ISSN 
2023 Vol. 52, No. 4
2023, 52(4): 482-488.
doi: 10.12178/1001-0548.2022220
Abstract:
In this paper, a quantum fuzzy inference system based on adaptive network (ANQFIS) is proposed based on ANFIS and quantum BP (QBP) neural network. Different from ANFIS, ANQFIS combines the strength of fuzzy rules with QBP in the way of quantum gate rotation, and finally takes the measurement probability of quantum states as the output. The addition of QBP makes the output accuracy of the model higher, and the calculation speed of the model is improved by virtue of the speed advantage of quantum computing. According to the gradient descent method, the parameters learning algorithm of the system is given. In the simulation experiment, low-dimensional data and high-dimensional data are used as data sets to train the model, and attack algorithms are used to generate adversarial examples for testing. The results show that ANQFIS is superior to ANFIS and QBP in output accuracy and robustness.
In this paper, a quantum fuzzy inference system based on adaptive network (ANQFIS) is proposed based on ANFIS and quantum BP (QBP) neural network. Different from ANFIS, ANQFIS combines the strength of fuzzy rules with QBP in the way of quantum gate rotation, and finally takes the measurement probability of quantum states as the output. The addition of QBP makes the output accuracy of the model higher, and the calculation speed of the model is improved by virtue of the speed advantage of quantum computing. According to the gradient descent method, the parameters learning algorithm of the system is given. In the simulation experiment, low-dimensional data and high-dimensional data are used as data sets to train the model, and attack algorithms are used to generate adversarial examples for testing. The results show that ANQFIS is superior to ANFIS and QBP in output accuracy and robustness.
2023, 52(4): 489-497.
doi: 10.12178/1001-0548.2022339
Abstract:
The coupling constraint of noisy intermediate-scale quantum hardware makes most quantum algorithms change quantum bit mapping by inserting additional quantum gates, so that quantum algorithms run directly on the hardware. In order to reduce the quantum circuit running time and improve the quantum circuit fidelity, this paper designs a SWAP-based prospective heuristic quantum circuit mapping algorithm. First, the prospective mechanism is used to consider the front layer information, which improves the stability of the additional gate count results. Secondly, the search strategy is designed to evaluate the physically close candidate SWAP gates to reduce the search space complexity of SWAP gates. Finally, a bidirectional traversal is used to globally consider the gate information of quantum circuits to obtain higher quality initial mappings. The proposed algorithm is suitable for arbitrary coupled quantum hardware architecture, and has the ability to select circuit depth and additional gate number. The experimental results show that compared with the mainstream algorithms A*-based algorithm (Algorithm based on A* search) and SABRE algorithm (SWAP-based BidiREctional heuristic search algorithm), the SPBHA (SWAP-based Prospective Bidirectional Heuristic) algorithm proposed in this paper can reduce the number of additional gates by about 68% and 34%, shorten the circuit execution time, and ensure the reliability of the quantum program results.
The coupling constraint of noisy intermediate-scale quantum hardware makes most quantum algorithms change quantum bit mapping by inserting additional quantum gates, so that quantum algorithms run directly on the hardware. In order to reduce the quantum circuit running time and improve the quantum circuit fidelity, this paper designs a SWAP-based prospective heuristic quantum circuit mapping algorithm. First, the prospective mechanism is used to consider the front layer information, which improves the stability of the additional gate count results. Secondly, the search strategy is designed to evaluate the physically close candidate SWAP gates to reduce the search space complexity of SWAP gates. Finally, a bidirectional traversal is used to globally consider the gate information of quantum circuits to obtain higher quality initial mappings. The proposed algorithm is suitable for arbitrary coupled quantum hardware architecture, and has the ability to select circuit depth and additional gate number. The experimental results show that compared with the mainstream algorithms A*-based algorithm (Algorithm based on A* search) and SABRE algorithm (SWAP-based BidiREctional heuristic search algorithm), the SPBHA (SWAP-based Prospective Bidirectional Heuristic) algorithm proposed in this paper can reduce the number of additional gates by about 68% and 34%, shorten the circuit execution time, and ensure the reliability of the quantum program results.
