2024年度优秀论文
为进一步提升学术影响力，激励和培育优秀作者，国防科技大学学报编辑部组织了2024年度优秀论文评选。经过编辑部筛选和专家评审，最终评选出10篇优秀论文。
关键词：暂无
2024, 46(1):121-130.
DOI: 10.11887/j.cn.202401013
Abstract:
With the increasing depth and the inconsistent length of processing sequences, the performance optimization of RNN(recurrent neural network) on different processors makes it difficult to researchers. An efficient RNN acceleration engine was implemented for the self-developed long vector processor FT-M7032. This engine proposed a row-first matrix vector multiplication algorithm and a data-aware multi-core parallel method to improve the computational efficiency of matrix vector multiplication. It proposed a two-level kernel fusion optimization method to reduce the overhead of temporary data transmission. Optimized handwritten assembly codes for multiple operators were integrated to further tap the performance potential of long vector processors. Experiments show that the RNN engine for long-vector processors is efficient, when compared with the multi-core ARM CPU and Intel Golden CPU, the RNN-like model long short term memory networks can achieve a performance acceleration of up to 62.68 times and 3.12 times, respectively.
With the increasing depth and the inconsistent length of processing sequences, the performance optimization of RNN(recurrent neural network) on different processors makes it difficult to researchers. An efficient RNN acceleration engine was implemented for the self-developed long vector processor FT-M7032. This engine proposed a row-first matrix vector multiplication algorithm and a data-aware multi-core parallel method to improve the computational efficiency of matrix vector multiplication. It proposed a two-level kernel fusion optimization method to reduce the overhead of temporary data transmission. Optimized handwritten assembly codes for multiple operators were integrated to further tap the performance potential of long vector processors. Experiments show that the RNN engine for long-vector processors is efficient, when compared with the multi-core ARM CPU and Intel Golden CPU, the RNN-like model long short term memory networks can achieve a performance acceleration of up to 62.68 times and 3.12 times, respectively.
2024, 46(1):22-31.
DOI: 10.11887/j.cn.202401003
Abstract:
To meet the computational requirements of safe airspace and the expected landing point in the recovery mission of the reentry capsule, a fast prediction algorithm of the flight pipeline based on the Koopman operator approach was proposed for the reentry capsule and the determination process for safe flight airspace of search and rescue helicopters was provided. The body-parachute dynamic model was constructed. A group of discrete state points was uniformly selected from the random state space by using the Halton sampling method, and the multiple possible trajectory was calculated. Based on the back pulling mechanism of Koopman operator, the initial probability density value was associated with the current state to obtain the flight pipeline and desired trajectory of the reentry capsule and its separation parts under uncertain conditions.The simulation results show that the fast prediction algorithm of the flight pipeline based on the Koopman operator approach is significantly better than the Monte Carlo method in terms of convergence speed and accuracy. After using the flight pipeline calculation results to plan the flight route of the rescue helicopter, the collision risk is reduced by 54% at most and the corresponding search time is reduced by 70%. The proposed algorithm has been successfully applied to the Chang′e-5 recovery mission.
To meet the computational requirements of safe airspace and the expected landing point in the recovery mission of the reentry capsule, a fast prediction algorithm of the flight pipeline based on the Koopman operator approach was proposed for the reentry capsule and the determination process for safe flight airspace of search and rescue helicopters was provided. The body-parachute dynamic model was constructed. A group of discrete state points was uniformly selected from the random state space by using the Halton sampling method, and the multiple possible trajectory was calculated. Based on the back pulling mechanism of Koopman operator, the initial probability density value was associated with the current state to obtain the flight pipeline and desired trajectory of the reentry capsule and its separation parts under uncertain conditions.The simulation results show that the fast prediction algorithm of the flight pipeline based on the Koopman operator approach is significantly better than the Monte Carlo method in terms of convergence speed and accuracy. After using the flight pipeline calculation results to plan the flight route of the rescue helicopter, the collision risk is reduced by 54% at most and the corresponding search time is reduced by 70%. The proposed algorithm has been successfully applied to the Chang′e-5 recovery mission.
2023, 45(6):100-131.
DOI: 10.11887/j.cn.202306014
Abstract:
Neural architecture search is a task that aims to automatically search for the optimal neural network structure for different tasks, which is of great importance and inevitability in the joint development of deep learning and computer vision to the current stage. A comprehensive review of the research on neural network search was provided. In specific, the definition and significance of neural architecture search were introduced, and the difficulties and challenges faced in relevant research were deeply analyzed. Based on this, the mainstream search strategies was elaborate and summarize; Finally, the potential problems and possible future research directions were summarized and discussed to promote further development in this field.
