Abstract:Time series anomaly detection is a key technology relied upon in applications such as network service, data security, and system monitoring. In order to address the limitations of effectiveness, rationality and stability in anomaly detection results caused by the fuzziness of time series context, complexity of data distribution, and the uncertainty of anomaly detection models in practical scenarios, this paper proposes a new anomaly detection method (AdcGAN), based on contextual generative adversarial network. Firstly, AdcGAN extracts conditional context for generating time series data by processing historical data. Secondly, AdcGAN constructs a context generative adversarial network following conditional generative adversarial network strategy to achieve conditional distribution prediction of the data at any moment. Meanwhile, AdcGAN uses Dropout to approximate model uncertainty and replacing point estimates with probability distribution as prediction result. Then, anomalies are measured based on the differences in observations (represented by the expected deviations) and the uncertainty of the model (represented by prediction variances). Finally, an automatic method for setting anomaly thresholds based on statistical information of the time series data is proposed to reduce the number of manually adjusted parameters. Our experimental results on 47 real-time series data of the NAB dataset compared with baselines show that, compared to similar benchmark algorithms, the proposed AdcGAN method can effectively detect anomalies in time series data. It outperforms other benchmark methods in most evaluation metrics and achieves better stability.

Abstract:Due to the overlapping functionality between communication devices and radar systems, communication devices can also be utilized for target detection. However, traditional communication protocol adopts explicit feedback to transmit information, leading to the failure of communication equipment detection to retain perceptual information. Therefore, the decomposition process of explicit feedback can be improved, so as to realize the awareness of environmental objectives in the communication gap. In MIMO systems, beamforming is a technology used at the transmitter to generate a steering matrix based on MIMO channel information, thereby enhancing the channel performance at the receiver. At present, display feedback beamforming is the mainstream approach of beamforming in communication. However, the feedback matrix obtained by the traditional decomposition method usually loses some important information, which limits the functionality of using Wi-Fi signals for target detection. It is found that the eigenvector obtained by the traditional singular value decomposition method can not retain the distance and velocity information of the target. In this paper, a novel decomposition method is proposed based on the covariance matrix of the channel matrix. Firstly, the left singular value matrix is obtained, and then an improved right singular value matrix is further obtained by using the corresponding relationship between the orthogonal vectors of the two unitary matrixes. The improved result retains the distance, speed and angle information of the target, and greatly improves the effectiveness of the feedback information. Finally, the effectiveness of the proposed method is verified through simulation.

Abstract:To address the problem of high-precision autonomous visual positioning of asteroid probes, this paper proposes an autonomous visual positioning method for deep space probes incorporating orbital dynamics, which can correct the positioning errors of the simultaneous localization and mapping (SLAM) algorithm. The method enables high-precision visual navigation of the probe and the creation of a dense 3D model of the asteroid surface in a scenario, where there is no prior information on the surface and no manual markers by incorporating orbital dynamics for orbit improvement. Firstly, the visual simultaneous localization and mapping method (VSLAM) is used to extract the asteroid surface features, estimate the probe attitude utilizing a factorisation algorithm, and design loopback detection to improve the localisation accuracy. Secondly, we reconstruct the 3D model of the surface of the planet and realise the irregular gravitational field modelling of the planet based on the polyhedral method. Finally, a pseudo-relative motion orbit optimisation algorithm based on orbital dynamics is proposed as a physical constraint to correct the accumulated visual positioning error, and the propagation process of the inverse visual initial orbiting error in orbital dynamics is analysed to correct the accumulated visual positioning error and improve the initial positioning results. The experimental simulation results show that the fused orbital dynamics can effectively improve the accuracy of visual positioning for asteroid detection, thus realising high-precision navigation and providing a reference for the future development of deep space exploration technology.