2023, 52(4): 498-505.
doi: 10.12178/1001-0548.2022283
Abstract:
Aiming at the estimation problem of the Direction of Arrival (DOA) of weak targets in the presence of strong mainlobe interference for digital sub-array radar, this paper proposes a sum-difference beam angle measurement method. Firstly, a receiving model of the mainlobe interference and target echo under the multi-subarray sum-difference beam structure is established. Then, according to the uncorrelated characteristics of the target echo and the interference signal in the time domain, blind source separation is used for each sub-array to separates the target echo and interference signal, and monopulse angle measurement technology is used to achieve target/interference DOA estimation via combining the target echo/interference signal separated by each sub-array. Finally, the effectiveness and analysis performance of this method is verified by numerical simulation experiment.
Aiming at the estimation problem of the Direction of Arrival (DOA) of weak targets in the presence of strong mainlobe interference for digital sub-array radar, this paper proposes a sum-difference beam angle measurement method. Firstly, a receiving model of the mainlobe interference and target echo under the multi-subarray sum-difference beam structure is established. Then, according to the uncorrelated characteristics of the target echo and the interference signal in the time domain, blind source separation is used for each sub-array to separates the target echo and interference signal, and monopulse angle measurement technology is used to achieve target/interference DOA estimation via combining the target echo/interference signal separated by each sub-array. Finally, the effectiveness and analysis performance of this method is verified by numerical simulation experiment.
2023, 52(4): 506-511.
doi: 10.12178/1001-0548.2023112
Abstract:
A highly stable network topology is an important guarantee for the collaborative functions of the Unmanned Aerial Vehicle (UAV) swarm system, including the distributed joint sensing, distributed information interaction and distributed cooperative control. In 3D dynamic application scenarios, fast and stable network topology construction is crucial for reliable applications of swarm systems, and current topology construction methods are not sufficiently studied in this aspect. In this paper, a fast distributed topology construction method for UAV swarm system is proposed, the method is based on a Particle Swarm Optimization (PSO) algorithm that maximizes the duration of network topology under the specific end-to-end communication delay performance requirement. To achieve fast convergence of the distributed topology construction method, the initial values are designed according to the static characteristics and dynamic trends of the swarm nodes, while the update direction and step size are optimized based on the feature similarity function. The simulation results show that the proposed method of topology construction has high stability under typical application scenarios and system configurations. With 100 nodes, the traditional PSO strategy requires an average of 5.5 iterations to obtain the optimal solution, while the proposed algorithm converges to the global optimal solution in an average of 2 iterations obtaining the same end-to-end delay and network throughput performance.
A highly stable network topology is an important guarantee for the collaborative functions of the Unmanned Aerial Vehicle (UAV) swarm system, including the distributed joint sensing, distributed information interaction and distributed cooperative control. In 3D dynamic application scenarios, fast and stable network topology construction is crucial for reliable applications of swarm systems, and current topology construction methods are not sufficiently studied in this aspect. In this paper, a fast distributed topology construction method for UAV swarm system is proposed, the method is based on a Particle Swarm Optimization (PSO) algorithm that maximizes the duration of network topology under the specific end-to-end communication delay performance requirement. To achieve fast convergence of the distributed topology construction method, the initial values are designed according to the static characteristics and dynamic trends of the swarm nodes, while the update direction and step size are optimized based on the feature similarity function. The simulation results show that the proposed method of topology construction has high stability under typical application scenarios and system configurations. With 100 nodes, the traditional PSO strategy requires an average of 5.5 iterations to obtain the optimal solution, while the proposed algorithm converges to the global optimal solution in an average of 2 iterations obtaining the same end-to-end delay and network throughput performance.
2023, 52(4): 512-522.
doi: 10.12178/1001-0548.2022205
Abstract:
In order to solve the problem that it is difficult to accurately estimate the fast time-varying channel state information in orthogonal frequency division multiplexing (OFDM) communication system of high-speed railway in high-speed mobile scene, an OFDM channel estimation method based on multi-scale residual attention network was proposed. Firstly, we design the multi-scale channel feature extraction structure. For the low-quality channel matrix, we apply the multi-scale convolution kernel to extracting the shallow multi-dimensional feature information, which can improve the extraction performance of channel feature information with different scales. Then, a multi-scale residual attention cascade depth network is constructed for channel feature reconstruction and mapping. The local residual feedback is combined with CBAM (convolutional block attention module) attention mechanism to promote the fusion and utilization of deep features and improve the reconstruction and mapping ability of OFDM channel matrix. Finally, the sub-pixel convolution reconstruction is used to generate a high-resolution channel matrix to complete the channel estimation. The analysis in both frequency domain and time domain show that the proposed channel estimation method is better than other methods in terms of accuracy and complexity of channel estimation and can satisfy the needs of OFDM channel estimation.
