Search

Multi-step station-level crowd flow prediction (Ms-SLCFP) is to predict the count of people that would depart from or arrive at subway/bus/bike stations in multiple future consecutive time periods. By providing a long term view, it benefits the decision making in related applications, such as public safety, traffic management, etc. However, performing Ms-SLCFP is challenging as complicated spatiotemporal correlations are formed among stations due to the flowing crowd. Besides, the crowd flow at a single station fluctuates a lot though the regularity is obvious at the regional level. To tackle such issues, we propose a deep neural networks-based model with spatiotemporal hierarchical attention mechanisms, called ST-HAttn for short, for Ms-SLCFP. The notable contributions are that ST-HAttn performs attention mechanisms (AM) in two ways: 1) implementing AM at both station level and regional level; 2) implementing AM to explicitly model the pairwise correlations of station-region instead of station-station. The intuition is to alleviate the negative impact on Ms-SLCFP due to the fluctuation of the crowd flow at the station level. Verified on three real-world datasets, ST-HAttn outperforms the state-of-the-art methods in terms of Ms-SLCFP.

Multi-step citywide crowd flow prediction (MsCCFP) is to predict the in/out flow of each region in a city in the given multiple consecutive periods. For traffic control and public safety protection, it can provide a long term view for taking measures. However, the spatial and temporal correlations in crowd movements and the lack of information make MsCCFP challenging. In this paper, a deep-learning based prediction model with spatial-temporal attention mechanism is proposed for MsCCFP. The model, called ST-Attn for short, follows the general encoder-decoder framework for modeling sequential data but adopts a multiple-output strategy to preserve the correlations characterizing between each predicted step. The spatial-temporal attention mechanism learns to globally determine the focus on those parts of the city at certain periods that are more relevant to the predicted region and time period. Besides, a pre-predicted result calculated by spatiotemporal kernel density estimation is fed to ST-Attn, which provides a reference for further accurate predicting. Experiments on three real-world datasets are carried out to verify ST-Attn's performance and the results show that ST-Attn outperforms the baselines in terms of MsCCFP.

In this paper, we propose a novel outlier detection method for robust manifold learning. First, we assume that the input high-dimensional data can be represented by an integration of local linear patches, and each patch consists of samples in the local neighborhood that maintains a linear relationship. Then, we use a hierarchical divisive clustering method to seek maximum linear patches (MLPs) and present a local density based scheme to detect outliers in each MLP. In order to evaluate the performance of outlier detection, we also propose an improved outlier detection evaluation method based on manifold distance, which is suitable for robust manifold learning. Last, we give several experiments to demonstrate the effectiveness of the proposed outlier detection method and evaluation method.

Rotation, anisotropic scaling, translation, outliers and noise are common problems in point set registration. In order to solve the point set registration under the above conditions, a robust weighted scaled iterative nearest point (WSICP) algorithm is proposed in this paper. Based on the Euclidean distance between point pairs, a weighted strategy of point pairwise is designed to reduce the influence of outliers and noise points on the solution of transformation matrix. Next, the brief and clear registration transform matrices of rotation, scale and translation are given. Experiments result show that our algorithm performs better than others on registration accuracy and convergence quality under scale stretching, outliers and noise case.

High crowd mobility is a characteristic of transportation hubs such as metro/bus/bike stations in cities worldwide. Forecasting the crowd flow for such places, known as station-level crowd flow forecast (SLCFF) in this paper, would have many benefits, for example traffic management and public safety. Concretely, SLCFF predicts the number of people that will arrive at or depart from stations in a given period. However, one challenge is that the crowd flows across hundreds of stations irregularly scattered throughout a city are affected by complicated spatio-temporal events. Additionally, some external factors such as weather conditions or holidays may change the crowd flow tremendously. In this paper, a spatio-temporal U-shape network model (ST-Unet) for SLCFF is proposed. It is a neural network-based multi-output regression model, handling hundreds of target variables, i.e., all stations&rsquo, in and out flows. ST-Unet emphasizes stations&rsquo, spatial dependence by integrating the crowd flow information from neighboring stations and the cluster it belongs to after hierarchical clustering. It learns the temporal dependence by modeling the temporal closeness, period, and trend of crowd flows. With proper modifications on the network structure, ST-Unet is easily trained and has reliable convergency. Experiments on four real-world datasets were carried out to verify the proposed method&rsquo, s performance and the results show that ST-Unet outperforms seven baselines in terms of SLCFF.

To analyze taxi origin-destination (OD) trips data and understand passengers flow patterns from the pick-up/drop-off distribution, is beneficial to many applications in taxi dispatching systems. Mapping the pick-up/drop-off locations onto taxi zones maps, relations between zones would come into being when zones are geographical adjacent, or exchanging many passengers, or sharing similar pick-up/drop-off patterns. We categorize these relations between zones as geo-neighbor, complementary and homogenous respectively. An analytic method based on dynamical network view to detect communities of the latter two relations is proposed in this paper. It makes an analogy between zones and words and performs a distributed representation learning method, mapping zones to a dense low-dimensional vector space where closely related zones (complementary/homogenous) are close. Then, community detection is easily performed by using the cosine of two vectors as a measurement. With New York City taxi trips data as a case of study and making a comparison between workdays and weekends, experiments based on the proposed method show interesting results.