Abstract:Motor power calculation plays an important role in ensuring production safety, formulating rolling schedule, and leveraging equipment capabilities. In order to improve calculation accuracy, a motor power prediction model combining mechanism model and data-driven method was proposed. The mechanism model was analytically solved by upper bound method, and then parts of the power analytical solutions (plastic deformation power, tension power, and shear power) and related variables were input into long short-term memory (LSTM) network for training, so as to extract depth characteristics and temporal correlations between variable data. Results show that the relative error of the proposed time series coupling prediction model was less than ±3.9%, and the prediction effect was obviously better than that of the upper bound method and artificial neural network (ANN) when the rolling schedule of adjacent strips changed greatly. According to the residual distribution histogram, the residual of upper bound method was concentrated around ±50 kW and deviating from a normal distribution, as it ignored the strong relationship between friction coefficient, loss power, and other parameters. Both ANN and the proposed model basically followed normal distribution, but the proposed model had higher fitting accuracy. In comparison with broader model evaluation indicators, the proposed model had better comprehensive prediction performance. Besides, the relationship between motor power, rolling length, and rolling speed was analyzed based on model output results, which indicated that the model not only has high prediction accuracy, but also strong consistency with known physical law between parameters.

Abstract:With the rapid development of micro-videos, the task of micro-video event detection is receiving more and more attention. Existing micro-video event detection studies commonly use deep neural networks to obtain definitive detection results. But these networks that ignore the effect of uncertainty may lead to false predictions yielding definitive results. To address these problems, in this paper, a micro-video event detection method with multimodal uncertainty network was proposed. Firstly, the proposed method embeds a variational layer in a traditional domain separation network, which was used to obtain predictive distributions. Then the visual modal information and the acoustic modal information was fed into the network, and the independence and correlation losses were constructed to obtain visual-audio shared domain predictive distributions and visual-audio private domain predictive distributions. Finally, an uncertainty discriminant was proposed to filter the prediction distribution of the four domains, so as to get the final prediction results. The experiments were performed on the public dataset(UCF-101 and HMDB51) and the newly constructed micro-video event detection dataset. Experimental results show that the proposed method not only has higher classification accuracy on different datasets but also can estimate the uncertainty of the output results. It also shows robustness against audio interference.

Abstract:To address the issues of not considering chromatic domain information and low accuracy in current video froensic compression methods, a surveillance video recompression forensics method for the latest versatile video coding (VVC) standard based on the information fusion of chroma domain and luminance domain, referred to as CLF-SVRF, is proposed. Based on the coding principles of the VVC standard, the basic bitstream characteristics in VVC video bitstream that are closely related to compression time are analyzed and determined from the dimensions of chroma domain and luminance domain in surveillance videos. The basic bitstream characteristics include the partitioning type and prediction mode of coding unit (CU) in chroma domain and luminance domain. Combining with the Lagrangian rate distortion optimization technique, the variations in the partitioning type and prediction mode of CU in chroma domain and luminance domain as the compression time increases are analyzed. It is further determined that the partitioning type and prediction mode of CU in chroma domain and luminance domain can be used as basic bitstream characteristics for detecting the video compression time. Then, considering the requirement of video surveillance applications for low complexity forensics methods, a low complexity advanced bitstream characteristics is constructed based on the partitioning type and prediction mode of CU in chroma domain and luminance domain. The advanced bitstream characteristics are input into the support vector machine to complete the recompression forensics of surveillance videos. The experimental results show that compared with the current advanced methods, the CLF-SVRF method can improve the accuracy of surveillance video recompression forensics by 13.53% on average. At the same time, it can significantly reduce the time required for forensic recompression, and reduce the recompression forensics time by 47.42% on average.

