Recent years have witnessed the emerging success of Graph Neural Networks (GNNs) for modeling graphical data. A GNN can model the spatial dependencies of nodes in a graph based on message passing through node aggregation. However, in many application scenarios, these spatial dependencies can change over time, and a basic GNN model cannot capture these changes. In this article, we propose a G raph S eq uence neural network with an A tt ention mechanism (GSeqAtt) for processing graph sequences. More specifically, two attention mechanisms are combined: a horizontal mechanism and a vertical mechanism. GTransformer, which is a horizontal attention mechanism for handling time series, is used to capture the correlations between graphs in the input time sequence. The vertical attention mechanism, a Graph Network (GN) block structure with an attention mechanism (GNAtt), acts within the graph structure in each frame of the time series. Experiments show that our proposed model is able to handle information propagation for graph sequences accurately and efficiently. Moreover, results on real-world data from three road intersections show that our GSeqAtt outperforms state-of-the-art baselines on the traffic speed prediction task.

Multi-task learning has been applied successfully in various applications. Recent research shows that the performance of multi-task learning methods could be improved by appropriately sharing model architectures. However, the existing work either identifies multi-task architecture manually based on prior knowledge, or simply uses an identical model structure for all tasks with a parameter sharing mechanism. In this paper, we propose a novel architecture search method to discover flexible and compact architectures for deep multi-task learning automatically, which not only extends the expressiveness of existing reinforcement learning-based neural architecture search methods, but also enhances the flexibility of existing hand-crafted multi-task learning methods. The discovered architecture shares structure and parameters adaptively to handle different levels of task relatedness, resulting in effectiveness improvement. In particular, for deep multi-task learning, we propose an architecture search space which includes a combination of partially shared modules at the low-level layer, and a set of task-specific modules with various depths at high-level layers. Secondly, a parameter generation mechanism is proposed to not only explore all possible cross-layer connections, but also reduce the search cost. Thirdly, we propose a task-specific shadow batch normalization mechanism to stabilize the training process and improve the search effectiveness. Finally, an auxiliary module is designed to guide the model training process. Experimental results demonstrate that the learned architectures outperform state-of-the-art methods with fewer learning parameters.

In era of big data, global shopping is very popular. Different from domestic shopping, global shopping requires the merchants to purchase goods from abroad in advance, thus knowing approximate sales of goods in the next period will be very helpful for merchants in determining the procurement amount. To solve this problem, in this paper, we are based on user performance data on commodity, commodity sales data and commodity promotion data to predict the commodity sales after 45 days. We extract two kinds of features, statistical feature and discrete feature and design a fusion model. This fusion model uses three different models, which is LR (Linear regression), XGBoost (Extreme Gradient Boosting) and LightGBM (Light Gradient BoostingMachine). The experimental results show that the prediction results of this model have a less bias compared to real value. It will provide some guidance for the decision of merchants.

Urban traffic passenger flows prediction is practically important to facilitate many real applications including transportation management and public safety. Sustained and rapid economic growth requires an orderly organization, and planning is an indispensable part of an orderly organization process. The reduction in travel efficiency due to traffic congestion, as well as energy and various pollution issues from the transportation sector, have become the bottleneck for the further development of the city and are the most troublesome topic for governments in all countries. Recently, deep learning performs the excellent ability to extract high dimensional spatial-temporal characters in regression and classification tasks. In this paper, we propose a deep learning model based on CNN and RNN, which takes matrixed traffic as input, uses CNN to extract traffic characteristics, and uses RNN to predict the evolution of features to achieve traffic flow prediction. Instead of traditional rnn models, we design a new type of RNN structure unit that can process time data in multiple time dimensions at the same time. Using a network-like RNN model, the evolution of traffic flow in different time dimensions is fully considered, and the interaction between different time dimensions is taken into account to predict the traffic flow of the target time series.The prediction of each data in the sequence has real data as input instead of merely taking the output of the previous moment as the input of the next moment.Experiments show that our model can significantly improve the prediction accuracy for real traffic passenger flow datasets.

