Search

This paper presents an enhanced double Quasi-PR controller to realize the passivity characteristic of output admittance in the whole frequency range from 0 Hz to Nyquist frequency. This passivity characteristic can be useful in the case of wind turbine generation system which grid-side inverter is connected to main grid with long transmission cable. To realize the passivity characteristic of output admittance, a constraint condition-based design method is proposed. Then, based on the proposed controller structure and design method, stability of grid-side inverter is promised regardless of grid impedance parameters. Finally, simulation results verify the accurateness and effectiveness of the proposed controller.

We study the problem of word-level confidence estimation in subword-based end-to-end (E2E) models for automatic speech recognition (ASR). Although prior works have proposed training auxiliary confidence models for ASR systems, they do not extend naturally to systems that operate on word-pieces (WP) as their vocabulary. In particular, ground truth WP correctness labels are needed for training confidence models, but the non-unique tokenization from word to WP causes inaccurate labels to be generated. This paper proposes and studies two confidence models of increasing complexity to solve this problem. The final model uses self-attention to directly learn word-level confidence without needing subword tokenization, and exploits full context features from multiple hypotheses to improve confidence accuracy. Experiments on Voice Search and long-tail test sets show standard metrics (e.g., NCE, AUC, RMSE) improving substantially. The proposed confidence module also enables a model selection approach to combine an on-device E2E model with a hybrid model on the server to address the rare word recognition problem for the E2E model., Comment: To appear in ICASSP 2021

Aiming at the problem of the large-scale weapon target distribution model, the conventional algorithms are low efficient to obtain the solution and can not obtain the fire optimization scheme untimely. Thus a new two-level task optimization distribution model is constructed, which can obtain a distribution scheme in a shorter time by combining the fuzzy clustering algorithm with the auction algorithm. The numerical example shows that the established collaborative task allocation model and the corresponding solution method can get the task allocation scheme in time and effectively, and solve the problem of the fire power optimal allocation of the large-scale weapon-target.

End-to-end (E2E) models have made rapid progress in automatic speech recognition (ASR) and perform competitively relative to conventional models. To further improve the quality, a two-pass model has been proposed to rescore streamed hypotheses using the non-streaming Listen, Attend and Spell (LAS) model while maintaining a reasonable latency. The model attends to acoustics to rescore hypotheses, as opposed to a class of neural correction models that use only first-pass text hypotheses. In this work, we propose to attend to both acoustics and first-pass hypotheses using a deliberation network. A bidirectional encoder is used to extract context information from first-pass hypotheses. The proposed deliberation model achieves 12% relative WER reduction compared to LAS rescoring in Google Voice Search (VS) tasks, and 23% reduction on a proper noun test set. Compared to a large conventional model, our best model performs 21% relatively better for VS. In terms of computational complexity, the deliberation decoder has a larger size than the LAS decoder, and hence requires more computations in second-pass decoding.

Thus far, end-to-end (E2E) models have not been shown to outperform state-of-the-art conventional models with respect to both quality, i.e., word error rate (WER), and latency, i.e., the time the hypothesis is finalized after the user stops speaking. In this paper, we develop a first-pass Recurrent Neural Network Transducer (RNN-T) model and a second-pass Listen, Attend, Spell (LAS) rescorer that surpasses a conventional model in both quality and latency. On the quality side, we incorporate a large number of utterances across varied domains [1] to increase acoustic diversity and the vocabulary seen by the model. We also train with accented English speech to make the model more robust to different pronunciations. In addition, given the increased amount of training data, we explore a varied learning rate schedule. On the latency front, we explore using the end-of-sentence decision emitted by the RNN-T model to close the microphone, and also introduce various optimizations to improve the speed of LAS rescoring. Overall, we find that RNN-T+LAS offers a better WER and latency tradeoff compared to a conventional model. For example, for the same latency, RNN-T+LAS obtains a 8% relative improvement in WER, while being more than 400-times smaller in model size.