Unmanned aerial vehicles (UAVs) have been increasingly used in earth observation, public safety, military and civilian applications due to its portability, high mobility and flexibility. In some GPS-denied environments, accurate drone position cannot be obtained due to occlusion, multi-path interference and other factors. While understanding and localization the content of the images is vital for earth observation, map revision, multi-source image fusion, disaster relief, smart city and other applications. The progress of computer vision and convolutional neural networks(CNNs) in image processing provide a promising solution to locate UAVs aerial image and mapping to the large-scale reference image. Firstly, key localization techniques based on image retrieval-----image description, image matching and position mapping are summarized considering the characteristics of UAVs aerial images. And then, image localization based on extracting deep semantic features and image localization based on classification method by subdividing areas are recommended. Throughout this paper, we will have an insight into the prospect of the UAVs image localization and the challenges to be faced.

Electromagnetic detection satellite (EDS) is a type of Earth observation satellite (EOS). Satellites observation and data down-link scheduling plays a significant role in improving the efficiency of satellite observation systems. However, the current works mainly focus on the scheduling of imaging satellites, little work focuses on the scheduling of EDSes for its specific requirements. And current works mainly schedule satellite resources and data down-link resources separately, not considering them in a globally optimal perspective. The EDSes and data down-link resources are scheduled in an integrated process and the scheduling result is searched globally. Considering the specific constraints of EDS, a coordinate scheduling model for EDS observation tasks and data transmission jobs is established and an algorithm based on the genetic algorithm is proposed. Furthermore, the convergence of our algorithm is proved. To deal with some specific constraints, a solution repairing algorithm of polynomial computing time is designed. Finally, some experiments are conducted to validate the correctness and practicability of our scheduling algorithms.

Continuous Data Protection (CDP) can restore data to any point-in-time, but high storage overhead and drastic system performance drop restricts its application. In this paper, we propose a file version based file level CDP system (FV-CDP) by using cheap distributed storage for backup to low down the storage costs and using local object cache and parralel asynchronous object sending to mask network storage latency. It designs special opration log to identify the file system hierarchy at any point-in-time and exploits parallel restoring in filesystem recovery. The experimental results show that parallel asynchronous objects sending makes the FV-CDP system max write ops to get improved by about 3.4 times, and the parallel recovery reduces file system recovery time by up to 57%. Under high frequency file syetem change workload, FV-CDP causes a large storage space overhead.

In this paper, we present a new way for evaluating the Continuous Data Protection (CDP) system. This method primarily describes the CDP systems in their snapshot views, which focus on the major issues affecting the CDP system's performance. To explain the application of the method, we present several representative snapshot views of the CDP systems and propose a model for analyzing the systems' performances. The main contribution of the present study is that we provide a methodological guide for CDP systems optimization.

With the development of human satellite observation system, the amount of satellite remote sensing data shows explosive growth. Massive remote sensing data storage capacity of the storage system, data availability and I / O performance and other aspects of the increasingly high demands. In this paper, we study the real-time storage technology of massive satellite data in the field analysis and processing of high real-time remote sensing data, and focus on the design of multi-thread data low-delay processing technology and complex exception processing technology Effectively guarantee the data analysis and processing of data on the continuous writing of the timeliness requirements. Experiments show that the proposed method can effectively improve the real-time access efficiency of massive satellite data.

It is secure for customers to store and share their sensitive data in the cryptographic cloud storage. However, the revocation operation is a sure performance killer in the cryptographic access control system. To optimize the revocation procedure, we present a new efficient revocation scheme which is efficient, secure, and unassisted. In this scheme, the original data are first divided into a number of slices, and then published to the cloud storage. When a revocation occurs, the data owner needs only to retrieve one slice, and re-encrypt and re-publish it. Thus, the revocation process is accelerated by affecting only one slice instead of the whole data. We have applied the efficient revocation scheme to the ciphertext-policy attribute-based encryption (CP-ABE) based cryptographic cloud storage. The security analysis shows that our scheme is computationally secure. The theoretically evaluated and experimentally measured performance results show that the efficient revocation scheme can reduce the data owner’s workload if the revocation occurs frequently.

This job focuses on the stroke regulation of a class of high-precision metering pumps. A parameter-tuning method of robust non-fragile PID (proportional-integral-derivative) controllers is proposed with the assumption that a PID controller has additive gain perturbations. An H -infinite robust PID controller can be obtained by solving a linear matrix inequality. This approach can guarantee that the closed-loop control systems is asymptotically stable and the H -infinite norm of the transfer function from the disturbance to the output of a controlled system is less than a given constant to attenuate disturbances. The simulation case shows that the control performance of the proposed strategy is significantly better than the traditional PID approach in the situation with perturbations of controller parameters.

Data protection has played an important role in maintaining the continuity and correctness of data processing application. We have designed the data protection architecture with data stream checking mechanism (DSCM) for copy-on-change. When the upper application requests a write to the disk, the DSCM computes the hash value of the data block by input the data stream in memory, and then compares the hash value with the pre-stored value, which has been generated in last data write request, in order to determine whether the data block need to be backup again. The experiments result shows that the DSCM can significantly reduce the disk storage overhead.

Multi-satellite data downlink resource scheduling is a complex combinatorial optimization problem. Current researches cannot handle the problem effectively when the satellite observation tasks increase over time. Considering the characteristic of the problem, a data downlink resource scheduling model adapting to observation task increments is established and a novel algorithm based on evolutionary computation is proposed. Finally, some experiments are conducted to validate the correctness and practicability of our algorithm.