Current Issue

• 2015, No. 12　Published:25 December 2015 Last issue    Next issue
  Columns in this issue: 论文

• 论文

  State of the art analysis of China HPC 2015 

  YUAN Guoxing1,YAO Jifeng2

  2015, 37(12): 2195-2199. doi:

  Abstract ( 297 )   PDF (780KB) ( 590 )   　　

  In this paper,according to the latest China HPC TOP100 rank list released by SAMSS in early November,we present the overall performance trends of China HPC TOP100 and TOP10 of 2015 from the aspects of overall performance,manufacturers,and application areas.

  Comparison of machine learning methods
  for disk failure prediction   

  DONG Yong，JIANG Yanhuang，LU Yutong，ZHOU Enqiang

  2015, 37(12): 2200-2207. doi:

  Abstract ( 266 )   PDF (702KB) ( 633 )   　　

  As disk is one of the most important data storage device, it is significant to improve disks’ reliability and data availability. Modern disks adopt the SMART protocol to monitor the internal operating status. We employ machine learning methods, including backpropagation neural networks, decision tree, supported vector machine and nave Bayes to analyze the SMART data of disks, which can predict disk failures. Real SMART data of disks are used in experiments to validate and analyze the effectiveness of those methods, and the effectiveness of different methods is compared. The results show that the decision tree method has best prediction rate while the supported vector machine method has the lowest false alarm rate.

  A dynamic SDRAM page policy based on
  advance information and prediction mechanism  

  Lv Hui,XIE Xianghui

  2015, 37(12): 2208-2215. doi:

  Abstract ( 120 )   PDF (1093KB) ( 216 )   　　

  We present a novel dynamic SDRAM page policy based on prediction mechanism and advance information. The policy makes full use of the addresses of memory accesses to accurately determine whether the next memory access is a page hit or not. If there is not a pending memory access, the policy predicts whether a page hit happens or not according to history records. Theoretical analysis shows that the policy can be easily implemented at low hardware cost. Experimental results confirm that all the three main categories of dynamic page policies based on advance information achieve the desired memory bandwidth. Under the best circumstances, the actual memory bandwidth is improved by 42%. Among the three policies, the policy based on both prediction mechanism and advance information is the best.

  Tide-bound water level computing and
  visualization platform based on Spark 

  QIN Bo1，ZHU Yong1，QIN Xue2

  2015, 37(12): 2216-2221. doi:

  Abstract ( 140 )   PDF (755KB) ( 261 )   　　

  Tidebound water level computing is an important part of ocean environment information processing, which features huge amount of data, high complexity, and prolonged computing time. The traditional computing model implemented by HPC has a number of problems, such as high computation cost, poor scalability and interactivity. Aiming at all these problems, we propose an interactive computing and visualization platform based on the Spark scheduling algorithm. We design a computing capacity scheduling algorithm, realize the parallel processing of largescale tidebound water level data, such as data retrieval, data extraction, numerical calculation, featurebased visualization, and achieve the purpose of parallel processing and visualization of largescale ocean environmental data on Spark. Experimental results show that the computing and visualization platform based on Spark can improve the traditional computing model, lessen the dependence of tidal level calculation on high performance cluster and reduce computation cost. In addition, the newlydeveloped task scheduling algorithm can make task allocation more rational and scientific, and therefore further enhance its efficiency. In conclusion, the proposed platform provides a new method for tidebound water level computing.

  A 1 GHz multi-port low-power register file design  

  LI Jiao1,2,WANG Lianghua1,BI Zhuo1,3,LIU Peng1

  2015, 37(12): 2222-2227. doi:

  Abstract ( 117 )   PDF (719KB) ( 333 )   　　

  Register files in superscalar processors usually adopt the multiport structure to support the wide issue, however, this structure brings in problems such as prolonging access speed, increasing in silicon areas and higher power consumption.We design a 64*64 bit multiport register file which can concurrently accomplish 8 read operations and 4 write operations in one single clock cycle.We improve the conventional singleended memory cell structure and purpose a new structure, which combines the powergating and the bitline floating techniques, and the transmission gate is used in all ports to accelerate the access speed.Simulations are conducted on Hspice with PTM 90 nm, 65 nm, 45 nm and 32 nm technology models compared with the conventional singleended structure, the proposed method  can significantly improve the performance of register files, the delay of write logic 1 decreases more than 32%, and the total power consumption decreases more than 45%; the stability of memory cells is also improved.

