基于专家知识对青藏铁路GPS测量数据的综合性粗差探测方法研究毕业论文外文翻译.docx

《基于专家知识对青藏铁路GPS测量数据的综合性粗差探测方法研究毕业论文外文翻译.docx》由会员分享，可在线阅读，更多相关《基于专家知识对青藏铁路GPS测量数据的综合性粗差探测方法研究毕业论文外文翻译.docx（36页珍藏版）》请在淘文阁 - 分享文档赚钱的网站上搜索。

1、外文翻译（外文文献）An integrated error-detecting method based on expert knowledge for GPS data points measured in QinghaiTibet RailwayDewang Chen *, Tao Tang, Fang Cao, Baigen CaiA b s t r a c tAs there are huge amounts of Global Positioning System (GPS) data points measured in the QinghaiTibet Railway (QTR)

2、 with a length of 1142 km, it was inevitable that some measuring errors existed due to various situations in measurement. It is very important to develop a method to automatically detect the possible errors in all data points so as to modify them or measure them again to improve the reliability of G

3、PS data. Four error patterns, including redundant measurement, sparse measurement, back-and-forth measurement, and big angle change, were obtained based on expert knowledge. Based on the four error patterns, four algorithms were developed to detect the corresponding possible errors in data points. T

4、o delete the repetitive errors by different algorithms and effectively display the possible errors, an integrated error-detecting method was developed by reasonably assembling the four algorithms. After four performance indices were given to evaluate the performance of the error-detecting method, si

5、x GPS track data sets between seven railway stations in the QTR were used to validate the method. Thirty-eight segments of some sequential points that are possibly wrong were found by the method and fourteen of them were confirmed by measurement experts. The detecting rate of the method was 100% and

6、 the duration time of the detecting process was less than half an hour compared with the 94 h manual workload. The validation results show that the method is effective not only in decreasing workload, but also in ensuring correctness by integrating the domain expert knowledge to make the final decis

7、ion.Keywords:QinghaiTibet Railway;GPS;Error-detecting;Expert knowledge;Error pattern1. IntroductionAs satellite positioning has many advantages (e.g., low cost,real-time, and no cumulative errors (Blomenhofer, 2004), it is often used in car-navigation systems (Skog & Handel, 2009) or in generating e

8、lectronic maps (Zhang, Chen, & Kruger, 2008). Furthermore, satellites are currently used to track the positions of train instead of using radio frequency systems (Santos, Soares, & Redondo, 2005) or track circuits (Oukhellou, Debiolles, Denoeux, & Aknin, 2010). Positioning systems using satellites c

9、an help in reducing the cost of installing and maintaining track-side equipment (George, Juliette, & Marion, 2004). Recently, the European Union has launched many projects (e.g., GADEROS (Urech, Diestro, & Gonzalez, 2002) and RUNE (Albanese, Marradi, Labbiento, & Venturi, 2005) using satellite posit

10、ioning for low-density railways. In USA, an incremental train control system using GPS positioning was employed in a Michigan railway (Baker & Clennan, 2005). The results of these projects have shown that satellite positioning has a better performance-cost ratio for low-density railways. In China, G

11、PS positioning was firstly adopted in the train control system for the QTR in 2006 where no track circuits are present. It greatly reduced track-side equipment and maintenance cost in the worlds highest railway (Li, 2005).Digital track maps (DTM) with high precision are the basis for accurate train

12、positioning. With the help of DTM, positioning error is reduced and positioning reliability is enhanced (Simsky, Wilms, & Franckart, 2004). Apparently, DTM generated from GPS track data points can be easily used in train positioning. To generate the DTM with high reliability, GPS track data points w

13、ith high accuracy and the integrity and strictness in recording are required as prerequisite. The high accuracy of GPS track data points was achieved by the differential GPS (DGPS) technology (Lee & Rizos, 2008) in the QTR. However, the integrity and strictness in recording mainly depended on the pa

14、tience and earnest of the measuring workers. As there are huge amounts of satellite location GPS data points in QTR with a length of 1142 km, it is inevitable that some measuring errors may exist due to various reasons in measurement.The basic method to find measuring errors for engineers is to obse

15、rve data points segment by segment, which is obviously time-consuming and easy to miss some errors. An effective errordetecting method should be developed to find the possible errors automatically and quickly to increase the reliability of the data. Expert knowledge play a key role in dealing with s

