ISSN 0253-2778

CN 34-1054/N

2015 Vol. 45, No. 10

Display Method:
Original Paper
Hospital outpatient visit analysis and forecasting using time series models
ZHU Shunzhi, WANG Dahan, HE Yanan, WNAG Yan
2015, 45(10): 795-803. doi: 10.3969/j.issn.0253-2778.2015.10.001
Analysis and forecasting of hospital outpatient visits are important in making correct and feasible decisions for hospital resources management and high quality patient care provision. However, research in outpatient visit analysis and forecasting has not drawn much attentions so far, and current research mainly focuses on the computational methods for forecasting only, lacking in comprehensive analysis, rules finding, and knowledge discovery for hospital outpatient visits. Thus it was propsed to construct autoregressive moving average models (ARMAX), neural network models, and hybrid models integrating ARMAX and NN for outpatient visit analysis and forecasting. By constructing these models, the rules of the daily outpatient visit of the Xiamen city, China were analyzed comprehensively. It was fund that outpatient visit data show a significantly upward time trend, a significant day-of-week effect, and a significant serial autocorrelation. By comparing the forecasting performance of these time series models, it was fund that the ARMAX+NN hybrid model achieves better performance, which is mainly due to the fact that the hybrid model can capture both linear and nonlinear parts of the outpatient visit data.
A novel combination recommendation method for solving sparse and cold start problems
Guo Xiaobo, Zhao Shuliang, Niu Dongpan, Wang Changbin, Pang Huanli
2015, 45(10): 804-812. doi: 10.3969/j.issn.0253-2778.2015.10.002
Considering the problems resulting from the traditional recommended approaches which are powerless to address the well-known cold-start and data sparseness, and the fact that most currently existing association rule mining(ARM) algorithms were designed with basket-oriented analysis in mind, which are inefficient for collaborative recommendation because they mine many rules that are not relevant to a given user, this paper introduces a novel association recommendation method based on combination similarity, and proposes a solution to the cold start problem by combining association rules and collaborative filtering techniques. The proposed method focuses on mining rules for only one target user or target item at a time, while utilizing the interest factor to balance the weight between active users (or items) and non active users (or items), which in order to recommend an optimal solution (rules) via weighted method. To recommend both high ratings and collection of items with high similarity, the similarity measurement method was used to filter low similarity items, and to provide the final results by combining the association rules and CF recommendation, realizing user-based or item-based collaborative filtering recommendation. Experiments on the MovieLens data set reveals that the results obtained from employing this method has significantly better than the publishecl results and that it is better able to deal with sparse data and cold start problems.
A composite index strategy for big marine data based on adaptive method of data merging strategy
HUANG Dongmei, SUN Le, ZHAO Danfeng
2015, 45(10): 813-821. doi: 10.3969/j.issn.0253-2778.2015.10.003
Marine data fall easily into category of Big Data. A basic requirement for various marine monitoring applications is quick retrieval and the establishment of a sound index structure is of great importance. A multi-layer index (ML-index, for short) with regard to time interval B+-tree and hybrid space partition tree (HSP-tree, for short) was proposed. It employs the adaptive method of data merging strategy to optimize the primary key index (i.e. B+-tree). An adaptive space partition method was also proposed on the basis of data characteristics, and data unit capacity particular, for building secondary index, namely, HSP-tree. The experiment result shows that ML-index saves about 2/3 of the time in comparison with two state-of-the-art index methods.
An EMD-based method for assessing and analyzing coal consumption of coal-fired units
SUN Hong, SUN Shuanzhu, ZHOU Chunlei, DAI Jiayuan, SUN Bin, WANG Qixiang
2015, 45(10): 822-828. doi: 10.3969/j.issn.0253-2778.2015.10.004
Operation habits of coal-fired unit workers often affect significantly the level of coal consumption. In practice, different operation teams usually consume different amounts of coal, thus it is necessary to carry out benchmarking management to guide the workers towards optimal operations. Conventionally, the level of coal consumption within a time period is evaluated by statistics such as the mean, the minima, and the maxima. However, these simple statistics are not informative enough to reveal the operation habits of workers. an EMD-based assessment method is proposed to analyze operation habits of teams. By taking the coal consumption distribution of the best operation team (i.e., the one with minimal amount of total coal consumption) as the baseline comparator, the method first estimates, for non-optimal teams, how far away their operations are from the optimum, and then finds out the primary parameters that cause the difference. Based on such an analysis, suggestions are provided to the operation teams, such that their behaviors can be adjusted. The proposed method achieves accurate assessment of coal consumption and may provide an analytical method for electric power enterprises. to save energy and reduce coal consumption. And it also may provide technical support for government supervision departments to carry out fine benchmarking management and performance evaluation of energy-saving for coal-fired units.
Discovery of hot regions about crowd activities based on mobility data
BAN Leiyu, HUO Huan, XU Biao
2015, 45(10): 829-835. doi: 10.3969/j.issn.0253-2778.2015.10.005
