Abstract:A self-supervised graph embedding approach based on hierarchical projection network called MeghenNet(multi-view heterogeneous graph projection network) was introduced to learn low-dimensional representations from multiple views. The concept of multiple-view heterogeneous graphs was defined to explicitly allow the model to simultaneously collect information from multiple data sources for modeling heterogeneous graphs. A hierarchical attention projection that involves a cross-relation projection to extract semantics information within each view was employed, followed by a cross-view projection to aggregate contextual information from other views. The mutual information loss function between each view embedding and the global embedding was computed to ensure the information consistency across views. Experimental results on several real-world datasets demonstrate that the proposed method outperforms state-of-the-art approaches when handling multi-view heterogeneous graphs.

Abstract:Existing research on sequential ad recommendations mainly focuses on user preferences for advertisement, insufficiently considering positive relationships between ads. Starting from the associations between ads, incorporates both ad networks and user networks into consideration, a multi-round social advertising influence maximization model based on triggering model was constructed. An ad edge based greedy strategy based on multi-round reverse influence sampling was proposed to enhance platform revenue, with theoretical proofs of its strict lower bound guarantee. Experiments show that compared to existing optimal methods, the proposed method increases the average ad propagation influence revenue by 35%, significantly enhancing ad recommendation effectiveness, providing a new solution for ad sequence recommendations.

Abstract:In order to reduce the marketing costs of merchants promoting products over multiple rounds on social networks,this study made a exploration on the selection of boosting nodes during the process of multi-round influence propagation. Based on the model of multi-round influence boosting propagation mode, an adaptive strategy for choosing boosting nodes was designed. Given known seed nodes, this strategy could find an efficient method to minimize the number of marketing rounds needed to reach a certain threshold of social influence, with nearly linear algorithmic complexity. Experimental results show that compared to existing heuristic algorithms and non-adaptive algorithms, the designed adaptive strategy can reduce the promotion rounds required to reach a specified threshold by 7.3%~18.3%, effectively reducing the promotion cost.

Abstract:Addressing the issue of anomaly detection on missing multivariate time series data in real IoT（Internet of things） environments, a novel method on multivariate time series anomaly detection algorithm intergrated with graph embedding of missing information was proposed. Using a joint learning framework of pre-interpolation and anomaly detection task fusion, a GNN（graph neural network） pre-interpolation module based on time series Gaussian kernel function was designed to realize the joint optimization of pre-interpolation and anomaly detection task. A graph structure learning method for embedding missing information in time series data was proposed, using graph attention mechanism to fuse missing information masking matrix and spatiotemporal feature vectors, effectively modeling the potential connections of missing data distribution in multivariate time series. The performance of the algorithm was verified on real IoT sensor datasets. Experimental results prove that the proposed method significantly outperform the mainstream two-stage methods on the task of missing multivariate time series anomaly detection. The comparative experiment of the pre-interpolation module fully prove the effectiveness of the GNN pre-interpolation layer based on the Gaussian kernel function.

Abstract:Traditional graph data models lack explicit temporal dimension representation, which may lead to complex temporal queries and potential loss of temporal information integrity. To address this limitation, a temporal property graph data model and a corresponding temporal graph query language called S-Cypher were proposed. The temporal graph data model represents utilized object nodes to represent entities, and introduced property nodes and value nodes to represent entity properties. Valid time was recorded on nodes and edges between object nodes to express temporal information, and the recorded valid time adhered a set of temporal constraints. S-Cypher served as a temporal extension to Cypher, ensured compatibility while providing a concise and comprehensive temporal graph query syntax, including temporal data types, temporal graph pattern matching, time window constraints, and temporal paths. An implementation scheme for executing S-Cypher temporal graph queries on Neo4j was also provided. Experimental results demonstrate that the query time of S-Cypher is on average 1.29 times that of Cypher, indicating that S-Cypher can effectively manage temporal graph data in Neo4j with satisfactory performance.

