Search

This paper deals with the problem of guidance law design for the single moving mass controlled reentry vehicle when impact angle constraints and maneuvering target are taken into consideration. More specifically, a modified rolling guidance law is proposed with the interactive virtual target and the landing point prediction strategy. First, considering the fact that the roll channel can be controlled directly, the relative motion between the single moving mass controlled reentry vehicle and the target is described by the error angle between the relative velocity and the line-of-sight. Second, a nonlinear error angle command is given to reduce the rotation rate. To satisfy impact angle constraints, an interactive virtual target is presented and the “S” formed velocity of the virtual target is given to abate the error angle tracking difficulty at the final stage of the reentry phase. Then, the landing point prediction strategy is employed and the motion variation trend is also taken into consideration. As the maneuvering target is replaced with the predicted landing point, the error angle tracking difficulty caused by the target velocity decreases, which is helpful to meet impact angle constraints and improve guidance accuracy at the same time. Finally, the finite-time rolling guidance law is proposed and proved via Lyapunov stability theorem. Compared with the existing method, lower-speed rotation, smaller missing distance and less impact angle errors are obtained, which can be demonstrated by numerical simulations.

Cloud is complicated due to its various types and distribution on different layers. Various classification principles and overlapping of clouds on multiple layers make it difficult to classify clou...

In machine translation evaluation, the traditional wisdom measures model's generalization ability in an average sense, for example by using corpus BLEU. However, the statistics of corpus BLEU cannot provide comprehensive understanding and fine-grained analysis on model's generalization ability. As a remedy, this paper attempts to understand NMT at fine-grained level, by detecting contextual barriers within an unseen input sentence that cause the degradation in model's translation quality. It proposes a principled definition of source contextual barriers as well as its modified version which is tractable in computation and operates at word-level. Based on the modified one, three simple methods are proposed for barrier detection by search-aware risk estimation through counterfactual generation. Extensive analyses are conducted on those detected contextual barrier words on both Zh $\Leftrightarrow$ En NIST benchmarks. Potential usages motivated from barrier words are also discussed.

Machines translation (MT) is an essential task in natural language processing or even in artificial intelligence. Statistical machine translation has been the dominant approach to MT for decades, but recently neural machine translation achieves increasing interest because of its appealing model architecture and impressive translation performance. In neural machine translation, an attention model is used to identify the aligned source words for the next target word, i.e., target foresight word, to select translation context. However, it does not make use of any information about this target foresight word at all. Previous work proposed an approach to improve the attention model by explicitly accessing this target foresight word and demonstrating substantial alignment tasks. However, this approach cannot be applied in machine translation tasks where the target foresight word is unavailable. This paper proposes several novel enhanced attention models by introducing hidden information (such as part-of-speech) of the target foresight word for the translation task. We incorporate the novel enhanced attention employing hidden information about the target foresight word into both recurrent and self-attention-based neural translation models and theoretically justify that such hidden information can make translation prediction easier. Empirical experiments on four datasets further verify that the proposed attention models deliver significant improvements in translation quality.

From the view of the fixed-trim reentry vehicle (FTRV) application, the integrated guidance and control (IGC) design is a key challenge. In order to improve the performance of the FTRV against the maneuvering target, an IGC system combining with the virtual target and the filter is proposed in this paper. To investigate the dynamics of the FTRV against the maneuvering target, a 7-DOF mathematical model is established and the error angle is employed to describe the relative motion between the FTRV and the target. Considering the controllability of the FTRV, a nonlinear differentiable error angle command is presented and its influence on the terminal velocity is discussed. To deal with the contradiction between the limited maneuverability and the demand of high guidance accuracy, the actual target is taken place with the virtual target, of which acceleration is estimated by the ESO. For improving the performance of the virtual target position prediction, a rolling updating strategy is proposed and derived analytically. Then, the back-stepping based IGC system is designed with the virtual target and the filter. The finite-time convergence of the IGC system is proved via the Lyapunov stability theorem. The numerical simulation results show the effectiveness of the proposed IGC system for the FTRV against the maneuvering target with time-varying acceleration.