  A thermal design for high density storage servers  

  SHA Chaoqun1,YOU Yang2,HU Changjun1,ZHENG Chenming1,LIU Xingkui3

  2015, 37(12): 2228-2232. doi:

  Abstract ( 171 )   PDF (768KB) ( 397 )   　　

  Big data and cloud computing applications have a booming demand for high density storage servers by  in recent years. Since temperature has a great influence on the performance and lifecycle of electronic components, and high density storage servers have higher power density, desired cooling solution design is in great need, so the servers can work at a proper temperature. We introduce a thermal design method for high density storage servers, in which the forced air cooling method is adopted. The thermal simulation software Flotherm is used to guarantee the cooling effect. We construct an engineering project and test the proposal under 35℃ ambient temperature. The results show that this solution can meet the thermal design requirements.

  A triple modular eager redundancy faulttolerant
  technique for distributed stream architecture 

  LI Xin1，3，4,LIN Yufei2,GUO Xiaowei1

  2015, 37(12): 2233-2241. doi:

  Abstract ( 99 )   PDF (896KB) ( 232 )   　　

  As computing systems continue to expand in size in the Internet environment, the reliability of the distributed stream architecture is facing serious challenges. Based on the Nmodular redundancy technique, we propose a triple modular eager redundancy faulttolerant method for the distributed stream architecture (TREFT). The TREFT employs three program copies to run the error detection and error correction processes efficiently. Experimental results on a prototype system of the distributed stream architecture show that the TREFT could enhance the reliability of the system at very low cost, increasing the faulttolerant cost by 10.77% on average.

  A fine grained grid-based maritime traffic density
  algorithm for mass ship trajectory data   

  NING Jianqiang,HUANG Tao,DIAO Boyu,ZHAO Ruilian,BI Jingping

  2015, 37(12): 2242-2249. doi:

  Abstract ( 138 )   PDF (793KB) ( 327 )   　　

  The attention to maritime trades is heating up with the strategic plan of “One Belt, One Road”. Maritime transport is significant for maritime trades, and thus getting a finegrained maritime traffic density is of great importance for extracting hot spots, analyzing global trade trends, inferring maritime traffic connectivity and handling maritime anomaly detection. We propose a finegrained gridbased maritime traffic density calculation method for massive ship trajectory data, which divides the global area into a standard finegrained grid of 0.01 * 0.01 degrees. On the one hand, in order to be adaptive to the granular characteristics of global ship real trajectory data, we design a corresponding pretreatment solution to deal with abnormal points and mooring. On the other hand, we design and implement a calculation method for multidimensional traffic density features utilizing mass trajectory data of the grid, including dynamic data count, message sending interval, vessel count, and the sum of crossing time, which shows good scalability. Experiments on more than 800 million records of two months in 2014 verify the desirable performance, accuracy and feasibility of the proposal.

  Data transmission scheme of realtime control CPS
  based on software defined resources 

  JIANG Jun，HUANG Chuanhe，HUA Chao，HU Haiqiao，PENG Hui

  2015, 37(12): 2250-2255. doi:

  Abstract ( 131 )   PDF (543KB) ( 202 )   　　

  CyberPhysical System (CPS) is currently one of the forefronts of cross research fields. It has wide application prospect, and its research and development cannot continue without the support of data transmission and management technology. In this paper, we propose a model of CPS based on software defined resources. The allocation and scheduling of resources is implemented through the way of using mapping resource description files. On the basis of this model, we present a data transmission mode of controlling the CPS in real time. This data transmission mode is implemented by middleware. This middleware employs a kind of serialization and deserialization methods in CPS, which is able to meet the requests of crossplatform, across networks and realtime communication in CPS. Besides, the paths of data transmission in CPS are designed in order to improve the efficiency of communication. Experimental results show that the serialization mechanism of data transmission is feasible.

  Public opinion analysis and implementation of network
  hot events based on value added theory 

  XU Yong

  2015, 37(12): 2256-2261. doi:

  Abstract ( 118 )   PDF (803KB) ( 283 )   　　

  We analyze the evolution of the network hot events based on the value added theory. The emergence of the precipitating factors, the formation of generalized belief, and the completion of mobilization for actions are three crucial phases of the evolution of the events. We also design a monitoring model to analyze sensitive factors, the formation and diffusion of the emotions of public opinions in the events. Based on this, we develop a network public opinion intelligent monitoring system (NPOIMS). And the public opinion monitoring system is built up by  takes the X city in the west of China as the instance object. All kinds of public opinion information of the X city are monitored. The hot words are refined, and the public opinion is analyzed. And the analysis results are provided to the government departments as a decision support to guard public opinions and dispose the events.