This work addresses a realization of a compact ROV (Remotely Operated Vehicle) for monitoring in a dark and turbid aquaculture environment. A six degree-of-freedom ROV with four thrusters is implemented, in which a novel structure is proposed to solve the heeling problem better. In addition, a lighting system with wide and uniform light field and a biaxial imaging platform are designed based on the principle of convex lens scattering analysis. The simulation and tank experiment results illustrate the effectiveness of the designed ROV on steadily and flexibly. It achieves moving forward at a velocity of 0.17m/s at 50HZ artificial flow, and cruising at a fixed depth with up to 5 hours battery life. The designed ROV exhibits better comprehensive performance than other small ROVs for the turbid aquaculture environments.

In this paper we introduce a new SVR (Support Vector Regression) based model for estimating air pollution at fine spatial granularity. Specifically we use historical data from (sparse) government monitoring sites and a (dense) wireless sensor network, along with SVR — a supervised regression learning method to estimate an air pollution surface for any given hour on any given day in Sydney. Further, we conduct trials and results which show that air pollution estimations from our estimation system has a high spatial resolution, and is more accurate than estimations from an ANN (Artificial Neural Network) model. Our model can benefit further medical studies on estimating dosage and its relation to health outcomes.

Cloud data centers contribute greatly to global warming, because most of the energy is generated by burning fossil fuels. In view of this, many cloud data centers are trying to power their data centers using renewable energy. In this paper, we propose a green scheduling architecture for the geographically distributed cloud data centers with time-varying and location-varying electricity prices. To lower down the energy cost and carbon emissions, each data center has its own wind turbines and solar panels. The generated renewable energy can be used to power data centers directly or stored into ESDs for latter use, or sold back to the power grid. However, it is hard to make decisions on the usage of each type of energy considering the dynamic incoming requests of users, fluctuating electricity prices, and intermittent energy supply in each time slot. Our problem is formulated as a mixed integer linear programming (MILP) problem: Given the arrival of incoming requests, schedule the requests, servers, and the usage of different energy sources, such that the total energy cost can be minimized while satisfying QoS requirement within certain carbon emission level. Our simulation is based on the traces from real world. Experiments show that our method can significantly lower down the energy cost for green cloud data centers by using ESDs and energy trading.

In this paper, an algebraic method which is based on the groebner bases theory is proposed to solve the selective harmonic elimination PWM(SHEPWM) problem. Firstly, the SHEPWM equations are transformed to polynomial equations by using the multiple-angle formulas and variables substitution, then, the polynomial equations are converted to an equivalent triangular form by computing the reduced groebner bases under the pure lexicographic monomial order, finally, the triangular equations can be solved by a successive back-substitution manner just like the Gaussian elimination which is used to solve the linear equations. A software package is developed under the symbolic computing software Maple, and it can solve the SHEPWM equations with no more than 8 switching angles for the single-phase inverters and 5 switching angles for the three-phase inverters. In contrast with the common used numerical and intelligent methods, the advantages of this method are there is no need to choose the initial values and all the real solutions can be found. Experimental results verify the correctness of the switching angles computed by the proposed method.

Recently, high spatial resolution is increasingly needed in air pollution forecast by governments and climatologists. However, data from fixed monitoring sites cannot meet these requirements. In this paper, we present a hybrid spatio-temporal model to improve the air quality forecast capacity with fine grid data from ubiquitous mobile sensor network.

In recent years several research groups, including ours, have demonstrated participatory systems that use wearable or vehicle-mounted portable units coupled with smartphones to crowdsource urban air pollution data from lay users. These systems have shown remarkable improvement in spatial granularity over government-operated monitoring systems, leading to better mapping and understanding of urban air pollution, at relatively low cost. In this paper we extend the paradigm to personalize the consumption of data by individuals. Specifically, we combine the pollution concentrations obtained from participatory systems with the individual's on-body activity monitors to estimate the personal inhalation dosage of air pollution. We show that the individual's activity, such as jogging, cycling, or driving, impacts their dosage, and develop an app that gives them this personalised information. Our system is a step towards enabling medical inferencing of the impact of air pollution on individual health.