In order to solve the problem that it is difficult to accurately estimate the fast time-varying channel state information in orthogonal frequency division multiplexing (OFDM) communication system of high-speed railway in high-speed mobile scene, an OFDM channel estimation method based on multi-scale residual attention network was proposed. Firstly, we design the multi-scale channel feature extraction structure. For the low-quality channel matrix, we apply the multi-scale convolution kernel to extracting the shallow multi-dimensional feature information, which can improve the extraction performance of channel feature information with different scales. Then, a multi-scale residual attention cascade depth network is constructed for channel feature reconstruction and mapping. The local residual feedback is combined with CBAM (convolutional block attention module) attention mechanism to promote the fusion and utilization of deep features and improve the reconstruction and mapping ability of OFDM channel matrix. Finally, the sub-pixel convolution reconstruction is used to generate a high-resolution channel matrix to complete the channel estimation. The analysis in both frequency domain and time domain show that the proposed channel estimation method is better than other methods in terms of accuracy and complexity of channel estimation and can satisfy the needs of OFDM channel estimation.
2023, 52(4): 523-529.
doi: 10.12178/1001-0548.2022342
Abstract:
The zero-IF receiver has I/Q imbalance in the quadrature mixer circuit, which leads to a sharp decline in the receiver's image rejection capability, and the signal-to-noise ratio and dynamic range of the system are affected. This paper designs a method to calculate and compensate the linear phase imbalance of I/Q signal. ADC sampling and Fourier transform are performed on the demodulated I/Q two-way baseband signals, and then the single-side spectrum is taken to calculate the cross-power spectrum and the phase spectrum of the cross-power spectrum. The three-point method is used for phase unwrapping, the phase compensation is performed according to the unwrapped group delay difference and phase offset, and finally the time domain signal is restored. After testing, the method has good compensation effect, simple implementation, low algorithm complexity and high real-time performance.
The zero-IF receiver has I/Q imbalance in the quadrature mixer circuit, which leads to a sharp decline in the receiver's image rejection capability, and the signal-to-noise ratio and dynamic range of the system are affected. This paper designs a method to calculate and compensate the linear phase imbalance of I/Q signal. ADC sampling and Fourier transform are performed on the demodulated I/Q two-way baseband signals, and then the single-side spectrum is taken to calculate the cross-power spectrum and the phase spectrum of the cross-power spectrum. The three-point method is used for phase unwrapping, the phase compensation is performed according to the unwrapped group delay difference and phase offset, and finally the time domain signal is restored. After testing, the method has good compensation effect, simple implementation, low algorithm complexity and high real-time performance.
2023, 52(4): 530-538.
doi: 10.12178/1001-0548.2022088
Abstract:
In this paper, a full-aperture linear tightly coupled active phased array antenna system with ultra-wide bandwidth and wide angle scanning performance is described. The design procedure and implementation method of the full-port feeding tightly coupled ultra-wideband wide angle scanning array without dumb element are studied. Two mirror symmetry dipole elements are fed by a Wilkinson power divider simultaneously. Extended dipole elements are utilized at the edge of array to eliminate the truncation effect and effectively reduce the active voltage standing wave ratio (VSWR) at low frequencies. Vertical metal walls are utilized at the other edge of array to replace the dummy elements. Meanwhile, vertical metal walls can also support the wide angle impedance matching (WAIM) layer. Finally, the full-aperture antenna array is realized. In order to meet the requirements of large instantaneous bandwidth and wide angle scanning performance, the radio frequency link of the T/R component uses a phase shifter plus a frequency-insensitive time delay device, so as to improve the beam pointing spatial dispersion and waveform time dispersion performance of large instantaneous bandwidth while large angle scanning. The ultra-wideband wide angle scanning array aperture, cooler plate, T/R components, beam control and power board and rear cover are integrated in high-density stack-type. A 1 × 10 full-aperture ultra-wideband active phased array is fabricated. The measurement results show that the array antenna can work across 4 octaves, the scanning range is greater than or equal to ±45°, and the instantaneous operating bandwidth can cover the full frequency band of the active phased array antenna system, which is of great significance for realizing integrated multi-function software defined satellite evolution in orbit, reducing cost and improving flexibility.