  Solving non-Lipschitz optimization problems by
  smoothing neural networks  

  YU Xin,XIE Mian,LI Chenyu

  2015, 37(12): 2262-2269. doi:

  Abstract ( 125 )   PDF (635KB) ( 184 )   　　

  In order to seek to optimal solution satisfying the necessary conditions of optimality, aiming at the optimization problems that objective functions are nonLipschitz and the feasible region consists of linear inequality or nonlinear inequality, we design a new smooth neural network by the penalty function method and the smoothing approximate techniques which convert nonsmoothing objective functions into smoothing functions. Detailed theoretical analysis proves the uniform boundedness and globality of the solutions to smooth neural networks, regardless of the initial points inside or outside of the feasible domain. Moreover, any accumulation point of the solutions of to smooth neural networks is a stationary point of the optimization promble. Numerical examples also demonstrate the effectiveness of the method.

  BP neural network incorporating selfadaptive differential
  evolution algorithm for time series forecasting 

  WANG Lin1,PENG Lu1,XIA De2,ZENG Yi1

  2015, 37(12): 2270-2275. doi:

  Abstract ( 208 )   PDF (517KB) ( 499 )   　　

  It is easy for a BP neural network (BPNN) to be trapped into a local minimum point for the time series forecasting problem. To improve the forecasting accuracy, we design a hybrid algorithm which combines the selfadaptive differential evolution algorithm (SDE) with the BPNN. We adopt the SDE algorithm to search for global initial weights and thresholds of the BPNN. These values are then employed to further search for the optimal weights and thresholds. The performance of the proposed SDE algorithm is  verified through benchmark functions and a wellknown real data set is used to verify the effectiveness of the hybrid algorithm. Compared with general neural network, ARIMA and other hybrid models,experimental results indicate that the proposed algorithm can be an effective way to improve forecasting accuracy.

  Multi-objective and multidisciplinary design
  optimization of satellite payload 

  ZHAO Hongwei，LIU Bo，XIE Guangqian，LIU Heng

  2015, 37(12): 2276-2281. doi:

  Abstract ( 137 )   PDF (607KB) ( 236 )   　　

  We present an approach to realize multiobjective and multidisciplinary design optimization (MDO) of satellite payload. We first analyze the key techniques in multiobjective and multidisciplinary design optimization  of satellite payload system design, and establish  multidisciplinary analysis models including  antennas, transponders, data transmission, reliability, costs and qualities. Then multiobjective genetic algorithm is adopted to optimize the reliability and cost of the satellite payload system and the Pareto optimal set is obtained. Collaborative optimization which is one of the best feasible multidisciplinary design optimization approaches is integrated with the genetic algorithm to carry out the reliability optimization of single objective design. Results show that relationship among different disciplines are well considered during the multiobjective and multidisciplinary design optimization. Designers can choose satisfactory optimization results from the Pareto set according to their own specific requirements, design efficiency is therefore significantly improved. Collaborative optimization can help realize discipline autonomy and parallel design, improve design flexibility and reduce design cycle.

  Automatic recognition of the absent topics in Chinese
  punctuation clauses based on maximum entropy model 

  LU Dawei1,SONG Rou2

  2015, 37(12): 2282-2293. doi:

  Abstract ( 86 )   PDF (619KB) ( 191 )   　　

  We focus on the task of the automatic recognition,which identify whether an absent topic of a punctuation clause is the subject or object of its previous sentence. We regard this task as the pointcut of the automatic recognition of absent topics in Chinese punctuation clauses. Several literal features and semantic features are summerized to achieve this task by combining the rules and the maximum entropy model. Experimental results show that Fscore of this recognition approach reaches 82% for the samples of some specific verbs. Experimental results analysis shows that verb features and semantic features play the most important role in the recognition process; neither rules nor statistics can be neglected, and refined knowledge has great influence on the performance of the recognition .

  Automatic extraction and mapping directionality of
  synaesthetic sentences of modern Chinese  

  LIU Hongchao,Francesca Striklievers,HUANG Churen

  2015, 37(12): 2294-2299. doi:

  Abstract ( 123 )   PDF (493KB) ( 178 )   　　

  This paper focuses on the extraction and mapping tendencies of synaesthetic sentences in modern Chinese. The extraction applies two kinds of methodologies both based on the perception related word lists. We have constructed five sense word lists of touch, taste, smell, hearing and vision respectively. By checking each list and extracting the sentences with two or more kinds of perception related words, the accuracy of this methodology is 20.78%; by introducing POS distributing tendencies checking, the accuracy rises to 46.37%. The difficulty lies in collecting and further selecting the perception related word and also in observing the POS distributing rules of each perception related word. Finally, we check the mapping directionality of one domain of sense to another one.