Abstract:To address the aerodynamic load reduction requirement when the launch vehicle flying in high wind zone during the ascending phase, an intelligent attitude control method with adaptive learning rate was proposed. Taking a certain type of launch vehicle as the research object, the dynamic model in the pitch plane was established. A deep reinforcement learning framework suitable for flight control of the launch vehicle during the ascending phase was developed based on soft actor-critic, and a reward function that comprehensively considers attitude tacking accuracy and stability, and load reduction effectiveness was designed. On this basis, an adaptive iteration of learning rate was implemented based on a step-size learning rate scheduler to quickly improve the convergence velocity and find the optimal solution of the controller. Besides, an early stopping mechanism which can automatically end the training process was designed to enhance the training efficiency. Simulations show that the proposed method can effectively achieve load reduction of the launch vehicle while ensuring attitude tracking accuracy and stability. Additionally, it has strong robustness and adaptability to random wind disturbance.

Abstract:For the time-varying trajectory tracking control problem of the relative motion configuration of spacecraft formation, a distributed coordinated control method was designed based on the PH(port-Hamiltonian) model and generalized canonical transformation, using IDA-PBC（interconnection and damping assignment passivity-based control） algorithm.Through the PH modeling and generalized canonical transformation of the linear dynamics of spacecraft formation, the trajectory tracking error PH system was obtained. Under the assumption that the topological structure is connected and fixed, a distributed coordinated control method for formation trajectory tracking considering relative errors between neighboring spacecraft was derived based on IDA-PBC algorithm and using the PH model of relative motion errors. Numerical simulation verifies the effectiveness of the control method. The results show that the PH method can complete the trajectory tracking control of spacecraft formation, which provides a new effective method for the distributed coordinated control of spacecraft formation.

Abstract:In order to intensive study the self-excited vibration phenomenon of rocket in the flight test, a two-degree-of-freedom linear system based on the coupling of elastic force and aerodynamic force was established, and the working mechanism of self-excited vibration caused by the displacement feedback between pitch and yaw degrees of freedom was explained. By analyzing the stability of the motion equation of the system, the criterion of losing asymptotic stability of equilibrium point and generating self-excited vibration was obtained. The time history curves of pitch and yaw angular displacement were obtained by solving ordinary differential equations with numerical method. The comparison between the simulation results and the measured data shows that the sudden amplification of vibration divergence phenomenon during the flight of rocket is the mode-coupling self-excited vibration caused by displacement feedback. In addition, the analysis results show that the beat or constant frequency vibration usually occurs before and after the self-excited vibration.

Abstract:Aiming at the problems that traditional effectiveness evaluation methods can not reflect the evolution, emergence and adaptability of the anti-missile equipment system, a data-driven effectiveness evaluation method of anti-missile equipment system was proposed. Based on the analysis of the characteristics of anti-missile equipment system and the shortage of traditional effectiveness evaluation method. the Bayes optimization algorithm was used to optimize the convolutional neural network hyperparameters, and the efficiency evaluation model ofBayes-CNN(Bayes convolutional neural network) was constructed. The flow and steps of Bayes-CNN system effectiveness evaluation algorithm were studied, and a set of completed efficiency evaluation algorithm was formed. Designed and validated the simulation experiment, input a lot of test data to Bayes-CNN model for training and learning, so as to obtain the simulation prediction of the effectiveness of anti-missile equipment system. The experimental results show that the error between the actual and expected output is very small, and the non-linear fitting effect is great so that it had a high degree of feasibility and reliability.

Abstract:Since the thrust characteristics directly affect the launch speed, altitude and flight range of the water rocket, in order to improve the thrust performance of the water rocket, a flexible and deformable self-pressurized elastic air pressure cabin scheme was designed on the basis of the existing fixed volume pressure chamber, and its performance was evaluated. With Bernoulli′s theorem and deformation coordination relation, a coupled model of internal pressure, nozzle velocity and thrust of water rocket was established. Moreover, the influence of different initial states (water volume ratio and inflation air pressure) on water rocket thrust was also studied with the numerical calculation method, and the thrust difference was compared and analyzed further between the fixed air pressure cabin and the elastic air pressure cabin under the same initial conditions. The research results show that the improved flexible and deformable self-pressurized elastic air pressure cabin can effectively increase the water jet velocity during the launch, and the thrust generated by the water rocket in the same initial state increases significantly by 46.95%. The designed scheme can provide important reference for improving the flight performance of water rockets and the optimal design of new flexible deformation water rocket scheme.