This paper describes our experiences with building and deploying a low-cost participatory system for urban air pollution monitoring in Sydney. Though air pollution imposes significant health costs on the urban community globally, today's published data on pollution concentrations is spatially too sparse, and does not allow for sufficiently accurate estimation of exposures for (potentially mobile) individuals in order to make medical inferences. The HazeWatch project described in this paper uses several low-cost mobile sensor units attached to vehicles to measure air pollution concentrations, and users' mobile phones to tag and upload the data in real time. The greater spatial granularity of data thus collected enables creation of pollution maps of metropolitan Sydney viewable in real-time over the web, as well as personalized apps that show the individual's exposure history and allow for route planning to reduce future exposure. We share the insights we obtained from building and trialling such a system in Sydney, and highlight challenges that can be addressed collaboratively by groups developing similar systems world-wide.

Linear Asset is taking an increasingly important role in daily life. Segmentation serves as the fundamental issue for maintenance and further analysis. In this paper, we propose a new condition based dynamic segmentation approach, which keeps the within-segment consistency of the asset condition while making segments easily managed according to user input. In order to measure the segmentation performance, we design new evaluation criterion and conduct a group of experiments to verify our approach with other algorithms.

Unvoiced-voiced portions of cochannel speech contain considerable amounts of both voiced and unvoiced speech and play a significant role in separation. Motivated by recent developments in separation of speech from nonspeech noise, we propose a classification-based approach for unvoiced-voiced speech separation. A new feature set consisting of pitch-based features and gammatone frequency cepstral coefficients is proposed to represent the characteristics of a time-frequency unit. The cepstral features do not rely on pitch and are thus more robust than the pitch-based features to pitch estimation errors. Speaker-independent support vector machines are trained for classification. Results based on the TIMIT corpus show that the proposed algorithm significantly improves unvoiced speech segregation compared to a recent algorithm.

In order to investigate the efficacy of combining radiation therapy with endostar on Lewis lung carcinoma in mice and the interaction mechanisms of combined therapy. Tumor models were established in the hind limb of female C57 BL/6N mice by subcutaneous transplantation. The tumor-bearing mice were randomly divided into 4 groups: control group, endostar group (20 mg·kg−1, once daily for 15 days), radiotherapy group (2 Gray per day to 10 Gray, d6-d10), and radiotherapy plus endostar group (combination group), (n=10 in each group). The tumors were applied to analysis of the tumor inhibitory rate and growth curve. Immunohistochemisty was adopted to check the expressions of microvessel density (MVD). The expression levels of Hypoxia-inducible factor 1α (HIF-1α) mRNA was tested by real-time fluorescence quantitative polymerase chain reaction (RT-PCR). The results demonstrated that the tumor inhibitory rate in combined treatment group was higher than that in other groups. Meanwhile, MVD and the level of HIF-1α mRNA were lower than that in other groups. We concluded that endostar significantly sensitized the function of radiation on Lewis lung carcinoma in mice, and this effect is working by decreasing HIF-1α, thereby increasing the killing effect of radiation on tumor cells.

In order to rationally design the mixed model assembly line and effectively solve the assembly line balancing problem, this paper analyses the characteristics of mixed model assembly lines and designs a new algorithm based on the process exchange. This algorithm could combine with research achievements of mixed model sequencing problems to make the assembly line more balanced owing to its high compatibility. This paper describes the principle and the executing process of this new algorithm in details. Lastly, the new algorithm was applied in an automobile harness of lamp assembly line, the results of which proved the validity and feasibility of the algorithm this paper proposed.

Software as a Service (SaaS) is an emerging model to deliver software over the Web as a service. With the fast growing of the SaaS business in the recent years, data mining application in SaaS environment becomes an attractive topic. While many existing Web mining technologies can be leveraged, SaaS has its own unique perspective. In this paper, we propose a model for SaaS tenant behavior analysis based on the unique perspective of SaaS environment. The corresponding algorithm to derive the metrics of the model by means of mining the tenant's collaboration indices from raw log is also presented. The derived collaboration indices can provide an important metric for efficiently launching data mining algorithm for customer analysis in a SaaS environment. Its effectiveness is verified through an experiment comparing the result for different types of tenants on a real SaaS service.