  Neural representation and processing of Uyghur morphologically
  complex words: evidence from psycholinguistics method    

  Abudoukelimu·Abulizi1，2,JIANG Minghu1，2，YAO Dengfeng1，2,Halidanmu·Abudukelimu3

  2015, 37(12): 2300-2305. doi:

  Abstract ( 120 )   PDF (461KB) ( 178 )   　　

  In this behavioral study, we adopt the lexical decision paradigm to investigated Uyghur inflectional and derivational word processing of mental lexicon. Our experiments consist of two parts: inflectional word processing for experiment 1, and derivational word processing for experiment 2.In experiment 1,the results support to the view that morphological parsing takes place during the inflected word processing. Contrary to experiment 1, experiment 2 revealed that processing cost of the derived words is almost the same with the monomorphemic words. Our experimental results show that the processing of the inflected and derived words is independent from each other, in which a Uyghur speaker represents and accesses inflected words in a morphologically decomposed form, while derived words are accessed as single entities.

  Construction and its recognition
  of Chinese relevant event  

  HUANG Yilong,LI Peifeng,ZHU Qiaoming

  2015, 37(12): 2306-2311. doi:

  Abstract ( 117 )   PDF (439KB) ( 230 )   　　

  There are many relevant events concerning a topic. In the era of big data, extracting those events which are relevant to a specific topic is helpful for many natural language processing applications, such as information extraction, text summarization, and text generation. We propose a method to annotate relevant events and construct a Chinese relevant event corpus. We then put forward a relevant event recognition approach based on various distances and semantic features. Experimental results on the annotated corpus show that the proposed approach outperforms the baseline by 4.08% in F1-measure.

  Research on Chinese irony detection in microblog  

  DENG Zhao1,JIA Xiuyi1,CHEN Jiajun2

  2015, 37(12): 2312-2317. doi:

  Abstract ( 148 )   PDF (664KB) ( 346 )   　　

  Irony detection has drawn much attention in recent years. However, most studies are based on foreign languages. In this paper, we focus on the Chinese irony detection in microblog. Considering the characteristics of both Chinese language and social networks, we build a set of discriminating features for Chinese irony detection. Information gain is applied to these six features to compare their efficiency, and several classifiers are employed to test their stability. Experimental results prove the efficiency of the proposed features.

  Study of modern Uyghur word stem POS tag set    

  Azragul1,2,Mirxat3,Yusup·Abaydula1

  2015, 37(12): 2318-2323. doi:

  Abstract ( 83 )   PDF (550KB) ( 184 )   　　

  Taking the Uyghur word stem POS tagging of the Uygur language textbooks which are in use in primary schools as the verification object, we validate the feasibility, adaptability and reliability of Modern Uyghur Word Stem POS Tag set which is made from the perspective of grammatical semantic combination. We first describe the electronic corpus of primary school Uyghur language textbooks; secondly, we discuss the basic situation of "the partofspeech and tagging set standards of modern Uyghur word stem information processing", and the design and algorithm of multistrategy modern Uyghur Words Stem tagging system model; finally, we analyse the experimental results, validate the scientificity of Modern Uyghur Word Stem POS Tag set, supplement and correct parts of the semantic classification and codes and recommend a substantial expansion of the standard.

   

  Recognition of Chinese prosodic phrases
  based on AdaBoost-SVM algorithm and chunk information

  QIAN Yili1，2,FENG Zhiru1

  2015, 37(12): 2324-2330. doi:

  Abstract ( 133 )   PDF (497KB) ( 245 )   　　

  We propose a recognition method for Chinese prosodic phrases based on Chunk and the AdaBoostSVM algorithm. Firstly, the initial chunks are marked on the corpus of automatic word segmentation and the part of speech tagging, and then they are merged using the rules based on the closeness between initial Chunks. Secondly, based on the block structure and the AdaBoostSVM integrated algorithm, a Chinese prosodic phrase recognition model is constructed. Meanwhile we utilize various algorithms to build different models which use or not use Chunk information. Comparative experimental results show that the shallow syntactic information chunks make a positive and effective contribution to Chinese prosodic phrase recognition, and the performance of the AdaBoostSVM model is better.