In this paper, a full-aperture linear tightly coupled active phased array antenna system with ultra-wide bandwidth and wide angle scanning performance is described. The design procedure and implementation method of the full-port feeding tightly coupled ultra-wideband wide angle scanning array without dumb element are studied. Two mirror symmetry dipole elements are fed by a Wilkinson power divider simultaneously. Extended dipole elements are utilized at the edge of array to eliminate the truncation effect and effectively reduce the active voltage standing wave ratio (VSWR) at low frequencies. Vertical metal walls are utilized at the other edge of array to replace the dummy elements. Meanwhile, vertical metal walls can also support the wide angle impedance matching (WAIM) layer. Finally, the full-aperture antenna array is realized. In order to meet the requirements of large instantaneous bandwidth and wide angle scanning performance, the radio frequency link of the T/R component uses a phase shifter plus a frequency-insensitive time delay device, so as to improve the beam pointing spatial dispersion and waveform time dispersion performance of large instantaneous bandwidth while large angle scanning. The ultra-wideband wide angle scanning array aperture, cooler plate, T/R components, beam control and power board and rear cover are integrated in high-density stack-type. A 1 × 10 full-aperture ultra-wideband active phased array is fabricated. The measurement results show that the array antenna can work across 4 octaves, the scanning range is greater than or equal to ±45°, and the instantaneous operating bandwidth can cover the full frequency band of the active phased array antenna system, which is of great significance for realizing integrated multi-function software defined satellite evolution in orbit, reducing cost and improving flexibility.
2023, 52(4): 539-548.
doi: 10.12178/1001-0548.2022142
Abstract:
The complex urban traffic environment leads to problems of low information transmission security, slow data rate, and limited ground relay in vehicular communication network. A secure transmission scheme for vehicular communication scheme is proposed with RIS-assisted (reconfigurable intelligent surface-assisted) unmanned aerial vehicle (UAV) relay cooperation. In this scheme, the UAV trajectory optimization problem is designed according to the maximum average transmission rate of the system. And then, the non-convex problem of the UAV trajectory is solved by introducing a sequential convex programming alternating iterative algorithm of relaxed variables, which yields the UAV flight trajectory of the maximizing average transmission rate. Finally, the capacity of the eavesdropping channel can be reduced by properly adjusting the position between the RIS and the eavesdropping user, and the wireless transmission environment can be improved. Numerical analysis shows that the maximum average transmission rate can be obtained by optimizing the trajectory of the UAV. With the assistance of RIS, the proposed secure transmission scheme can effectively improve the transmission rate of the communication system and ensure the safety performance though reasonable parameter settings.
The complex urban traffic environment leads to problems of low information transmission security, slow data rate, and limited ground relay in vehicular communication network. A secure transmission scheme for vehicular communication scheme is proposed with RIS-assisted (reconfigurable intelligent surface-assisted) unmanned aerial vehicle (UAV) relay cooperation. In this scheme, the UAV trajectory optimization problem is designed according to the maximum average transmission rate of the system. And then, the non-convex problem of the UAV trajectory is solved by introducing a sequential convex programming alternating iterative algorithm of relaxed variables, which yields the UAV flight trajectory of the maximizing average transmission rate. Finally, the capacity of the eavesdropping channel can be reduced by properly adjusting the position between the RIS and the eavesdropping user, and the wireless transmission environment can be improved. Numerical analysis shows that the maximum average transmission rate can be obtained by optimizing the trajectory of the UAV. With the assistance of RIS, the proposed secure transmission scheme can effectively improve the transmission rate of the communication system and ensure the safety performance though reasonable parameter settings.