Conventional research trend detection methods require a large number of papers to perform analysis, while the trends detected are usually retrospective and cannot represent the emerging topic which is in embryonic stage, because detection method based on word features or citation relationships have to wait for the trend to come into form and become popular to detect it. Besides, the acquisition of paper full text is usually costly. We employ the paper metadata, which is relatively easy to acquire and process, to detect potentially valuable papers in specific field once the papers are published or their metadata become accessible and extract emerging cutting-edge trend from them. By comparing the result with conventional methods, we prove the effectiveness of the method.

Whereas adversarial training is employed as the main defence strategy against specific adversarial samples, it has limited generalization capability and incurs excessive time complexity. In this paper, we propose an attack-agnostic defence framework to enhance the intrinsic robustness of neural networks, without jeopardizing the ability of generalizing clean samples. Our Feature Pyramid Decoder (FPD) framework applies to all block-based convolutional neural networks (CNNs). It implants denoising and image restoration modules into a targeted CNN, and it also constraints the Lipschitz constant of the classification layer. Moreover, we propose a two-phase strategy to train the FPD-enhanced CNN, utilizing $\epsilon$-neighbourhood noisy images with multi-task and self-supervised learning. Evaluated against a variety of white-box and black-box attacks, we demonstrate that FPD-enhanced CNNs gain sufficient robustness against general adversarial samples on MNIST, SVHN and CALTECH. In addition, if we further conduct adversarial training, the FPD-enhanced CNNs perform better than their non-enhanced versions.

Recently many efforts have been devoted to interpreting the black-box NMT models, but little progress has been made on metrics to evaluate explanation methods. Word Alignment Error Rate can be used as such a metric that matches human understanding, however, it can not measure explanation methods on those target words that are not aligned to any source word. This paper thereby makes an initial attempt to evaluate explanation methods from an alternative viewpoint. To this end, it proposes a principled metric based on fidelity in regard to the predictive behavior of the NMT model. As the exact computation for this metric is intractable, we employ an efficient approach as its approximation. On six standard translation tasks, we quantitatively evaluate several explanation methods in terms of the proposed metric and we reveal some valuable findings for these explanation methods in our experiments., Comment: Accepted to ACL 2020, 9 pages

A generic feature of bounded confidence type models is the formation of clusters of agents. We propose and study a variant of bounded confidence dynamics with the goal of inducing unconditional convergence to a consensus. The defining feature of these dynamics, which we name the No one left behind dynamics, is the introduction of a local control on the agents which preserves the connectivity of the interaction network. We rigorously demonstrate that these dynamics result in unconditional convergence to a consensus. The qualitative nature of our argument prevents us quantifying how fast a consensus emerges, however we present numerical evidence that sharp convergence rates would be challenging to obtain for such dynamics. Finally, we propose a relaxed version of the control. The dynamics that result maintain many of the qualitative features of the bounded confidence dynamics yet ultimately still converge to a consensus as the control still maintains connectivity of the interaction network.

Multilayer architectures are currently the gold standard for large-scale neural machine translation. Existing works have explored some methods for understanding the hidden representations, however, they have not sought to improve the translation quality rationally according to their understanding. Towards understanding for performance improvement, we first artificially construct a sequence of nested relative tasks and measure the feature generalization ability of the learned hidden representation over these tasks. Based on our understanding, we then propose to regularize the layer-wise representations with all tree-induced tasks. To overcome the computational bottleneck resulting from the large number of regularization terms, we design efficient approximation methods by selecting a few coarse-to-fine tasks for regularization. Extensive experiments on two widely-used datasets demonstrate the proposed methods only lead to small extra overheads in training but no additional overheads in testing, and achieve consistent improvements (up to +1.3 BLEU) compared to the state-of-the-art translation model.