Abstract:The hole at the joint and boundary is an important component of the membrane structures and plays a key role in connecting the relevant components. However, the stress concentration is easy at the edges of holes, which has a serious effect on tearing behavior of the coated fabric. To investigate the effect of holes in polyvinyl chloride (PVC) coated fabric on crack propagation behavior. Uniaxial tensile tests of PVC-coated fabric with cracks and holes were carried out. The whole process of tear propagation and tearing mechanism are analyzed. A prediction model for the tear strength of PVC-coated fabric with multiple defects is proposed and the applicability of the model is verified. The results show that there are two stages of crack-end expansion and hole-end re-expansion during the center tearing process of the PVC-coated fabric with cracks and holes. Four tear characteristic indicators are defined to measure the tear resistance of coated fabric with cracks and holes, namely, the critical tearing load at the crack-end and hole-end, the ultimate tearing load at the crack-end and hole-end. The existence of holes leads to the increase in the bearing capacity of the specimen at the re-expansion stage. However, with the size of the hole increasing, the bearing capacity enhancement effect of the hole-end weakens and the ultimate tearing load at the crack-end becomes the maximum tearing load. The proposed quadratic functionexponential function stress field model can effectively predict the critical tear stress at the crack-end and the critical tear stress at the hole-end of the coated fabric.

Abstract:In order to solve the problems of poor real-time performance, significant tracking errors and poor robustness to clutter changes with maneuvering multi-target tracking with nonlinear measurement, a multi-model δ-generalized labeled multi-Bernoulli (δ-GLMB) filter is proposed in light of the random finite set (RFS) theory, capitalizing on measurement transformation and a fuzzy algorithm. Initially, the interactive multiple model (IMM) δ-GLMB filter realizes the unbiased conversion of position measurement from polar coordinate system to Cartesian coordinate system by initiating an uncorrelated unbiased measurement conversion and removes the filter estimation deviation caused by the correlation between measurement errors and their covariances on the basis of the predicted value, thus realizing the maneuvering multi-target tracking in nonlinear scenarios. Then, the number of clutter measurements is reduced by constructing a joint gate that takes into account track and measurement correlation and target maneuver constraints. Finally, an improved fuzzy algorithm is introduced, which takes the posterior model probability of the target as the input, to adaptively adjust the process noise of the motion model according to the separation degree of the model, thereby increasing filtering accuracy. The research result shows that in the clutter environment, compared with CKF-JMS-δ-GLMB, CKF-IMM-δ-GLMB, and other nonlinear filters, the proposed algorithm performs better in terms of computation time, tracking accuracy and robustness. The proposed algorithm sidesteps the computational burden typically associated with traditional nonlinear processing methods, and has better clutter suppression characteristics, which improves the performance of maneuvering multi-target tracking with nonlinear measurement.

Abstract:In view of the phenomenon of violent galloping and line-to-line discharge of the additional conductors of the overhead contact system (OCS) of the Lanzhou-Xinjiang high-speed railway in the strong wind environment, according to the structural characteristics of the additional conductors of the overhead contact system, a finite element calculation model of the additional conductors is established, and the type finding calculation of the additional conductors is carried out. The random wind field at the location of the additional conductors in the overhead contact line is simulated using the harmonic superposition method. Wind loads are applied to the conductor model. The unconditional stable Newmark method and load increment method are used to analyze the influence of line parameters on additional conductors galloping and line distances. Corresponding measures to prevent galloping are proposed. The results show that a span of 40 m or smaller has an obvious inhibitory effect on the galloping of the conductors under both low and high wind speeds. Increasing the tension of the two conductors to 6.5 kN and the damping ratio of the conductors to 1.5% or higher can effectively reduce the galloping amplitude of the two conductors. When the span of OCS additional conductors is too large, and the initial tension and damping ratio are too small, it can lead to excessive tension at the suspension point during galloping, increasing the risk of conductor disconnection or breakage accidents. With the decrease of span and the increase of operating tension and damping ratio of additional conductors, the minimum distance between lines shows an increasing trend. Under different line parameters (span, operating tension and damping ratio), the minimum distance between lines decreases with the increase in wind speed. This research provides an effective theoretical basis for the prevention and control of additional conductors galloping and discharge between lines of OCS in the strong wind sections of the Lanzhou-Xinjiang high-speed railway.