2023, 52(4): 549-554.
doi: 10.12178/1001-0548.2022335
Abstract:
In order to solve the computation bottleneck of electromagnetic scattering in the extremely electrically large sea-ship scene, this paper studies the implementation technology of the shooting and bouncing rays (SBR) based on the parallel acceleration technology of multiple graphs processing units (Multi-GPU). This method utilizes the multi-process service (MPS) of compute unified device architecture (CUDA) to build the framework of Multi-GPU acceleration technique. The task division method is based on radiation beams in the projected grid region. In addition, the task division technology based on matrix grid is studied to improve maximally GPU global memory utilization. Finally, dynamic load balancing algorithm based scheduling system, which can enhance the usage of the computational resources and solve the asynchronization of computation caused by differences between GPUs has been designed. Some simulations show that compared with existing SBR method based on parallel computation technique, the speedup rate of proposed approach reach 200% with certain accuracy. Therefore, this method can effectively solve the electromagnetic scattering problem of extremely electrically large sea-ship scene.
In order to solve the computation bottleneck of electromagnetic scattering in the extremely electrically large sea-ship scene, this paper studies the implementation technology of the shooting and bouncing rays (SBR) based on the parallel acceleration technology of multiple graphs processing units (Multi-GPU). This method utilizes the multi-process service (MPS) of compute unified device architecture (CUDA) to build the framework of Multi-GPU acceleration technique. The task division method is based on radiation beams in the projected grid region. In addition, the task division technology based on matrix grid is studied to improve maximally GPU global memory utilization. Finally, dynamic load balancing algorithm based scheduling system, which can enhance the usage of the computational resources and solve the asynchronization of computation caused by differences between GPUs has been designed. Some simulations show that compared with existing SBR method based on parallel computation technique, the speedup rate of proposed approach reach 200% with certain accuracy. Therefore, this method can effectively solve the electromagnetic scattering problem of extremely electrically large sea-ship scene.
2023, 52(4): 555-561.
doi: 10.12178/1001-0548.2021323
Abstract:
Significant radiations in the frequencies ranging from gamma rays and X-rays to radio frequency can be generated during the interactions between powerful lasers with solid targets. Gamma rays are able to penetrate metal walls of diagnostic setups and induce secondary electrons, which can further radiate Electromagnetic Pulses (EMPs). In this study, emissions of secondary electrons due to gamma-metal interactions are quantitatively simulated. The results indicate that secondary ions are strongly pertinent to the incident Gamma rays, metal contents and metal shapes. The Maxwell’s equations confirm that electromagnetic pulses partially stem from the secondary charges, and its average energy and spatial distribution are also discussed. The resulting conclusions are beneficial to deepen understanding of EMP generations and also offer more experimental supports to achieve an effective shielding design for various physical diagnostics.
Significant radiations in the frequencies ranging from gamma rays and X-rays to radio frequency can be generated during the interactions between powerful lasers with solid targets. Gamma rays are able to penetrate metal walls of diagnostic setups and induce secondary electrons, which can further radiate Electromagnetic Pulses (EMPs). In this study, emissions of secondary electrons due to gamma-metal interactions are quantitatively simulated. The results indicate that secondary ions are strongly pertinent to the incident Gamma rays, metal contents and metal shapes. The Maxwell’s equations confirm that electromagnetic pulses partially stem from the secondary charges, and its average energy and spatial distribution are also discussed. The resulting conclusions are beneficial to deepen understanding of EMP generations and also offer more experimental supports to achieve an effective shielding design for various physical diagnostics.
2023, 52(4): 562-567.
doi: 10.12178/1001-0548.2022306
Abstract:
The multi wire drift chamber (MWDC) is used for track measurement of the external target experimental terminal at heavy ion research facility in Lanzhou-Cooler storage ring. The amplification chip in the front-end electronics (FEE) adopts SFE16 chip. At present, the number of single board channels in the front-end electronics is few; In addition, the slow control configuration module is obsolete and cannot be compatible with new devices, resulting in low configuration efficiency. Therefore, the FEE is upgraded in this paper, and the single board realizes 32 channels; and a configuration board based on field programmable gate array (FPGA) is designed to realize fast and efficient configuration of multiple SFE16 chips through USB interface of host computer. The electronics test results show that the upgraded FEE guarantees the original performance on the basis of increasing the number of channels, and realizes the configuration of 248 SFE16 chips at a time. It is simple to use, efficient to configure, and practical.