Many Data Augmentation (DA) methods have been proposed for neural machine translation. Existing works measure the superiority of DA methods in terms of their performance on a specific test set, but we find that some DA methods do not exhibit consistent improvements across translation tasks. Based on the observation, this paper makes an initial attempt to answer a fundamental question: what benefits, which are consistent across different methods and tasks, does DA in general obtain? Inspired by recent theoretic advances in deep learning, the paper understands DA from two perspectives towards the generalization ability of a model: input sensitivity and prediction margin, which are defined independent of specific test set thereby may lead to findings with relatively low variance. Extensive experiments show that relatively consistent benefits across five DA methods and four translation tasks are achieved regarding both perspectives.

Prior researches suggest that neural machine translation (NMT) captures word alignment through its attention mechanism, however, this paper finds attention may almost fail to capture word alignment for some NMT models. This paper thereby proposes two methods to induce word alignment which are general and agnostic to specific NMT models. Experiments show that both methods induce much better word alignment than attention. This paper further visualizes the translation through the word alignment induced by NMT. In particular, it analyzes the effect of alignment errors on translation errors at word level and its quantitative analysis over many testing examples consistently demonstrate that alignment errors are likely to lead to translation errors measured by different metrics.

This paper treats the problem of attitude control for the fixed-trim reentry vehicle (FTRV) in the presence of actuator input and power constraints. More specifically, prescribed performance controller (PPC) is developed for the FTRV with the auxiliary system and the nonlinear filter. Compared with the conventional auxiliary system, the auxiliary system in this paper can response before the saturation occurs with the saturation risk prediction, decreasing the saturation effectively. Then the power constraint is converted by the mathematical derivation and the nonlinear filter is employed. Considering the controller failure of the conventional PPC caused by the discontinuous command, an improved performance function is presented in this paper. Then, a disturbance observer is employed to improve the system robustness. The finite-time convergence of the proposed control system is proved by the Lyapunov stability theorem. Compared with conventional controller, lower saturation and less energy consumption can be achieved, which can be demonstrated by numerical simulations.

With the development of the ages and social progress, our country has enormously developed in the field of smart power system. In most of the power systems, the hydraulic actuator, the operating mechanism and the spring energy storage operating mechanism are all the core issues in the current relay protection circuit breaker opening and closing. The effect of its operation will have a direct impact on the power system and circuit breaker performance. In this paper, by classifying the circuit breaker operating and analyzing the relationship between the blocking action and the relay protection, we can further analyze the blocking action of the circuit breaker in the relay protection, so that we can choose the better action position, and analyze and make researches by following the circuit wiring and dual configuration.

Compared with traditional ways of control for reentry vehicle, single moving mass control provides simple atmospheric layout, which means the erosion can be decreased when the vehicle is hypersonic in the reentry phase. However, the path-control of single mass control reentry can be only achieved by rolling-axis with the magnitude of lift force uncontrollable. Modified rolling guidance law based on sliding mode is taken out taken the limit of rolling speed into consideration. With the exponential decay LQR method, tracking normal trajectory is added into the guidance law at the phase where the lift force is not able to provide enough control force. The simulation is taken to testify the effect of the rolling guidance law for the whole reentry phase. The results show the robustness of the compound guidance law presented in this paper.

In order to alleviate data sparsity and overfitting problems in maximum likelihood estimation (MLE) for sequence prediction tasks, we propose the Generative Bridging Network (GBN), in which a novel bridge module is introduced to assist the training of the sequence prediction model (the generator network). Unlike MLE directly maximizing the conditional likelihood, the bridge extends the point-wise ground truth to a bridge distribution conditioned on it, and the generator is optimized to minimize their KL-divergence. Three different GBNs, namely uniform GBN, language-model GBN and coaching GBN, are proposed to penalize confidence, enhance language smoothness and relieve learning burden. Experiments conducted on two recognized sequence prediction tasks (machine translation and abstractive text summarization) show that our proposed GBNs can yield significant improvements over strong baselines. Furthermore, by analyzing samples drawn from different bridges, expected influences on the generator are verified.