Abstract:This paper focuses on the problems of computing correspondences among non-rigid 3D shape collections with a low accuracy rate, poor consistency, and difficult bijectivity. The novel approach proposed in this paper is based on the optimized spectral alignment algorithm and the canonical consistent latent basis. Firstly, we use the adjoint operator derived from the functional map to align the information between shapes in the spectral domain. The functional map matrix of each shape pair in the shape collections is calculated, resolving the problem of inconsistent direction between functional map and pointwise map. Secondly, we adapt the improved collections of shape maps approach to assign corresponding weights to the functional map matrix of each shape pair, reducing the impact of initialization parameter noise on the shape collection matching calculation results. Finally, we add the canonical consistent latent basis of the limit shape to compute the shape collections correspondence. The limit shape can be seen as a type structure of all shapes in the shape collection, which is an intermediate model with geometric variability, improving the consistency and bijectivity of the algorithm. The experimental results show that compared with the existing algorithms, this algorithm has the lowest geodesic error and the highest accuracy of global correspondence on FAUST, SCAPE, TOSCA, and SHERC’16 Topology datasets. Meanwhile, our method can reduce the noise of initialization parameters, solve the symmetric ambiguity problem, and more accurately compute the consistent and bijective correspondence of non-rigid 3D shape collections.

Abstract:In order to investigate the changes in rock cutting force and wear of disc cutters under sliding states, a rolling circumferential cutting model considering both the rotation of hob and the revolution of disc cutters is established based on the discrete element method. A slip ratio parameter η is defined to describe the sliding state of disc cutters. The force and wear of disc cutters are compared and analyzed under different slip rates η, and the numerical simulation conclusion is verified with an engineering example. The results show that the vertical force FV and rolling force FR in the numerical simulation fluctuate near the calculated value of CSM model, which are in good agreement, indicating the rationality of the model in this paper. The numerical simulation results show that with the increase of slip ratio η, the vertical force FV decreases slightly and the rolling force FR increases significantly. From cutting rock in rolling state to cutting rock in sliding state, the vertical force FV decreases by 23.6% and the rolling force FR increases by 83.7%, indicating that cutting rock under sliding state will lead to flat wear of disc cutters. The engineering data show that the increase of thrust is the main manifestation of a large number of disc cutters in the normal wear. When a large number of cutters are in flat wear, the main mainfestation is the increase of torque. Specifically, when the proportion of flat wear cutters is 19.05% and 28.57%, the increase of torque is 55.85% and 261.51% respectively. When there are a large number of normal wear and eccentric wear of cutters, the torque and thrust increase synchronously. The torque increases by 80.89% when the flat wear of cutters accounts for 21.43%. The numerical simulation and measured data show a high level of consistency. Based on the results of four opening, the increase of cutterhead torque by more than 50% can be used as an important basis for determining the flat wear of a large number of cutters.

Abstract:The water transportation environment poses challenges in terms of complexity, making it difficult to achieve clear and diverse visual target detection in water surface images captured by conventional optical cameras. This difficulty is particularly prominent when detecting medium- and small-scale objects in visible light visual signals. For the development of smart maritime applications, we proposed a multi-scale ship object detection (MS-SOD) algorithm to improve the performance of multi-scale ship object detection in complex waters. MS-SOD is built based on the mainstream framework of one-stage object detection models. The convolutional block attention module is embedded into its backbone network to optimize the ability of ship feature extraction. The shallow features with rich detailed information are added to the multi-scale feature fusion network, and cross-stage-partial residual structure is used to enhance the fusion mechanism of multi-scale ship object features. Additionally, a focal loss function is employed to optimize the training process of the model, and an adaptive anchor clustering algorithm is designed to optimize the prior anchor and improve the detection capability for multi-scale ship objects. Extensive experiments are conducted on a self-built large-scale ship object dataset to validate the effectiveness and efficiency of the proposed MS-SOD algorithm. Experimental results show that the accuracy of MS-SOD outperforms various mainstream comparative methods on test dataset. Especially, compared with the YOLOv4 algorithm, the detection accuracy of large-, medium-, and small-scale ship objects improve by 11.3%, 6.0%, and 10.5%, respectively.