Urban traffic passenger flows prediction has always been a great challenge in transportation field. Efficiently and correctly predicting the future flows of various regions can improve traffic resources scheduling and reduce the possibility of accidents. However, factors which affect the change of traffic passenger flows are complex, including interlaced lines and stations in large areas, diversified traveling demands for people, accidents and bad weathers. So the predicting algorithms or models should be more sensitive to multiply elements and their effecting patterns. Recently, deep learning performs the excellent ability to extract high dimensional spatial-temporal characters in regression and classification tasks. In this paper, we propose a new modeling method for urban traffic passenger flows. Instead of the grid matrices, we quantify the relationship between stations and represent it by a undirected graph. Then we sort the stations by their passenger flows and construct the two-channels graph flows matrices as the input of deep convolutional neural networks. To increase the temporal information of inputs, we also combine the input matrices with recent historical samples. In addition, we add date markers to correct the final prediction flows to further improve the accuracy. Finally we evaluate our model with the real Beijing subway data and compare with other traditional models on short-term passenger flows prediction tasks. Experiments show that our model including multidimensional flows graph matrices and the deep learning model can significantly improve the prediction accuracy.

The large amount of traffic data collected by urban traffic mobile terminals and sensing equipments provides us the opportunity to study group travel patterns and laws. In this paper, we built a New Transit Require Design Module (NTRDM) from the perspective of passenger flow conversion based on multi-source traffic data, which realized the adjustment and optimization of the current bus network. Specifically, CTDaaS was used for data fusion and processing to protect the passengers’ privacy. Then we established the NTRDM with minimum transfer time as the optimization goal, and proposed the Three-Step site adjustment method, which was solved with ant colony algorithm. Finally, we verified the calculating results with real data. Experimental results demonstrated the effectiveness of our method.

With the development of the city, the data amount is more and more large, the data source is more and more complex, the dimension of the data is getting bigger and bigger, how to find the potential value of these data and how to describe the multi-source data in a comprehensive way becomes a hot research point. The challenges of storing data separately and disparate understanding of different systems prevent integration. In order to overcome these problems, aiming at the data of these different sources, this paper adopts a data-as-a-service framework and is a service based on related data. This framework can integrate data from different sources and provide. Data services in a transparent manner. Consumers use data services without having to know the details. Our framework is transparent. Transparent integration of data resources, transparent data fusion and transparent data services. The data model pool and data resource pool can evolve because new data models and data sets are generated as data services are provided.

In order to improve the performance of tasks with dependencies in distributed environment and to overcome the shortcomings of existing table scheduling algorithms, the idea of table scheduling and task replication is combined to propose a heuristic task based on critical path and task replication Scheduling Algorithm (HCPTD). The algorithm improves the calculation method of task weight, and obtains the scheduling sequence according to whether it is the mission-critical or descending order of weight. The processor chooses the earliest task completion time and the shortest task-to-exit node distance. Experimental results show that HCPTD effectively improves the scheduling performance of distributed systems.

The first problem of traffic incident management is the detection and confirmation of traffic accidents. The detection method based on coil and video data is limited to practical application due to its high cost and insignificant detection effect. This paper presents a traffic incident detection algorithm based on outlier mining. The algorithm extracts characteristics of traffic event and builds a set of eigenvectors by using the new real-time traffic information released by NavInfo. The algorithm is simple, efficient and easy to deploy. Experimental results show that compared with traffic incident detection based on pattern recognition, the proposed algorithm has higher accuracy and can effectively distinguish between conventional congestion and traffic incidents.

We used kmeans algorithm to classify the drivers' traffic behaviors and combined the concrete effect to determine the k value. After the classification, we generated cluster label and then correlated the user information table to the cluster label. Finally, we analyzed the behaviors of the group user. The experimental results will be displayed and analyzed in a pie chart.

With the advancement of smart city, the development of intelligent mobile terminal and wireless network, the traditional text information service no longer meet the needs of the community residents, community image service appeared as a new media service. "There are pictures of the truth" has become a community residents to understand and master the new dynamic community, image information service has become a new information service. However, there are two major problems in image information service. Firstly, the underlying eigenvalues extracted by current image feature extraction techniques are difficult for users to understand, and there is a semantic gap between the image content itself and the user's understanding; secondly, in community life of the image data increasing quickly, it is difficult to find their own interested image data. Aiming at the two problems, this paper proposes a unified image semantic scene model to express the image content. On this basis, a collaborative filtering recommendation model of fusion scene semantics is proposed. In the recommendation model, a comprehensiveness and accuracy user interest model is proposed to improve the recommendation quality. The results of the present study have achieved good results in the pilot cities of Wenzhou and Yan'an, and it is applied normally.