Many algorithms have been developed to optimize police patrol services. Previous studies have mainly focused on determining the important locations (e.g., crime hotspots) and identifying important routes based on the topology of road networks. However, the impact of the patterns of hotspots on patrol route selections and the collective performance of patrol activities were rarely considered in these studies. In addition, some algorithms lack a mechanism to ensure the randomness in patrolling and some are not efficient enough for real-time applications. In this study, we propose an approach to determine sets of patrol routes, which can help to optimize the collective efforts of multiple patrol activity. In our approach, we integrate the Getis-Ord Gi* with the Cross Entropy approach to produce randomized optimal patrol routes. We also address how to ensure the randomness in the route identification using the Cross Entropy approach. Our results indicate that our approach can help to improve the collective performance of police patrol service and it is also efficient enough for real-time application.

Model-based methods for sequential organization in cochannel speech require pretrained speaker models and often prior knowledge of participating speakers. We propose an unsupervised approach to sequential organization of cochannel speech. Based on cepstral features, we first cluster voiced speech into two speaker groups by maximizing the ratio of between- and within-group distances penalized by within-group concurrent pitches. To group unvoiced speech, we employ an onset/offset based analysis to generate time-frequency segments. Unvoiced segments are then labeled by the complementary portions of segregated voiced speech. Our method does not require any pretrained model and is computationally simple. Evaluations and comparisons show that the proposed method outperforms a model-based method in terms of speech segregation.

A new method for modeling and analyzing scheduling of multi-variety and multi-process job-shop based on multi-agent was proposed to effectively solve the dynamic and complex scheduling of multi-variety and multi-process job-shop problem. A physical and mathematic scheduling model of multi-variety and multi-process job-shop based on multi-agent was established. The analyzing scheduling system obtained the target function based on six kinds of agents such as planning agents, searching agents, negotiating agents, deciding agents, logistics agents and database agents. A system was designed to realize the analyzing scheduling of multi-variety and multi-process job-shop with Java programming language. The scheduling of the job-shop in a certain national automobile liner production enterprise was taken as an example. Results proved the validity and reliability of this multi-agent model and scheduling system.

Taxi is an indispensable component of urban transportation system, which aims to provide a convenient and efficient means of transport. However, unlike bus service that has fixed station and schedule, to most taxi-seekers, the formula for catching a taxi at the nearest place with the shortest waiting time is experience plus luck, especially in peak hour or adverse weather conditions. In this demo paper, we present Taxi-Viewer, a service-oriented online system to help users pinpoint the hot spots where usually have more chances to catch a taxi, and keep users informed of the real-time location and trajectory of taxis in a given neighborhood. With the functions provided in the demo, taxi-seekers can reduce waiting time, and taxi operator can also benefit from the system to improve service quality and increase profit as well.

Hypoxia-inducible factor 1 (HIF-1) and vascular endothelial growth factor (VEGF) play important roles in tumor angiogenesis. We examined the effect of endostar, an inhibitor of angiogenesis on the expression of HIF-1 and VEGF and on angiogenesis and radio sensitization. Flow cytometry (FCM) was used to evaluate the cell cycle phase distribution affected by endostar. The cell growth inhibiting rate was detected by methyl thiazolyl tetrazolium (MTT). Hoechst staining displayed the apoptotic cells, the expression of HIF-1 and VEGF was determined by enzyme linked immunosorbent assay (Elisa). To explore the anti-tumor efficacy in vivo, Lewis lung cancer models were established in the hind limb of female C57BL/6J mice by subcutaneous transplantation. We calculated the tumor inhibitory rate and detected the microvessel density (MVD) and VEGF expression by Immunohistochemistry. Whether in vitro or in vivo, endostar in combination with radiotherapy (RT) had the most significant inhibitory effect. Endostar had the function of block cell periods on A549 and stopped the cell cycle at G2/M and S periods. It also decreased the expression of HIF-1 and VEGF induced by radiotherapy. Endostar combined with irradiation had synergistic anti-tumor effect. And the mechanism may be that endostar can arrest the cell cycle at stage of G2/M and S, thereby, inhibiting the proliferation of tumor cells. In addition, the inhibition of HIF-1 and VEGF which play important role in resistant of radiotherapy may contribute to the synergistic effect.