Abstract:The component parameters of the circuit significantly affect the control performance of Buck converters, especially the inductance and capacitance. A reference voltage pulse injection method is proposed for the online estimation of inductance and capacitance. The full-rank state equation is constructed using the original steady state and the new steady state to improve the estimation accuracy of inductance and capacitance in the Buck converter. The parasitic parameters and load resistance required by the algorithm are derived online. The precise discrete-time converter model based on the volt-second characteristic of the inductor and the charge balance characteristic of the capacitor is studied, which lays the foundation for parameter estimation. A short pulse signal is injected into the reference voltage, and using a PID controller to regulate the inductor current and output voltage in the circuit, a transient state and a new steady state for parameter estimation are established. The inductance and capacitance values are estimated using the voltage and inductor current sampled in the transient state, which avoids the convergence problem in the steady state. Based on the proposed average inductor current estimation algorithm, the current sampling frequency is reduced to the switching frequency. Finally, the simulation verification is carried out on Matlab/Simulink platform, and the results show that the maximum estimation errors of the inductor and capacitor values are less than 2% and 4.2%, respectively, even when considering the actual noise. Compared with other parameter estimation algorithms, the introduction of the reference voltage pulse injection method can effectively improve the estimation accuracy of parameter identification and contribute to the improvement of the control performance of the Buck converter system.

Abstract:For the improvement of the spectral efficiency of traditional pulse position modulation (PPM) symbol, a novel Intra-Chip 4-PPM symbol is proposed, which not only achieves a communication rate of 1 Gbit/s, but also greatly reduces the required spectrum resources. While in demodulation, the bit error rate performance of this modulation symbol is greatly affected by the frequency offset between the transmitter’s clock and the receiver’s clock. To address this issue, an algorithm and a circuit are proposed to compensate for the symbol frequency offset, realize symbol synchronization and enable high-speed data demodulation in the analog domain. The circuit system compensates for the frequency offset by eliminating the initial phase difference between the received data and the local clock, extracting their frequency offset information, and periodically changing the instantaneous phase of the local clock. Meanwhile, the local clock is utilized to demodulate the received data in the third step. In order to improve the linearity of the phase interpolator (PI), a delay-locked loop with the PI is introduced in this paper. Within the interpolation range of 2π, the circuit achieves 32 interpolation intervals, a step size of 992, a resolution of 2.016 ps, the maximum differential nonlinearity (DNL) of 0.183°, and the maximum integral nonlinearity (INL) of 0.325°. In addition, the phase control algorithm proposed in this paper effectively avoids the output phase jump caused by current glitch. Based on UMC 40 nm CMOS RF LP process, simulation results show that the proposed algorithm and circuit reduce the frequency deviation between received data and local clock from 50×10^-6to 1.03×10^-6 and the accuracy of frequency offset compensation reach 97.94% in the typical corner, enabling a demodulation rate of 1 Gbit/s. This method has significant engineering application value for synchronization and demodulation of high-speed PPM data.

Abstract:In view of the problems of target loss and low tracking accuracy due to scale transformation during long-term UAV tracking, a scale-aware spatial tracker based on moth-flame optimization (MFO) was proposed. First, the Gaussian initialization was used to replace the random initialization strategy of the original moth-flame optimization algorithm, so as to reduce the high computational complexity and waste of computing power of the optimization algorithm in the tracking problem. Second, on the basis of the characteristics of fast gradient histogram, an improved moth-flame optimization tracker was constructed. Then, considering the problem of target scale change under the long-term tracking of UAV aerial photography, a discriminative scale space tracking (DSST) algorithm combined with adaptive scale transformation was designed. A scale-aware spatial tracker was further proposed to solve the problem of tracking drift caused by the fixed aspect ratio of the scale filter. In addition, the variation of the filter response peak value under different backgrounds was analyzed, and an index that can reflect the tracking confidence under environmental changes was proposed. The moth-flame optimization tracking framework was combined with the scale-aware spatial tracker through confidence, which can solve the problems of scale change and target loss in long-term tracking. Finally, the performance of the algorithm was verified on the UAV long-term tracking dataset. Results show that the proposed algorithm can effectively prevent the occurrence of drift and improve the tracking efficiency. Compared with 12 similar algorithms in the tracking field, the proposed algorithm can effectively solve the scale change and the target loss of the long-term UAV tracking, and meet the requirement of real-time with high accuracy.