16、pecific problems in different domains, such as in analyzing degraded terrain (Genske & Heinrich, 2009), in earthquake resistant design of reinforced concrete buildings (Berrais, 2005), in fraud detection in communication networks (Hilas, 2009), etc. Therefore, it is reasonable to think that some rul

17、es can be concluded by measuring experts from which some error-detecting algorithms can be designed. After these error-detecting algorithms find the possible errors and display them, operators will make a final decision from their experience on what are real errors to increase the work efficiency an

18、d ensure the correctness of the judgment. Therefore, expert knowledge is very important for this issue not only in increasing efficiency, buy also in ensuring correctness.In this paper, some error-detecting algorithms based on expert knowledge will be developed and integrated into an error-detecting

19、 method to find the possible errors in GPS track data points of some railway stations in the QTR. The structure of this paper is as follows. In Section 2, after the brief description of QTR, the six field GPS data sets between seven railways stations in QTR are described.In Section 3, four error dat

20、a patterns are concluded from the expert knowledge and four corresponding error-detecting algorithms are developed based on the four patterns respectively. In Section 4, after an error-detecting method is developed by assembling the four error-detecting algorithms to effectively display the errors a

21、nd delete the possible repetitious errors detected by different algorithms, four performance indices are given to evaluate the performance of the method. In Section 5, some possible errors in data points are found by the error-detecting method from the field GPS track data sets after setting up the

22、threshold values by expert knowledge. The computational results for the six data sets are listed and summarized after being compared with the manual judgement. In Section 6, some conclusions and its field application are concluded.2. Descriptions on QinghaiTibet Railway and field data2.1. A brief de

23、scription for the QTRThe QTR mentioned in this paper is a high-altitude railway that connects Golmud in Qinghai Province to Lhasa in Tibet Autonomous Region, which was inaugurated on July 1, 2006 by Jintao Hu, the President of China. Fig. 1 illustrates a train which was running on the QTR after the

24、inauguration, and Fig. 2 shows the major railway stations on the QTR and the main mountains alongside the QTR, which are available from the following website: Fig. 1. A train on the QTR after inauguration.Fig. 2. Main stations and main mountains along the QTR.The QTR includes the Tanggula railway st

25、ation at an elevation of 5072 m, the worlds highest railway station, and the Fenghuoshan tunnel at an elevation of 4905 m, the highest rail tunnel in the world. Approximately 500,000 GPS data points with one centimeter precision were obtained from DGPS technology along the QTR with a length of 1142

26、km. The distance of two adjacent points is between 1.5 m and 3 m and the average distance is about 2.5 m.2.2. Field data descriptionSix data sets between seven railway stations were chosen to be studied in this paper. Because there is strict security requirement for the GPS track data of the QTR, a

27、specialized software was utilized to transform the latitude, longitude and height GPS data into xyz coordinate in Descartes coordinate system and the beginning data in first railway station was set as the original point of the coordinate system.It is a common sense that the gradient in railway is ve

28、ry small which means the change in height (z coordinates) is much smaller compared with the changes in x _ y plane (Cheng, Davydova, Howlett,& Pudney, 1999). Through pre-processing data analysis from the surveyed data sets, it was found that the distance between two adjacent points is between 1.5 m

29、and 3 m. However, after projecting all data points into z-axis, the average distance between two adjacent points in the z coordinates is only 0.0011 m and the maximal value is only 0.023 m. Therefore, the change in z-axis can be omitted compared with the change in xy plane. To display the data point

30、s clearly, only the GPS data projected in the x _ y plane are chosen to be shown in this paper. The projected data sets in x _ y plane of all surveyed GPS data points in six sections among seven railway stations in the QTR are illustrated in Figs. 35. The length of each section is between 20 km and

31、30 km. Six data sets of six sections are defined as =, (i = 1,2,. . . ,6) which respectively include 9279,9768, 10257, 11233, 9475, and 8595 data points. These data points were recorded in measuring order from the beginning point to the ending point of all data points and the DGPS technique was used

32、 to achieve one-centimeter precision for the GPS data points.Fig. 3. GPS data points in xy plane of Sections 1 and 2.Fig. 4. GPS data points in xy plane of Sections 3 and 4.Fig. 5. GPS data points in xy plane of Sections 5 and 6.3. Four error patterns and error-detecting algorithms3.1. Four error pa