Abstract:An anti-off-target control method for small video satellite target tracking based on visual field zoning was proposed to address the problem of large initial relative attitude deviation and angular velocity between small video satellite and observation targets, which can easily deviate from the camera field of view and cause off target. In this method, the rectangular imaging field of the satellite-carried camera was divided into inner and outer parts according to the tangent circle. The tracking controller of the inner and outer parts of the tangent circle was designed based on the potential function and the quasi-Euler rotation method respectively. By using the Barbalat′s lemma, the asymptotic stability of control laws in both regions was proved, and it was also proved by theory that the potential function controller can ensure that target would not leave the inscribed circle of visual field after entering it. Through controller comparison and simulation, the results show that the quasi-Euler rotation method has a stronger ability to suppress target deviation from the field of view compared with PD control. Compared with the full quasi-Euler rotation method, the field of view partition control combining the quasi-Euler rotation method and potential function method can effectively achieve anti-off-target control for faster maneuvering targets, thereby achieving continuous tracking and observation.

Abstract:Aiming for missile trajectory planning, an applicable Gym training evironment was established. An intelligent agent network structure and its reward functions were designed based on twin delayed deep deterministic policy gradient framework and according to terminal and process constraints, forming an intelligent trajectory planning method. Through deploying the algorithm on an embedded GPU computing acceleration platform, bias simulation and comparison tests were conducted. The results show that the method can reach the requirements of missile capability and process constraints under different range tasks and effectively overcome environmental disturbances with adaptability to distinct object models. Meanwhile, the method has an extremely fast calculation speed, far surpassing the popular GPOPS-Ⅱ toolbox. The computation time for single step trajectory command is less than a millisecond so that it can support real-time online trajectory generation, which provides an effective implementation path and technical support for engineering applications.

Abstract:Based on the differential impact of the spatial distribution of impact points on target damage effectiveness, the probability distribution model of missile hitting different important areas of target was constructed to realize expansion of the traditional hit accuracy concept. Aimed at the reality of actual missile hitting targets with complex process, high cost and low frequency, Bayesian method was used to fuse multi-source information, and missile hit accuracy was estimated on the basis of the idea of region division, distribution determination, prior fusion and posterior solution. The Dirichlet distribution was selected as the prior distribution of hit accuracy parameters, the D-S(Dempster-Shafer) evidence theory was used to fuse the prior information and the posterior distribution of accuracy parameters was solved by MCMC(Markov chain Monte Carlo) method. Example results show that this method can describe the probability of missile hitting different important areas of the target in detail, and scientifically integrate multiple types of prior information about hit accuracy, which provides theoretical references for missile hit accuracy estimation methods and test schemes optimization.

Abstract:To address the resource consumption issue of obtaining precise paired samples in existing fully supervised learning, while also considering the quality of network map generation, a novel semi-supervised online map generation model based on generative adversarial networks was proposed, which aimed to realize the direct generation of intelligent remote sensing images into network maps by using only a few precisely matched data and a large amount of unpaired data. In addition, a semi-supervised learning strategy based on transformation consistency regularization and sample enhanced consistency was designed, which overcomed the inconsistency problem caused by imprecise paired data and derives better generalization performance of the model. Adequate comparison experiments were conducted on different map datasets. The generated online maps outperform the competing methods on the quantitative metrics and visual quality, which validate the effectiveness and speed of semi-supervised network map generation methods.

Abstract:Aiming at the problems of large estimation error and poor anti-interference ability in state estimation and parameter learning of time-varying nonlinear systems, a batch state estimation and parameter learning method for accurate sparse Gaussian variational inference for nonlinear systems was proposed. A loss function was proposed based on Gaussian variational reasoning, and the state estimation problem was transformed into an approximation problem to the true posterior, and parameters that need to be learned were introduced. The parameters of the state probability distribution were iteratively updated using the Gauss-Newton optimizer method, and a complete state estimation iterative scheme was obtained by using Steins lemma, the sparsity of the covariance matrix and the Gaussian volume method. The noise parameters of the measurement model were learned through expectation maximization, and the inverse Wishart prior was introduced to reduce the influence of measurement noise and outliers on parameter learning and state estimation results. The simulation experiment was carried out on the UAV simulation model, and the UAV trajectory can be accurately estimated without adding the UAV movement and the real value of the measurement noise, and the impact of measurement noise and measurement outliers on trajectory estimation accuracy is effectively suppressed.