The multi wire drift chamber (MWDC) is used for track measurement of the external target experimental terminal at heavy ion research facility in Lanzhou-Cooler storage ring. The amplification chip in the front-end electronics (FEE) adopts SFE16 chip. At present, the number of single board channels in the front-end electronics is few; In addition, the slow control configuration module is obsolete and cannot be compatible with new devices, resulting in low configuration efficiency. Therefore, the FEE is upgraded in this paper, and the single board realizes 32 channels; and a configuration board based on field programmable gate array (FPGA) is designed to realize fast and efficient configuration of multiple SFE16 chips through USB interface of host computer. The electronics test results show that the upgraded FEE guarantees the original performance on the basis of increasing the number of channels, and realizes the configuration of 248 SFE16 chips at a time. It is simple to use, efficient to configure, and practical.
2023, 52(4): 568-577.
doi: 10.12178/1001-0548.2022248
Abstract:
Magnetic Resonance Imaging (MRI) has been extensively employed as an auxiliary means to diagnose pathological deterioration of brain, spinal cord and heart related diseases clinically. Nevertheless, imaging noise induced by both internal and external impacts restrict further improvement on diagnostic accuracy. This paper carries out a review on technological innovations ranging from earlier conventional approaches based on filter technique o state-of-the-art alternatives utilizing the deep learning network. Finally, some inductive summaries of the medical image quality assessments have been introduced. It also points out that existing deep learning methods, which rely on a large amount of data and manual annotation of medical image samples, are poorly interpretable. It is vital that clinical-oriented evaluation mechanism should be explored for clinical demands.
Magnetic Resonance Imaging (MRI) has been extensively employed as an auxiliary means to diagnose pathological deterioration of brain, spinal cord and heart related diseases clinically. Nevertheless, imaging noise induced by both internal and external impacts restrict further improvement on diagnostic accuracy. This paper carries out a review on technological innovations ranging from earlier conventional approaches based on filter technique o state-of-the-art alternatives utilizing the deep learning network. Finally, some inductive summaries of the medical image quality assessments have been introduced. It also points out that existing deep learning methods, which rely on a large amount of data and manual annotation of medical image samples, are poorly interpretable. It is vital that clinical-oriented evaluation mechanism should be explored for clinical demands.
2023, 52(4): 578-587.
doi: 10.12178/1001-0548.2022183
Abstract:
Aiming at the problem that the traditional crawler methods are easy to fall into local optima of the search and rarely consider modifying the crawling path based on historical crawling experience, a focused crawler method based on Wang−Landau (WL) sampling is proposed. This method uses the vector space model (VSM) and PageRank algorithm to evaluate the relevance and importance of links, respectively. Regional competition strategy is used to select the target link from the link set containing the topic−related links and links with potential value. Based on probability density function, the WL algorithm is used to sample the selected target links in the set, and guides the subsequent crawling of the crawler according to the historical statistical experience, so as to optimize the search path. The experimental results show that the WL-based focused crawling method can search more topic-relevant webpages than other methods in the literature, and the climbing accuracy and standard deviation of topic relevance of all downloaded pages are also significantly improved.
Aiming at the problem that the traditional crawler methods are easy to fall into local optima of the search and rarely consider modifying the crawling path based on historical crawling experience, a focused crawler method based on Wang−Landau (WL) sampling is proposed. This method uses the vector space model (VSM) and PageRank algorithm to evaluate the relevance and importance of links, respectively. Regional competition strategy is used to select the target link from the link set containing the topic−related links and links with potential value. Based on probability density function, the WL algorithm is used to sample the selected target links in the set, and guides the subsequent crawling of the crawler according to the historical statistical experience, so as to optimize the search path. The experimental results show that the WL-based focused crawling method can search more topic-relevant webpages than other methods in the literature, and the climbing accuracy and standard deviation of topic relevance of all downloaded pages are also significantly improved.
2023, 52(4): 588-594.
doi: 10.12178/1001-0548.2022119
Abstract:
Based on the artificial intelligence deep reinforcement learning algorithm, this paper proposes an intelligent selection mode with high fairness, expansibility and intelligence. On the basis of the artificial intelligence deep reinforcement learning algorithm, innovative mechanisms such as action inhibition, quadruple Q-learning (QQL) and normalized Q-value are introduced. With the research results of this paper, the IoT (Internet of Thing) terminal can more intelligently select its access or handover edge server under the principle of meeting the service delay requirements and fairness. This scheme reduces service delay, improves service response efficiency, and has good value significance for improving service security and operation management level.