Neural architecture search is a task that aims to automatically search for the optimal neural network structure for different tasks, which is of great importance and inevitability in the joint development of deep learning and computer vision to the current stage. A comprehensive review of the research on neural network search was provided. In specific, the definition and significance of neural architecture search were introduced, and the difficulties and challenges faced in relevant research were deeply analyzed. Based on this, the mainstream search strategies was elaborate and summarize; Finally, the potential problems and possible future research directions were summarized and discussed to promote further development in this field.
2023, 45(5):1-11.
DOI: 10.11887/j.cn.202305001
Abstract:
Cognitive electronic warfare is usually defined as a form of electronic warfare that is based on electronic warfare equipment with cognitive performance and focuses on autonomous interactive electromagnetic environment learning capability and dynamic intelligent confrontation task processing capability. Since it was first proposed, it has attracted extensive attention from researchers and scholars at home and abroad for its advantages of accurate perception, strong reasoning and fast decision-making. With the continuous emergence of new concepts, technologies and applications of artificial intelligence, cognitive electronic warfare has stepped into a brand new stage of development. In order to capture its future development direction, the connotation of the concept of cognitive electronic warfare was summarized and enriched from the perspective of artificial intelligence, the development of cognitive electronic warfare and typical foreign projects were sorted out, the framework and architecture of cognitive electronic warfare system was built, a comprehensive and systematic review of the key technologies of cognitive electronic warfare was conducted from the aspects of perception, judgment, decision-making, etc., and the challenges and development trends of cognitive electronic warfare were summarized.
Cognitive electronic warfare is usually defined as a form of electronic warfare that is based on electronic warfare equipment with cognitive performance and focuses on autonomous interactive electromagnetic environment learning capability and dynamic intelligent confrontation task processing capability. Since it was first proposed, it has attracted extensive attention from researchers and scholars at home and abroad for its advantages of accurate perception, strong reasoning and fast decision-making. With the continuous emergence of new concepts, technologies and applications of artificial intelligence, cognitive electronic warfare has stepped into a brand new stage of development. In order to capture its future development direction, the connotation of the concept of cognitive electronic warfare was summarized and enriched from the perspective of artificial intelligence, the development of cognitive electronic warfare and typical foreign projects were sorted out, the framework and architecture of cognitive electronic warfare system was built, a comprehensive and systematic review of the key technologies of cognitive electronic warfare was conducted from the aspects of perception, judgment, decision-making, etc., and the challenges and development trends of cognitive electronic warfare were summarized.
2023, 45(1):157-166.
DOI: 10.11887/j.cn.202301018
Abstract:
Based on the concept of the intelligent combat of UAV (unmanned aerial vehicle) swarms, the UAV swarms intelligent combat simulation environment was established. Aiming at the problem that it is difficult to accurately control the speed and attack angle of UAVs in the confrontation process through reward signals in traditional reinforcement learning algorithms, the RIC-MADDPG (rule and intelligence coupling constrained multi-agent deep deterministic policy gradient) algorithm was proposed. The algorithm uses rules to constrain the actions of UAVs in reinforcement learning. The simulation results show that the wining-rate of red UAV swarm, trained by the method based on the RIC-MADDPG, can be improved from 53% to 79%. This proves that the strategy of "agent training—problem finding—rule making—agent training again—problem finding again—rule making again" is effective for the optimization of agent combat strategy. The research results can be a reference for establishing the training system of the intelligent combat strategy of UAV swarms and conducting the research of swarm tactics coupling rule and intelligence.
Based on the concept of the intelligent combat of UAV (unmanned aerial vehicle) swarms, the UAV swarms intelligent combat simulation environment was established. Aiming at the problem that it is difficult to accurately control the speed and attack angle of UAVs in the confrontation process through reward signals in traditional reinforcement learning algorithms, the RIC-MADDPG (rule and intelligence coupling constrained multi-agent deep deterministic policy gradient) algorithm was proposed. The algorithm uses rules to constrain the actions of UAVs in reinforcement learning. The simulation results show that the wining-rate of red UAV swarm, trained by the method based on the RIC-MADDPG, can be improved from 53% to 79%. This proves that the strategy of "agent training—problem finding—rule making—agent training again—problem finding again—rule making again" is effective for the optimization of agent combat strategy. The research results can be a reference for establishing the training system of the intelligent combat strategy of UAV swarms and conducting the research of swarm tactics coupling rule and intelligence.