Abstract:As lithium-ion batteries are widely used in mobile electronics and electric vehicles, accurate prediction of their remaining useful life (RUL) is of great importance for health management of lithium batteries. A prediction method based on particle filter (PF) and gated recurrent unit (GRU) neural network fusion method (PF-GRU) is proposed to predict the RUL of lithium batteries. This fusion method combines the advantages of PF in estimating the probability distribution of RUL and the ability of GRU to perform long-term prediction of time series to obtain the fusion prediction result. Using the capacity prediction results of each prediction period, the GRU model is retrained by iteratively updating the training dataset with a sliding window, which improves long-term prediction performance of GRU. The above prediction steps are iteratively performed until the capacity decays below the lifetime threshold. Finally, prediction results represented by particles are extrapolated to the lifetime threshold, and RUL distribution histogram of the battery is obtained. The lithium battery data provided by the NASA prognostics center of excellence (PCoE) laboratory is adopted to verify the proposed method, which compares the proposed fusion method with GRU, PF and fusion method without sliding window for RUL prediction. Experimental results show that the proposed fusion method has good performance, which is obviously better than PF and fusion method without sliding window in terms of state and parameter estimation and RUL prediction accuracy.

Abstract:To reduce the impact of noise and outliers on ensemble pruning and robustly select a more sparse subset of base classifiers to improve the performance of facial expression recognition, a robust sparse and low-redundancy ensemble pruning method with dependency scores is proposed in this paper for facial expression recognition. First, the method treats the prediction results of sample instances as base classifier features, and evaluates the dependency and priority between pairs of base classifiers using mutual information and entropy, respectively. Second, the priority dependency is added to the regression-based objective equation to prune redundant base classifier, which uses the l_2,1-norm to increase the robustness of the classifier subset and thus improve the generalization performance of the algorithm. Then, an inner product regularization term is introduced into the target equation to select sparse and low-redundant base classifiers by computing the sum of the absolute values of the inner products of the classifier feature coefficient vectors. Finally, the majority voting method is used to integrate the selected subset of base classifiers to obtain the final recognition results. The recognition accuracy of the proposed method on four public facial expression datasets, FER2013, JAFFE, CK+, and KDEF, is 3.29%, 10.39%, 1.76%, and 4.89% higher than those obtained by integrating all base classifiers, respectively, indicating that the method can select a subset of classifiers with better recognition results and lower redundancy, and improve the ensemble pruning generalization ability.

Abstract:Based on a case of wind induced damage to glass curtain wall of a tower building group, the motion trajectory of glass curtain wall debris is predicted, and the influence of four different wind field conditions on the flight trajectory of glass curtain wall debris are studied. Firstly, based on the validation of the large eddy simulation method and parameter setting, the unsteady flow field around the tall buildings is obtained through the large eddy simulation of the unsteady flow around the tower group in the turbulent boundary layer wind field. Then, the fifth-order Runge-Kutta method is used to solve the governing equations of three-dimensional rigid body motion, and the validity of the method is verified by comparing the trajectory calculation method of the debris in a uniform flow field with the experimental results in the literature. Finally, the influence of different flow field conditions on the flight path and velocity of particles are analyzed. The results show that the trajectory calculation method with large eddy simulation can predict the trajectory effectively. For the debris released from the same position, the predicted flight speed and flight distance of the debris in three-dimensional free flow field are relatively lower than those in an ideal uniform flow field. When the turbulent component is ignored in the three-dimensional free flow field, the flight speed of the debris is further reduced and the flight distance is increased. For the three-dimensional unsteady flow around the tower buildings, there is a large difference between the trajectory prediction for the ideal uniform flow and the three-dimensional free flow. Moreover, the trajectory of the debris is greatly affected by the different initial positions in the flow field, which may significantly affect the final evaluation results. Therefore, the three-dimensional unsteady flow around the buildings and the initial failure position should be fully considered in the prediction of trajectory of the debris.