Based on the artificial intelligence deep reinforcement learning algorithm, this paper proposes an intelligent selection mode with high fairness, expansibility and intelligence. On the basis of the artificial intelligence deep reinforcement learning algorithm, innovative mechanisms such as action inhibition, quadruple Q-learning (QQL) and normalized Q-value are introduced. With the research results of this paper, the IoT (Internet of Thing) terminal can more intelligently select its access or handover edge server under the principle of meeting the service delay requirements and fairness. This scheme reduces service delay, improves service response efficiency, and has good value significance for improving service security and operation management level.
2023, 52(4): 595-601.
doi: 10.12178/1001-0548.2022133
Abstract:
Long-Sequence forecasting aims to model and predict future long-term time series trends by leveraging historical knowledge and patterns and has many practical applications in various industries. To fully utilize long-time series industrial data characteristics, this paper presents an improved self-attention mechanism suitable for modeling and forecasting long sequence industrial data. Our model builds a new embedding representation learning module, combined with the pooling operations, and uses the generative inference for long-range dependency modeling and time-series signal prediction. Compared with the previous self-attention-based method, the proposed model effectively solves the problems of insufficient prediction accuracy and high training cost in long sequence prediction. Our model significantly improves long-sequence water level prediction accuracy and efficiency compared with other benchmark methods. Experiments conducted on the real-world water level data from a large-scale hydropower station proved the superior performance of the proposed model in terms of both effectiveness and efficiency over existing state-of-the-art models.
Long-Sequence forecasting aims to model and predict future long-term time series trends by leveraging historical knowledge and patterns and has many practical applications in various industries. To fully utilize long-time series industrial data characteristics, this paper presents an improved self-attention mechanism suitable for modeling and forecasting long sequence industrial data. Our model builds a new embedding representation learning module, combined with the pooling operations, and uses the generative inference for long-range dependency modeling and time-series signal prediction. Compared with the previous self-attention-based method, the proposed model effectively solves the problems of insufficient prediction accuracy and high training cost in long sequence prediction. Our model significantly improves long-sequence water level prediction accuracy and efficiency compared with other benchmark methods. Experiments conducted on the real-world water level data from a large-scale hydropower station proved the superior performance of the proposed model in terms of both effectiveness and efficiency over existing state-of-the-art models.
2023, 52(4): 602-609.
doi: 10.12178/1001-0548.2022146
Abstract:
Attacks against IoT (Internet-of-Things) infrastructure, applications and end devices have increased significantly. Typical malware in IoT generates a high volume of malicious traffic. Thus, this paper improves the malware byte sequence-based MalConv model. A malicious traffic feature-based Bi-LSTM (Bidirectional Long Short-Term Memory) model is integrated. Finally, we design a fused malware detection model applicable for end devices in IoT. The experiment results demonstrate that the fused Network Traffic-based MalConv (NT-MalConv) achieves higher detection performance with 95.17% accuracy. NT-MalConv outperforms the improved MalConv and is 10.31% better in accuracy when detecting adversarial samples.
Attacks against IoT (Internet-of-Things) infrastructure, applications and end devices have increased significantly. Typical malware in IoT generates a high volume of malicious traffic. Thus, this paper improves the malware byte sequence-based MalConv model. A malicious traffic feature-based Bi-LSTM (Bidirectional Long Short-Term Memory) model is integrated. Finally, we design a fused malware detection model applicable for end devices in IoT. The experiment results demonstrate that the fused Network Traffic-based MalConv (NT-MalConv) achieves higher detection performance with 95.17% accuracy. NT-MalConv outperforms the improved MalConv and is 10.31% better in accuracy when detecting adversarial samples.
2023, 52(4): 610-619.
doi: 10.12178/1001-0548.2022164
Abstract:
This paper proposes a sequence recommendation algorithm based on self-attention mechanism and Fourier transform, named CSFTRec. By filtering the noise in the original data, this algorithm maximizes the feature capturing ability of the self-attention mechanism on the sequence data. According to the characteristics of contrast learning, a new contrast loss is introduced on the basis of Bayesian personalized ranking for joint training, which can shorten the distance between different similar sequences. Experiments on eight public data sets show that CSFTRec converges faster and improves the recommendation accuracy by 3% to 5%, which indicates that CSFTRec is more suitable for processing sequence data.