Mobility data records the change of location and time about crowd activities, showing semantic knowledge about human mobility. From the perspective of regional semantic knowledge, mining the hot regions visited frequently by moving crowds is essential to understand regional characteristics in the smart city applications. This paper studied how to discover hot regions and how to constraint their coverage size. Based on an analysis of the location sequence of moving crowd, a discovery method for discovering hot regions based on kernel function was proposed. This discovery method uses the grid as a spatial data indexing structure and the Top-k sorting method. A discovery algorithm of hot regions was presented based on the discovery method. Finally, experimental results validate accurately the feasibility and effectiveness of the method on practical datasets.
PipelineJoin:A new MapReduce-based multi-table join algorithm
LIN Ziyu, LI Yuqian, LI Can, LAI Yongxuan
2015, 45(10): 836-845. doi: 10.3969/j.issn.0253-2778.2015.10.006
MapReduce, a parallel and distributed computing model, has been widely used to process join operations for two or more large tables. The existing MapReduce-based multi-table join algorithms all have some limitations when dealing with chain join. Some methods can not process join operations for multi large tables, and others involve sequentially running too many MapReduce tasks, which leads to low efficiency. Here a new MapReduce-based multi-table join algorithm, PipelineJoin, is proposed to process chain join of a number of tables. PipelineJoin adopts a pipeline model and a scheduler to allow the overlapping execution of a series of Map tasks and Reduce tasks in the whole join process so as to enhance the efficiency of multi-table join, while effectively overcoming the deficiency of the existing methods. Extensive experimental results based on various synthetic datasets show that the proposed algorithm can greatly reduce join operation time compared with the existing chain join algorithms.
An intelligent tutoring platform for educational assessment
HUANG Zhenya, SU Yu, WU Runze, LIU Yuping, LIU Qi, CHEN Zhigang, HU Guoping
2015, 45(10): 846-854. doi: 10.3969/j.issn.0253-2778.2015.10.007
K-12 education is an important part of educational psychology. Recently, online-learning has been widely accepted because of its significant effect on K-12 education. However, mostly based on the educational database, though existing online-learning systems and intelligent tutoring systems can provide useful resources for teachers and students, they seldom utilize offline test data to offer personalized services. To the end, an intelligent tutoring platform for educational assessment(ITPEA) was proposed and implemented. The platform combines both offline examinations and online resources to offer analysis from the perspectives of test papers, students and teachers. Specifically, based on the data from offline examinations, educational theories were first employed to evaluate the quality of the questions. Then, diagnostic models were constructed on the key knowledge points that students should master to meet one’s personalized demands. Finally, new analytical methods for evaluating teacher’s influence on class abilities were proposed with data mining technologies to help find "unusual" students in class. The core technology of ITPEA is currently operating on an online-learning system, and obtains good results.
Power measurement based on VSLMS improved adaptive filters
2015, 45(10): 855-863. doi: 10.3969/j.issn.0253-2778.2015.10.008
A new algorithm is proposed to detect and extract in real-time signals with fundamental and harmonic wave components in the power grid, which is applicable to power measurement. The proposed method is based on the concept of adaptive filter, and adaptively decomposes the measured power signal into its constituting components, resulting in a fast convergence rate. The fundamental and harmonic wave components in the power grid can be decomposed into a series of sinusoidal signals. The frequency of the power grid is measured by energy operator. A model of voltage and current wave in the power grid is constructed, a new step-size LMS algorithm for improving the adaptive filter is proposed and the stability of the proposed method is discussed. The effectiveness of the proposed method is demonstrated by simulation examples.
Grid-like radar detection based on the distribution of key points
DU Binbin, LING Qiang, LI Feng, SUN Tao
2015, 45(10): 864-870. doi: 10.3969/j.issn.0253-2778.2015.10.009
Grid-like radars have been widely used for military applications, and their detection is of great importance. A novel method is proposed to detect grid-like radars even with large appearance variation. In our method, the key points of grid-like radars are first treated as small objects and detected by the classical sliding window method. Then a possible radar area is located based on the distribution density of the detected key points. Finally the decision regarding the presence/absence of grid-like radars will be made based on the spatial distribution relation of the detected key points. Experiments were done on our dataset, including 42 grid-like radar images and 154 non-radar images, and our approach achieved a 7.1% miss rate and 12.3 FPR(false positive rate). The method based on the distributions of key points is more robust against the appearance variation caused by the types of radar, deformations and viewpoint changes, and demonstrates better performance than classical method, such as “BOF+SIFT” and “HOG”.
Consensus of second-order multi-agent systems with directed topologies and time-varying delays
OU Meiying, GU Shengwei, DONG Kexiu
2015, 45(10): 871-880. doi: 10.3969/j.issn.0253-2778.2015.10.010
The consensus problem of second-order multi-agent systems with directed topologies and time-varying delays is investigated,