2023, 45(1):95-101.
DOI: 10.11887/j.cn.202301010
Abstract:
In order to reveal the aging mechanism of solid propellant grain during the long-term storage, a HTPB (hydroxyl-terminated polybutadiene) propellant rocket motor which was moulded by casting and stored naturally for 19 years was dissected. Samples were taken along the radial position of the propellant grains, and were tested and analyzed by means of SEM(scanning electron microscope), EDX(energy dispersive X-ray spectrometer), infrared absorption spectroscopy and cross-linking density. Results show that the aging degree of propellant is not the same at different positions of grain. The closer the position is to the non-metallic case, the more serious the aging degree of propellant is. Further analysis shows that, in addition to the effects of CC double bond oxidation chain breaking and condensation of —OH on propellant aging, Al3+ ions produced by the oxidation of Al powder catalyzes the addition reaction of HTPB′s CC double bond with Cl2 or water to form CCl double bond and secondary alcohol, and the uneven migration and distribution of aluminum ions will cause the uneven aging of grain. The research results are of great significance to the improvement of propellant aging performance and the research of anti-aging measures.
In order to reveal the aging mechanism of solid propellant grain during the long-term storage, a HTPB (hydroxyl-terminated polybutadiene) propellant rocket motor which was moulded by casting and stored naturally for 19 years was dissected. Samples were taken along the radial position of the propellant grains, and were tested and analyzed by means of SEM(scanning electron microscope), EDX(energy dispersive X-ray spectrometer), infrared absorption spectroscopy and cross-linking density. Results show that the aging degree of propellant is not the same at different positions of grain. The closer the position is to the non-metallic case, the more serious the aging degree of propellant is. Further analysis shows that, in addition to the effects of CC double bond oxidation chain breaking and condensation of —OH on propellant aging, Al3+ ions produced by the oxidation of Al powder catalyzes the addition reaction of HTPB′s CC double bond with Cl2 or water to form CCl double bond and secondary alcohol, and the uneven migration and distribution of aluminum ions will cause the uneven aging of grain. The research results are of great significance to the improvement of propellant aging performance and the research of anti-aging measures.
2023, 45(1):67-73.
DOI: 10.11887/j.cn.202301007
Abstract:
In order to achieve the miniaturization and low power consumption of the low-frequency electromagnetic transmitter, a system design and performance verification method based on multi-physics coupling simulation was proposed for the theoretical research and technical practice of RMBMA(rotating-magnet based mechanical antennas). A RMBMA system scheme including a permanent magnet source, a high-efficiency drive motor and its controller was designed. By developing an experimental prototype, the system performance and near magnetic field characteristics were tested experimentally, and the feasibility and effectiveness of the proposed design method and system scheme were verified. Results show that the main indexes of the prototype, such as magnetic field intensity and size, meet the requirements of the penetrating radio frequency technical area in the AMEBA(a mechanically based antenna) project of the United States, which provides a feasible idea for the scheme research and prototype design of RMBMA.
In order to achieve the miniaturization and low power consumption of the low-frequency electromagnetic transmitter, a system design and performance verification method based on multi-physics coupling simulation was proposed for the theoretical research and technical practice of RMBMA(rotating-magnet based mechanical antennas). A RMBMA system scheme including a permanent magnet source, a high-efficiency drive motor and its controller was designed. By developing an experimental prototype, the system performance and near magnetic field characteristics were tested experimentally, and the feasibility and effectiveness of the proposed design method and system scheme were verified. Results show that the main indexes of the prototype, such as magnetic field intensity and size, meet the requirements of the penetrating radio frequency technical area in the AMEBA(a mechanically based antenna) project of the United States, which provides a feasible idea for the scheme research and prototype design of RMBMA.
2023, 45(1):57-66.