Recently, considerable attention has been focused on the investigation of the use of wireless sensor networks (WSN) to drive mobile robot for efficient exploration in unknown environments. The process of navigation depends heavily on estimating the position of the features provided by sensor within the entire environment. This paper proposes a novel navigation algorithm based on RF wireless sensor networks to simultaneous localization and mapping (SLAM) for an indoor autonomous mobile robot. Triangulation localization method is employed to make the robot locates itself and knows its pose. A simple yet efficient two-step particle filter is applied for map building. Simulation and experimental results show that good localization can be achieved using the proposed method.

Bilateral Filter has demonstrated great effectiveness for a variety of applications in image process, computer vision and computer graphic due to its property of edge preserving while removing noise. Since it is not as fast as linear filter because of its nonlinear nature, lots of researchers have proposed their own fast versions for Bilateral Filter. In this paper, we focus on the fast versions of Fast and Piecewise-linear Bilateral Filter. With the idea of weight averaging the 3D function which is defined on the image surface, we obtained the Fast Bilateral Filter through a more intuitionistic way than the pure mathematic derivation method which is used by Paris. By varying the 3D function form, we derived another 3D expression of Bilateral Filter, then demonstrated that the Piecewise-linear version is a downsampled version of ours, and proposed some theoretical interpretation for this downsampling strategy.

Ultra-fine copper powder was prepared by ball milling solid phase reaction in ambient condition. Ctystallinity of the synthesized copper powder was investigated using an X-ray diffraction, while the morphology of the copper particles was studied via scanning electron microscopy. The energy dispersive spectroscopy peaks and the analysis of resultant particles' components were also presented. The resultant particles were protected from oxidation by atmospheric air, by the use of benzotriazole. Face-centered cubic copper particles with about 200 nm in diameter were obtained. The resultant powder was impurity-free and shows good stability.

Unvoiced speech poses a big challenge to current monaural speech segregation systems. It lacks harmonic structure and is highly susceptible to interference due to its relatively weak energy. This paper describes a new approach to segregate unvoiced speech from nonspeech interference. The system first estimates a voiced binary mask, and then performs unvoiced speech segregation in two stages: segmentation and grouping. In segmentation, time-frequency units labeled as 0 in the voiced binary mask are first used to estimate the noise energy and spectral subtraction is then performed to generate time-frequency segments in unvoiced intervals. Based on the type of noise, unvoiced segments are grouped either by selecting segments consistent with those generated by onset/offset analysis or by Bayesian classification of acoustic-phonetic features. Systematic evaluation and comparison show that the proposed approach improves the performance of unvoiced speech segregation considerably.

In this paper, we propose new rotary equipment for interconnected operations of power systems. That is a doubly-fed rotary frequency converter, which consists of two machines that are directly connected each other with shaft and they are both doubly-fed machines, or a synchronous machine and a doubly-fed machine. Analytical models were newly developed in order to analyze and understand the dynamic behavior of the proposed doubly-fed rotary frequency converters. Basic control scheme for the converters is proposed. Results of studies on dynamic characteristics of the control systems applied in the converters are shown. Moreover, merits and demerits were discussed in comparison with static DC converter with thyristor valves from the practical point of view.

The doubly-fed (or adjustable speed) pumped storage generating systems which have been newly developed with recent progress of power electronic technology as its setting can contribute not only to power system frequency control at night, but also to improvement of power system stability. In this paper, as an application of doubly-fed machine technologies of large capacity, a new rotary frequency converter is proposed that consists of MG set of a doubly-fed machine and an induction machine. Compared with a conventional rotary frequency converter that consists of a couple of synchronous machines, the new doubly-fed rotary frequency converter can control power transfer actively. The mathematical models of the proposed new converter and basic considerations for their fundamental control strategy based on the models are described, and the results of dynamic simulations for investigating the characteristics of control and operation of the converter in power system are presented.