Urban public transportation network is a typically complex network and the local fault of network often leads to serious systemic impact, causing cascading failure. Research on cascading failure of the bus network, is advantageous to understand of the potential key individuals of the network, so as to guide the rational planning of transit network. At first, this paper proposed the path navigation strategy based on the transfer bus to describe the flow propagation law of transit network, and after which an improved failure model of the bus network based on capacity-load model was put forward. Finally, the experimental analysis was conducted based on real traffic data of Beijing, and under the improved path navigation strategy, the correlation between node load and real data reached 99.61%, accord with the real law more than the traditional path navigation strategy. Cascading failure simulation illustrated that even small-scale attacks could lead to systemic paralysis, causing serious impacts on the structure and function of network, meanwhile the damage to the functional integrity is more severe than structural integrity. Conclusion is conducted that results can guide the bus lines and citizens' travel.

The invention relates to a data management service system based on large data. The data management service system comprises a heterogeneous data normalized-description module, a data semantization module, a data storage performance module, a data logic-management module, a data scenarization and service matching module and a data display module. The data management service system based on scenes solves the problems as follows: first, the data volumes are large at present, the data variety is large, the data is from various data sources, and the data categories and formats are rich; as a result, the problem of difficult storage is formed; second, the description of data heterogeneity: the data with multi-source large data forms data islands; different data structures exit in each data source, and at the same time, different designation systems also exist in each data source; the homogeneous data can also not interoperate; third, the data matching problem: the data matching problem is formed as the structures of data categories are different, the precision ratio and the recall rate are low, and the query cost is high.

“计算机组成原理”是计算机科学与技术专业核心课程，也是师生普遍反映难教难学的课程之一。本文简要介绍了“计算机组成原理”课程教学实施方案的总体框架，并结合科学型、工程型和应用型不同类型计算机人才培养的特点和要求，对“计算机组成原理”教学过程中重点难点的讲授、教学实验设置与要求等问题进行了详细的说明。

To improve the reliability of the servers which supply long-running and persistent computing and service,it is necessary to detect and analysis the performance degradation of the operation system which run in the server.In this paper,a system performance degradation detection method for the seasonal characteristic system of servers is proposed,which integrates the time-based method of degradation detection with metric-based one.The performance degradation indexes of each system running cycle are calculated by the new method base on AHP(Analytical Hierarchy Process).Then the method uses regression to analysis the time series of the indexes.The results will reflect the state of current system performance degradation and give the approximate downtime of the server in the future.Finally,a case will be introduced in the application of BRT(Bus Rapid Transit).

On the basis of fully analysing the characters of Bus Rapid Transit System,such as closed lane,platform selling and checking ticket,perfect intelligence system,studies how to solve the problem of optimizing the sending frequency of the prime sending plan,establishes the optimal model and then brings forward an algorithm to solve the model.According to the characters of BRT,focuses on using the self -adaptive genetic algorithm to calculate the ending inters.The algorithm is proved that the optimal model can raise the satisfaction of waiting for a bus and the comfortable of taking a bus.

"计算机组成原理"课程教学历来是计算机专业教学中的重点,本文分析了"计算机组成原理"课程的定位和特点,提出了一种从整体功能推进到局部组成、从微观实现抽象到宏观结构的层次化教学内容设计模式,探讨了"计算机组成原理"教学与学生能力培养之间的关系,探索并实践了一种研究性的教学方法。实践证明,这样的方法也的确收到了良好的效果。

The paper first analyzes the characteristic of predicting BRT vehicle travel time,and builds the prediction model.Then contraposed to the disadvantages of the tradition Kalman filter in predicting travel time,the paper presents an improved Kalman filter based on the fuzzy regression adaptive historical data samples of vehicle travel time.Finally the paper uses actual data collected from BRT Transport of South Axis Street in Beijing on Oct 9,2008 for experiment.The results show that the improved filter effectively reduces the error of the original algorithm.