DOI: 10.11887/j.cn.202301006
Abstract:
Matrix transpose is one of the common matrix operations, which is widely employed in various fields such as signal processing, scientific computing, and deep learning. With the popularization of Phytium heterogeneous multi-core DSPs(digital signal processors) developed by National University of Defense Technology, there is a strong demand for high-performance matrix transpose implementations for Phytium multi-core DSPs. Based on the architecture of multi-core DSPs and the characteristic of matrix transpose operations, a parallel matrix transpose algorithm (called ftmMT) for matrices with different element bit widths (8 B, 4 B, and 2 B) was proposed. In ftmMT, the main optimizations include vectorization based on vector Load/Store functions, core-level parallelization based on matrix blocking, and overlapping between vectorization and memory access through implicit ping-pong methods. The experimental results show that ftmMT can significantly improve the performance of matrix transpose operations, and achieve a speedup of up to 8.99 times in comparison with the open-source transpose library HPTT running on CPU.
Matrix transpose is one of the common matrix operations, which is widely employed in various fields such as signal processing, scientific computing, and deep learning. With the popularization of Phytium heterogeneous multi-core DSPs(digital signal processors) developed by National University of Defense Technology, there is a strong demand for high-performance matrix transpose implementations for Phytium multi-core DSPs. Based on the architecture of multi-core DSPs and the characteristic of matrix transpose operations, a parallel matrix transpose algorithm (called ftmMT) for matrices with different element bit widths (8 B, 4 B, and 2 B) was proposed. In ftmMT, the main optimizations include vectorization based on vector Load/Store functions, core-level parallelization based on matrix blocking, and overlapping between vectorization and memory access through implicit ping-pong methods. The experimental results show that ftmMT can significantly improve the performance of matrix transpose operations, and achieve a speedup of up to 8.99 times in comparison with the open-source transpose library HPTT running on CPU.
2022, 44(4):22-31.
DOI: 10.11887/j.cn.202204003
Abstract:
In order to improve the efficiency of the UAV (unmanned aerial vehicle) swarm in cooperative search for moving targets, a UAV swarm cooperative search method was proposed on the basis of the moth pheromone courtship mechanism. According to the courtship behavior of moth in choosing flight direction by pheromone, a wind direction model in pheromone map and a moth pheromone courtship model were established. Considering the constraint of collision avoidance between UAV swarm, a map from moth pheromone courtship mechanism to UAV swarm distributed cooperative search was proposed, and the specific implementation process was given. Simulation results show the effectiveness and stability of the proposed method in solving the cooperative search problem of single moving target, and the outdoor flight experimental results verified the feasibility of the proposed method in practice.
In order to improve the efficiency of the UAV (unmanned aerial vehicle) swarm in cooperative search for moving targets, a UAV swarm cooperative search method was proposed on the basis of the moth pheromone courtship mechanism. According to the courtship behavior of moth in choosing flight direction by pheromone, a wind direction model in pheromone map and a moth pheromone courtship model were established. Considering the constraint of collision avoidance between UAV swarm, a map from moth pheromone courtship mechanism to UAV swarm distributed cooperative search was proposed, and the specific implementation process was given. Simulation results show the effectiveness and stability of the proposed method in solving the cooperative search problem of single moving target, and the outdoor flight experimental results verified the feasibility of the proposed method in practice.
2022, 44(1):55-62.
DOI: 10.11887/j.cn.20220109
Abstract:
Plasma sheath formed during the reentry of near space hypersonic vehicle will interfere the electromagnetic wave detection. In order to solve this problem, the model of the typical near space vehicle was built, and the flow field distribution under different flight conditions during the reentry process was simulated. Based on the flow field distribution, the plasma parameter distribution was modeled, and the transmission characteristics of terahertz wave in the plasma sheath were theoretically calculated by using the scattering matrix method. The results show that high frequency terahertz wave can penetrate plasma sheath effectively. In addition, terahertz active imaging experiments were carried out in the laboratory environment. The experimental results show that the plasma exerts little effect on the terahertz active imaging results. The simulation and experimental results preliminarily prove the potential of terahertz technology for hypersonic vehicle detection, which is of great significance for national defense.
Plasma sheath formed during the reentry of near space hypersonic vehicle will interfere the electromagnetic wave detection. In order to solve this problem, the model of the typical near space vehicle was built, and the flow field distribution under different flight conditions during the reentry process was simulated. Based on the flow field distribution, the plasma parameter distribution was modeled, and the transmission characteristics of terahertz wave in the plasma sheath were theoretically calculated by using the scattering matrix method. The results show that high frequency terahertz wave can penetrate plasma sheath effectively. In addition, terahertz active imaging experiments were carried out in the laboratory environment. The experimental results show that the plasma exerts little effect on the terahertz active imaging results. The simulation and experimental results preliminarily prove the potential of terahertz technology for hypersonic vehicle detection, which is of great significance for national defense.
