scispace - formally typeset
Search or ask a question
Author

Kun Niu

Bio: Kun Niu is an academic researcher from Beijing University of Posts and Telecommunications. The author has contributed to research in topics: Computer science & Mobile phone. The author has an hindex of 4, co-authored 15 publications receiving 57 citations.

Papers
More filters
Journal ArticleDOI
TL;DR: A novel neural network is proposed, named L-CNN based on CNN and LSTM, and an effective real-time prediction model is developed to forecast the most likely potential passenger for taxi drivers and the results prove the efficiency of the predicting system.
Abstract: The GPS trajectories are rich with potential information that could be used to explore the regulation of traffic to serve the public. While that past approaches for short-term traffic prediction have existed for some time, emerging smart transportation technologies require the traffic prediction capability to be both fast and scalable to full urban networks. In this paper, we propose a novel neural network, named L-CNN based on CNN and LSTM, and develop an effective real-time prediction model to forecast the most likely potential passenger for taxi drivers. It is noteworthy that our model can be easily extended to other real-time traffic prediction problems, such as road traffic and flow prediction. Finally, we test our method based on GPS trajectories generated by Cheng Du taxi. The method presented provides passenger prediction over 15-min intervals for up to 1 h in advance and the results prove the efficiency of our predicting system.

29 citations

Journal ArticleDOI
TL;DR: A novel spatio-temporal model named L-U-Net based on U-Net as well as long short-term memory architecture is proposed and an effective speed prediction model is developed, which is capable of forecasting city-scale traffic conditions.
Abstract: City-scale traffic speed prediction provides significant data foundation for the intelligent transportation system, which enriches commuters with up-to-date information about traffic condition. However, predicting on-road vehicle speed accurately is challenging, as the speed of the vehicle on the urban road is affected by various types of factors. These factors can be categorized into three main aspects, which are temporal, spatial, and other latent information. In this paper, we propose a novel spatio-temporal model named L-U-Net based on U-Net as well as long short-term memory architecture and develop an effective speed prediction model, which is capable of forecasting city-scale traffic conditions. It is worth noting that our model can avoid the high complexity and uncertainty of subjective features extraction and can be easily extended to solve other spatio-temporal prediction problems such as flow prediction. The experimental results demonstrate that the prediction model we proposed can forecast urban traffic speed effectively.

27 citations

Proceedings ArticleDOI
01 Aug 2016
TL;DR: This paper defines Logic File Name (LFN) and proposes the Small file Merge Strategy Based LFN (SMSBL), a new idea and a new perspective on hierarchy that improves the correlation of small files in the same block of HDFS effectively based different file system hierarchy.
Abstract: Hadoop Distributed File System (HDFS) is designed for reliable storage and management of very large file and low-cost storage capability. As HDFS architecture based on master (NameNode) to handle metadata for multiple slaves (DataNode), NameNode often becomes the bottleneck, especially when handing large number of small files. It is a common solution to merge many small files into one big file about this problem. To solve the large small files problem and improve the efficiency of accessing small files, in this paper, we define Logic File Name (LFN) and propose the Small file Merge Strategy Based LFN (SMSBL). SMSBL is a new idea and a new perspective on hierarchy, it improves the correlation of small files in the same block of HDFS effectively based different file system hierarchy, so the performance is amazing facing large small files when HDFS adopted SMSBL with prefetching mechanism. The system efficiency analysis model is established and experimental results demonstrate that SMSBL can solve small file problem in HDFS and has appreciable high hit rate of prefetching files.

16 citations

Proceedings ArticleDOI
06 Jan 2017
TL;DR: Experimental results proved that FMA is more effective than Apriori on time consuming with similar accuracy.
Abstract: Apriori is the most famous frequent pattern mining method. It scans dataset repeatedly and generate item sets by bottom-top approach. In order to reduce time complexity, we proposed a modified algorithm named as Frequent Matrix Apriori (FMA). Firstly, FMA scans the dataset only once to store frequent item information in a frequent matrix. Then, FMA discretize the matrix by the minimum support parameter which is generated automatically. Thirdly, it scans the discretized frequent matrix and find the most frequent item sets recursively. Experimental results proved that FMA is more effective than Apriori on time consuming with similar accuracy.

6 citations

Proceedings ArticleDOI
01 Aug 2016
TL;DR: K-means+ is presented, which effectively reduces time costs of clustering modeling through block operation and redesigning of distance function, which aggregates instances as blocks to cluster afterwards.
Abstract: Clustering is one of the most important task in data mining. But for big data application, clustering models are faced with the problem of high complexity for low respond time requirement. This paper focuses on velocity criterion of big data modeling, presents a developed k-means algorithm, k-means+, which effectively reduces time costs of clustering modeling through block operation and redesigning of distance function. Block operation aggregates instances as blocks to cluster afterwards. Manhattan distance is used instead of common Euclidean distance to simplify calculation. Experimental results show that k-means+ works well on most testing datasets and executes much faster than original k-means.

