlefield communication, an action elimination dueling double deep Q network intelligent jamming decision algorithm is designed. Based on the framework of double deep Q network, this algorithm utilizes a neural network with a dueling structure for determining the optimal jamming action in combination with the advantage function to judge the relative pros and cons of each jamming action. And then on the basis of the above mentioned, an invalid jamming action elimination mechanism is introduced to speed up the learning of the best jamming strategy. The method can learn a better jamming strategy in the face of an unknown communication antijamming strategy. The simulation results show that when the enemy changes communication strategy, the method can adaptively adjust the jamming strategy and have stronger robustness. Compared with the existing methods, this method can achieve a higher jamming success rate with greater jamming efficacy.

Abstract:In order to explore the dynamic strength and failure mechanism of red sandstone under condition of hydrochemical damage, the static uniaxial compression and dynamic uniaxial impact tests of natural, dry and water saturated red sandstone samples are carried out. Combined with the SEM images of rock fragments, the strength characteristics of rocks under condition of different water bearing and strain rate loading levels are analysed, and based on the damage fracture theory, the initiation and propagation mechanism of micro cracks in water bearing rock are analysed as well. The test results show that the dynamic compressive strength of red sandstone decreases with the increase of water content and increases with the increase of strain rate, and the saturated sample has significant strain rate effect; under condition of impact load, the stressstrain curve of saturated sample has significant volume compression phenomenon, with the maximum peak strain and obvious plastic deformation, while the dry sample has the largest elastic deformation and the smallest plastic deformation before the peak; under the influence of pore water, the grain structure of saturated sample is loose and porous, and the cementation is weakened by dissolution of cementation material. According to the theory of maximum circumferential normal stress, the conditions of initiation and propagation direction of micro cracks in water bearing rock are discussed, and a modification to the dynamic stress intensity factor at crack tip is made.

Abstract:In view of the characteristics of complex stress and high tensile stress in anchorage zone of crosstensioned prestress concrete pavement, the experimental study on local compressive performance of fiber reinforced concrete is carried out, and the typical failure mode of fiber reinforced concrete under local compressive load is obtained. Furthermore, the influence of the supporting conditions and fiber content on the cracked and ultimate strength is analyzed. The failure mechanism of fiber reinforced concrete under local compressive load is revealed. Additionally, the enhancement coefficient of local compressive bearing capacity of fiber reinforced concrete with ducts is obtained based on the pullrod model. The results show that the supporting conditions and fiber content of fiber reinforced concrete have significantly influence on its local compressive bearing capacity and failure mode. The local compressive bearing capacity of fiber reinforced concrete increases gradually with the increase of fiber content for the same support condition. The local compressive bearing capacity of the fiber reinforced concrete under full supporting condition is significantly higher than that under partial supporting condition for the same fiber content. The ratio of local compressive area and the size of reserved ducts significantly influence the enhancement coefficient of local bearing capacity.