This paper proposes a sequence recommendation algorithm based on self-attention mechanism and Fourier transform, named CSFTRec. By filtering the noise in the original data, this algorithm maximizes the feature capturing ability of the self-attention mechanism on the sequence data. According to the characteristics of contrast learning, a new contrast loss is introduced on the basis of Bayesian personalized ranking for joint training, which can shorten the distance between different similar sequences. Experiments on eight public data sets show that CSFTRec converges faster and improves the recommendation accuracy by 3% to 5%, which indicates that CSFTRec is more suitable for processing sequence data.
2023, 52(4): 620-630.
doi: 10.12178/1001-0548.2022113
Abstract:
The construction algorithm of small-world (WS) network was adopted to add randomly rewired hyperedges to the hypernetwork (also called hypergraph) to construct a small-world hypernetwork model. This model was used as the underlying network for information propagation, and a small-world SIR information propagation model in hypernetworks was proposed. A simulation was conducted to investigate the influences of hyperedge rewiring probability, number of neighboring nodes in the hypernetwork, and propagation and recovery rates on the hypernetwork's information propagation process. A comparative analysis is performed with the information propagation process in a normal network, and it is found that information spreads faster and reaches a wider range in hypernetworks. Furthermore, the small-world characteristics and information propagation rules of hypernetworks are verified on three types of empirical hypernetworks. In the era of everything interconnected, where the world is becoming smaller, there is certain reference significance for the in-depth study of more complex information and disease propagation mechanisms in the real world.
The construction algorithm of small-world (WS) network was adopted to add randomly rewired hyperedges to the hypernetwork (also called hypergraph) to construct a small-world hypernetwork model. This model was used as the underlying network for information propagation, and a small-world SIR information propagation model in hypernetworks was proposed. A simulation was conducted to investigate the influences of hyperedge rewiring probability, number of neighboring nodes in the hypernetwork, and propagation and recovery rates on the hypernetwork's information propagation process. A comparative analysis is performed with the information propagation process in a normal network, and it is found that information spreads faster and reaches a wider range in hypernetworks. Furthermore, the small-world characteristics and information propagation rules of hypernetworks are verified on three types of empirical hypernetworks. In the era of everything interconnected, where the world is becoming smaller, there is certain reference significance for the in-depth study of more complex information and disease propagation mechanisms in the real world.
2023, 52(4): 631-640.
doi: 10.12178/1001-0548.2022111
Abstract:
In order to improve the performance of existing higher-order community detection algorithms, a higher-order community detection algorithm based on motif-based modularity optimization is proposed. By quantifying the number of motifs as the weight between nodes, the higher-order community detection based on motifs is transformed into lower-order weighted network community detection based on edges, and a weighted modularity optimization problem is constructed. Based on the meta-heuristic algorithm as the optimization strategy, the lower-order topology structure and higher-order weight information are comprehensively utilized to design the neighborhood community modification operation and local search operation of nodes, so as to improve the quality of community partitions and prevent the algorithm from falling into local optimum. Experimental results on synthetic and real-world networks show that the utilization of motifs is helpful to improve the detection performance under the condition of fuzzy community structure. The proposed algorithm can effectively realize motif-based community detection and has certain advantages in accuracy and quality compared with existing typical motif-based algorithms, which helps to deepen the understanding of the higher-order structure and functional characteristics of complex networks.
In order to improve the performance of existing higher-order community detection algorithms, a higher-order community detection algorithm based on motif-based modularity optimization is proposed. By quantifying the number of motifs as the weight between nodes, the higher-order community detection based on motifs is transformed into lower-order weighted network community detection based on edges, and a weighted modularity optimization problem is constructed. Based on the meta-heuristic algorithm as the optimization strategy, the lower-order topology structure and higher-order weight information are comprehensively utilized to design the neighborhood community modification operation and local search operation of nodes, so as to improve the quality of community partitions and prevent the algorithm from falling into local optimum. Experimental results on synthetic and real-world networks show that the utilization of motifs is helpful to improve the detection performance under the condition of fuzzy community structure. The proposed algorithm can effectively realize motif-based community detection and has certain advantages in accuracy and quality compared with existing typical motif-based algorithms, which helps to deepen the understanding of the higher-order structure and functional characteristics of complex networks.