33、tterns based on expert knowledgeAfter consulting some GPS measuring experts and the field measuring engineers, we found that four error patterns often occurred in the measurement and recording. Although the measuring density in straight rail is lower than that in the curved rail for the GPS data poi

34、nts in the QTR based on the real measurement experience,the distance in two adjacent data points is within a certain range. Therefore, the two error patterns are obtained in the following.(1) Error pattern 1 includes three situations, which are redundant measurement, dense measurement and repetitive

35、 recording, as the point illustrated in Fig. 6.(2) Error pattern 2 is referred to as sparse measurement, missing points and forgotten recording, like the missing point between point and illustrated in Fig. 7.(3) Error pattern 3 is defined as back-and-forth measurement or circular measurement, as poi

36、nt and illustrated in Fig. 8.In measurement, there existed a phenomenon of the socalled back-and-forth measurement caused by false recording or the worry of missing points. This kind of error will result in many redundant points in a short railway segment and make DTM unreasonable as the later segme

37、nt is in front of the former segment.(4) Error pattern 4 is defined as the big angle change caused by the data point obviously departing from the railway track, as the point in Fig. 9.As a rail track is either a straight line or a curve with big curvature radii, the angle change between two adjacent

38、 lines is very small. If a data point is found to be obviously departing from the track, then the angle change will be big. It is necessary to point out that the first two error patterns are not severely wrong. Error pattern 1 only finds some redundant data points which can be deleted to save the st

39、orage space. Error pattern 2 has no bad effect on the straight track, but can remind measuring engineers of adding some measuring data points if necessary.Fig. 6. Error pattern 1.Fig. 7. Error pattern 2.Fig. 8. Error pattern 3.Fig. 9. Error pattern 4.3.2. Four error-detecting algorithmsTo make descr

40、iptions for error-detecting algorithms more accurate and easy to be understood, we define some notations as follows:(1) , the ith point; ，and are the coordinates of pi, thus, = (，);(2) N, the number of all points in a data set;(3) , the distance between and ; , the distance between and ;(4) , the li

41、ne connecting and; , the inclination between and ;(5)，, the inner product of and .Aimed at the four error patterns, four algorithms were developed to detect them respectively.(1) As to the error pattern 1, can be utilized to find the possible errors. If is smaller than a threshold value,as shown in

42、Eq. (1), then it is thought that a false or redundantdata point appeared. (2) As to the error pattern 2, it can also be found by. If is bigger than a threshold value , as shown in Eq. (2), then it is thought that a false data point or a missing data point appears. (3) As to the error pattern 3, it c

43、an be found by the as shown in Eq. (3) which should be multiplied by 2 in the normal situation. If the back-and-forth measurement happened, then will be very small, even smaller than the . As to the threshold value , it is approximately equal to multiplied by 2.(4) As to the error pattern 4, it can

44、be found by the angle change in degree between two adjacent lines.where is the inverse cosine of the elements of x in radians. In the normal situation, is very small as the railway track is very straight. Therefore, the absolute value of should be less than a threshold value . Otherwise, a false dat

45、a point will probably appear as shown in Eq. (5).The threshold vales, in Eq. (1), in Eq. (2), and in Eq. (3) and in Eq. (5), can be decided by experienced experts in measurement. Thus, expert knowledge has great influence on error - detecting algorithms. Besides, the scatter plots of , and could be

46、used to double-check the threshold values given by experts.4. Integrated error-detecting method and performance indices4.1. Integrated error-detecting method: the integration of the fourerror-detecting algorithms If a possible error is found by any algorithm at , then we collect the five points near

47、by pi as a suspicious point set ， (if such nearby points exist) for the convenience of observation. Some errors can appear in different error patterns, for example, error pattern 3 will result in error pattern 4. To decrease the repetitious counting for the possible errors in GPS data points, it is

48、necessary to combine all possible errors found by all 4 algorithms together and delete the repetitious points. After the combining process, the possible errors are divided into some segments each of which contains several continuous data points. Through the four stages of extending, combining, delet

49、ing, and displaying, the four algorithms are integrated into the so-called integrated error-detecting method, which can effectively find and display the possible errors hidden in the huge amounts of the measured GPS data points in the QTR.Then, the possible false points found by the integrated errordetecting method will be displayed for measurement experts to make the f