Abstract:In order to improve the prediction accuracy of the displacement and deformation of reservoir, the displacement and deformation of non-linear and non-stationary reservoir was predicted by changing the decomposition method of VMD(variational mode decomposition) and integrating VMD and long short-term memory. A MVMDLSTM (mixed variational mode decomposition long short-term memory) model prediction method was proposed. The reliability of the new method was verified with multi-source datasets for different single prediction models and combined models.The experimental results show that the MVMDLSTM model can effectively attenuate the bias of the single prediction model and the empirical mode decomposition combination model estimation, and the prediction accuracy of the MVMDLSTM model is better, which provides an effective data decision-making for the stable monitoring of the prediction and warning of the reservoirs slow sliding and creeping and other small deformations.

Abstract:In the few-shot recognition scenario of space targets observed at low frequency, the drastic changes in the image representation of space targets in different poses challenges to the extraction of discriminative features and the correlation of features between images. To address these issues, the few-shot space target recognition method based on adaptive cross fusion of local features was proposed. Based on the existing few-shot learning framework, the feature cross fusion module based on self-attention and cross-attention was used to adaptively learn the correlation between local features, improve the discriminant and robustness of feature in different poses, effectively explore the similarity between the support set and the query set, and improve the accuracy of feature association with representation differences. Meanwhile, the sample label weight based on neighborhood density was employed into the loss function to solve the learning bias problem of the network model caused by unbalanced space target datasets. Through the verification on different datasets, the proposed method is proved to achieve higher recognition accuracy．

Abstract:In order to solve the problem that many traffic flow prediction research methods are unable to comprehensively explore the dynamic hidden correlations in traffic data, the dynamic spatio-temporal variation characteristics were studied and an encoder-decoder-based traffic prediction model was proposed. In the model, both encoder and decoder mainly consisted of multi-head spatio-temporal attention mechanism modules, and a connection attention mechanism was added in between to analyze the spatio-temporal correlations of the road network. The model also used a dynamic embedding module consisting of a combination of both spatio-temporal embedding coding and adaptive graph convolution to analyze the dynamic and static information of nodes. Experiments on two real datasets demonstrate that the spatio-temporal model outperform other models for long-and short-term traffic prediction. Thus, the spatio-temporal encoder-decoder model can effectively handle complex spatio-temporal sequences and improve the traffic flow prediction accuracy.

Abstract:Based on the impressed current cathodic protection system and its current output mode, in order to achieve the purpose of taking into account the underwater electric field stealth on the basis of hull anticorrosion function, the “integrated” method was adopted to replace the traditional “zonal” anodic current control method, and the current adjustment strategy based on underwater electric field and reference potential measurement information was proposed. The main idea is to transform the cathodic protection current optimization problem with respect to natural corrosion state into the minimization problem of the electric field increment under water with respect to electric field stealth. The effectiveness of the proposed current optimization control method was verified through the simulation calculation and shrinkage ratio model test.The results show that when the peak-peak potential is taken as the evaluation criterion and the whole ships electrical potential meets the anticorrosion requirements, the peak-peak potential of the 1.0B depth measurement plane can be reduced by more than 20% compared with that in the natural corrosion state of the ship. Moreover, the proposed method can quickly calculate the output current of each anode required by the impressed current cathodic protection system for both electric field stealth and hull anticorrosion.

Abstract:FG-CNTRC demonstrate significant engineering value in advanced equipment manufacturing due to their exceptional mechanical properties and designable characteristics. The critical problem of nano-reinforcement scale effects on mechanical response mechanisms was addressed through integration of nonlocal theory with the Eshelby-Mori-Tanaka method, resulting in the development of a nano-to-macro multiscale constitutive model. Based on mathematical characterization of spatially gradient-distributed CNTs(carbon nanotubes), the thermo-mechanical coupling effects from environmental temperature and visco-Pasternak substrates were incorporated. Vibration governing equations for nanocomposite structures were established through Kirchhoff plate theory and energy variational principles, with characteristic frequencies of simply-supported plates subsequently solved. The influence mechanisms of CNTs′ characteristic parameters and thermo-mechanical coupling effects on the natural frequency of structural systems was analyzed. Results demonstrate that the constitutive model effectively characterizes the stiffness-weakening effect induced by CNTs′ scale effects. This effect simultaneously suppresses the stiffness enhancement from substrate elastic parameters while significantly increasing sensitivity to temperature variations. Moreover, the critical volume fraction for structural reciprocating vibration shows positive correlation with substrate damping parameters.