A novel single-wire power transmission system (SWPT) was introduced in this paper. This system consists of the transmitter and the receiver which are connected via only single-wire. The Tesla's transformer is used in both the transmitter and the receiver. The electric energy was transmitted through the spatial electric field between the transmitter and the receiver. In this paper, the lumped-parameter circuit model of the SWPT system was built according to the current type and its rout. Based on the proposed model, the guideline of operation frequency selection was addressed and the transmission ratio was expressed analytically. Two SWPT systems with different scaled Tesla's transformers were designed and demonstrated in experiments. The experimental results for these two SWPT systems showed the capability of power transmission in 2m distance with 40W power and 66.7% efficiency and in 30m distance with 100W power and 10% efficiency, respectively.

In order to realize accurately calculating and automatically grading of plant disease severity, a single-leaf disease severity automatic grading system based on image processing was developed by using MATLAB GUIDE platform. Using this system, the single-leaf disease severity could be automatically assessed and graded via image development technologies including segmentation processing technologies of plant disease images and related data mining technologies. Structural diagram of the system, algorithms used in the system and realization of the system functions were described. The problems in the current version of the system were discussed and further research on this subject was suggested. The usefulness and adaptability of the system was evaluated using the images of grape downy mildew caused by Plasmopara viticola. The results showed that the effectiveness of the system was favorable with high accuracy.

Plant disease identification based on image processing could quickly and accurately provide useful information for the prediction and control of plant diseases. In this study, 21 color features, 4 shape features and 25 texture features were extracted from the images of two kinds wheat diseases (wheat stripe rust and wheat leaf rust) and two kinds of grape diseases (grape downy mildew and grape powdery mildew), principal component analysis (PCA) was performed for reducing dimensions in feature data processing, and then neural networks including backpropagation (BP) networks, radial basis function (RBF) neural networks, generalized regression networks (GRNNs) and probabilistic neural networks (PNNs) were used as the classifiers to identify wheat diseases and grape diseases, respectively. The results showed that these neural networks could be used for image recognition of these diseases based on reducing dimensions using PCA and acceptable fitting accuracies and prediction accuracies could be obtained. For the two kinds of wheat diseases, the optimal recognition result was obtained when image recognition was conducted based on PCA and BP networks, and the fitting accuracy and the prediction accuracy were both 100%. For the two kinds of grape diseases, the optimal recognition results were obtained when GRNNs and PNNs were used as the classifiers after reducing the dimensions of feature data with PCA, and the prediction accuracies were 94.29% with the fitting accuracies equal to 100%.

In order to realize automatic disease diagnosis and provide related information for disease prediction and control timely and accurately, the identification and diagnosis of grape downy mildew and grape powdery mildew was conducted based on image recognition technologies. The method based on K_means clustering algorithm was used to implement unsupervised segmentation of the disease images. Fifty shape, color and texture features were extracted from the images of the diseases. Support vector machine (SVM) classifier for the diseases was designed based on thirty-one effective selected features. The training recognition rates of these two kinds of grape diseases were both 100%, and the testing recognition rates of grape downy mildew and grape powdery mildew were 90% and 93.33%, respectively. The recognition results using the SVMs with different kernels indicated that the SVM with linear kernel was the most suitable for image recognition of the diseases. This study provided an effective way for rapid and accurate identification and diagnosis of plant diseases, and also provided a basis and reference for further development of automatic diagnosis system for plant diseases.

Virtual radio space control system is the necessary tendency to spacecraft control Spacecraft control system based on Virtual radio is different from trandkional software radio in the data paralleled transmission and time synchronization. Dynamic time-division multiplexing transmission mode is presented by analyzing the data paralleled and synchronous transmission. The problem caused by traditional software radio frame that paralleled high speed data being blocked by low speed data is solved in the transmission time delay. According to uncertainty problems of data transmission and dealing time delay caused by the asynchronous work model of the virtual radio spacecraft control system, the method of time synchronization based on the data frame structure is presented, and the measure synchronization is preformed in a general way.

Based on the normal form of chaotic system, the compound nonlinear feedback control is proposed to generate more complex chaotic system. In the general three-dimensional continuous-time system, a novel nonlinear feedback control system has been designed. Simulation results show that the novel system can generate more complicated chaotic attractors.