5 citations


Cited by
More filters
Journal ArticleDOI
TL;DR: Predictive results show that the Prophet model and Keras stateful LSTM perform better than neural network models, where the optimal size of the training data is found to be three years.
Abstract: Big data analytics (BDA) is a systematic approach for analyzing and identifying different patterns, relations, and trends within a large volume of data. In this paper, we apply BDA to criminal data where exploratory data analysis is conducted for visualization and trends prediction. Several the state-of-the-art data mining and deep learning techniques are used. Following statistical analysis and visualization, some interesting facts and patterns are discovered from criminal data in San Francisco, Chicago, and Philadelphia. The predictive results show that the Prophet model and Keras stateful LSTM perform better than neural network models, where the optimal size of the training data is found to be three years. These promising outcomes will benefit for police departments and law enforcement organizations to better understand crime issues and provide insights that will enable them to track activities, predict the likelihood of incidents, effectively deploy resources and optimize the decision making process.

73 citations

Posted Content
TL;DR: This paper identifies two essential spatial dependencies in traffic forecasting in addition to distance, direction and positional relationship, for designing basic graph elements as the smallest building blocks, and suggests DDP-GCN (Distance, Direction, and Positional relationship Graph Convolutional Network) to incorporate the three spatial relationships into prediction network for traffic forecasting.
Abstract: Traffic speed forecasting is one of the core problems in Intelligent Transportation Systems. For a more accurate prediction, recent studies started using not only the temporal speed patterns but also the spatial information on the road network through the graph convolutional networks. Even though the road network is highly complex due to its non-Euclidean and directional characteristics, previous approaches mainly focus on modeling the spatial dependencies only with the distance. In this paper, we identify two essential spatial dependencies in traffic forecasting in addition to distance, direction and positional relationship, for designing basic graph elements as the smallest building blocks. Using the building blocks, we suggest DDP-GCN (Distance, Direction, and Positional relationship Graph Convolutional Network) to incorporate the three spatial relationships into prediction network for traffic forecasting. We evaluate the proposed model with two large-scale real-world datasets, and find 7.40% average improvement for 1-hour forecasting in highly complex urban networks.

48 citations

Journal ArticleDOI
24 Jan 2020
TL;DR: Simulation results illustrate that the proposed LSTM-based method can extract spatial and temporal traffic features of hotspot with higher accuracy, compared with some existing deep and non-deep learning approaches.
Abstract: To meet the extremely stringent but diverse requirements of 5G, cost-effective network deployment and traffic-aware adaptive utilization of network resources are becoming essential. In this paper, a hotspot prediction based virtual small cell (VSC) operation scheme is adopted to improve both the cost efficiency and operational efficiency of 5G networks. This paper focuses on how to predict the hotspots by using deep learning, and then demonstrates how the predictions can be leveraged to support adaptive beamforming and VSC operation. We first leverage the feature extraction capabilities of deep learning and exploit use of a long short-term memory (LSTM) neural network to achieve hotspot prediction for the potential formation of the VSCs. To support the operation of VSCs, large-scale antenna array enabled hybrid beamforming is adaptively adjusted for highly directional transmission to cover these hotspot-based VSCs. Within each VSC, an appropriate user equipment is selected as a cell head to collect the intra-cell traffic in the unlicensed band and relays the aggregated traffic to the macro-cell base station by using the licensed band. Our simulation results illustrate that the proposed LSTM-based method can extract spatial and temporal traffic features of hotspot with higher accuracy, compared with some existing deep and non-deep learning approaches. Numerical results also show that VSCs with hotspot prediction and hybrid beamforming can improve the energy efficiency dramatically with flexible deployment and low latency, compared with the scenario of the convolutional fixed small cells.

39 citations

Journal ArticleDOI
TL;DR: The value of trajectory data in understanding on-demand services is highlighted, the procedures of retrieving information for the demand part and the supply part from raw trajectory data are discussed, and four types of factors that influence the spatial-temporal patterns of demands are summarized.
Abstract: With the development of information technique and wireless communication, a vast number of taxis' and ride-sharing cars' trajectory data that provide a rich and detailed source to study on-demand services have been collected. The increasing available trajectory data bring benefits and new challenges to the studies of on-demand services. To provide an overview of the benefits and challenges brought by the trajectory data, we provide a survey on recent studies of trajectory analysis (refer to analyzing trajectory datasets) for on-demand services in this paper. Our purposes are at least trifold. First, we highlight the value of trajectory data in understanding on-demand services and discuss the procedures of retrieving information for the demand part and the supply part from raw trajectory data. Second, we categorize related studies into three parts (the demand part, the supply part, and the mixed part) and review the significant findings. For the demand part, we focus on the models proposed for describing and explaining the spatial-temporal characteristics of observed trips. Methods or models proposed for describing trip statistics, scaling laws of trips, and dynamics of ridership are reviewed. We summarize four types of factors that influence the spatial-temporal patterns of demands. For the supply part, we focus on the models proposed for describing the spatial-temporal characteristics of available taxis/ride-sharing cars and modeling the behavior of drivers (i.e., passenger-search behavior and route choice behavior) to explain the spatial-temporal patterns of taxi/ride-sharing supplies. For the mixed part, we focus on studies that apply the uncovered demands/supplies patterns to design recommendation systems and pricing strategies. Third, we discuss the future directions on collecting/releasing trajectory data and future research directions to advance the understanding of on-demand services.

32 citations