Abstract:Aiming at the reliability-based design optimization problem of complex structural systems, an efficient optimization method based on subset simulation and Markov chain simulation in augmented space was proposed. Considering the reliability-based design optimization problem in which the design parameters were distributed parameters of basic random variables, the target failure probability was transformed into a posterior density function of the design parameters in the augmented space, obtained a set of initial failure samples in the whole design domain through subset simulation, and then adopted the efficient Markov chain simulation to generate more failure samples in the gradually smaller design domain under the sequential approximate optimization framework. The target posterior density function was estimated and updated, and the decoupling approach was used to solve the transformed optimization problem to finally obtain the optimum. Compared with the existing methods, the proposed method requires only one reliability analysis and can avoid local optimal solution, resulting in the global optimal solution. Examples were given to illustrate the applicability of the proposed method in engineering and its superiority in the accuracy and efficiency of analysis and calculation.

Abstract:To thoroughly investigate the underwater motion characteristics of rod jets, the effects of liner thickness, material, and charge length-to-diameter ratio on the underwater motion characteristics of rod jets were systematically explored by combining experimental methods with numerical simulations. The results show that after entering the water, the rod jet undergoes head upsetting and experiences mass erosion effects. The effective length of the jet initially increases and then decreases during its motion, while its average velocity decays exponentially. Further analysis indicates that increasing the liner thickness and charge length-to-diameter ratio can significantly enhance the jets resistance to erosion and its ability to maintain velocity. The optimal range for liner thickness is 0.036D_k to 0.055D_k. When the charge length-to-diameter ratio exceeds 1.25, the influence of charge structure on the underwater motion characteristics of the rod-shaped jet gradually diminishes. Additionally, material density has a significant impact on the velocity decay law of the rod jets during underwater penetration: the higher the density, the stronger the jets ability to maintain velocity; when material densities are similar, the velocity decay laws of the jets tend to be consistent. The study also demonstrates that liners made of copper, tantalum, and tungsten are all suitable for underwater shaped-charge warheads. This research provides important theoretical support and reference for the design optimization of underwater shaped-charge warheads.

Abstract:In order to comprehensively and scientifically evaluate the performance of suspension system in high-speed maglev train, a comprehensive evaluation method was designed based on fuzzy analytic hierarchy process. According to the operation law and capability characteristics of suspension system, an evaluation index system was proposed to characterize the performance of suspension system. The fuzzy analytic hierarchy process was used to determine the weight of the performance index of the suspension system, and a multi-layer fuzzy comprehensive evaluation model of the suspension system was established based on exponential scaling method and trapezoidal distribution method. A comprehensive performance evaluation system of suspension system based on operation data was developed, and the dynamic operation test was carried out on the Shanghai 1.5 km test line by using a 600 km/h high-speed maglev train. Results show that the comprehensive evaluation method of suspension system based on fuzzy analytic hierarchy process can comprehensively evaluate the performance of suspension system and realize the quantification and visualization of the evaluation results.

Abstract:Continuous-wave LDV (laser Doppler velocimeters) are limited in low-altitude velocity measurement due to factors such as the low power of continuous-wave lasers and optical system diffraction. By introducing the concept of "virtual distance" to expand the Feuilleté model, the time-domain echo signal model for pulsed LDV had been established. Simulation results indicate that pulsed LDV can perform velocity measurement through the accumulation of hard target echo signals. Pulsed LDV can also utilize longer laser pulse widths for detection without restricted by spatial resolution, the feasibility of high-precision velocity measurement using long pulses with pulsed LDV is verifies, laying a theoretical foundation for future experimental validation of the pulsed LDV. Pulsed LDV is capable of detecting echo signals scattered from targets at 5 km and beyond, significantly extending the working distance range of LDV. This makes LDV applicable in integrated navigation of low-altitude aircrafts and planetary surface landing navigation for spacecrafts, and other scenarios that require long-distance high-precision velocity measurement.