期末報告
鐵路路線入侵事故之分析– 以臺鐵為例
計 畫 類 別 ： 個別型計畫
計 畫 編 號 ： MOST 105-2221-E-006-054- 執 行 期 間 ： 105年08月01日至106年07月31日 執 行 單 位 ： 國立成功大學交通管理科學系（所）
計 畫 主 持 人 ： 胡守任
計畫參與人員： 碩士班研究生-兼任助理：黃品芸 碩士班研究生-兼任助理：陳思予 碩士班研究生-兼任助理：蔣佳蓉
報 告 附 件 ： 出席國際學術會議心得報告
中 華 民 國 106 年 10 月 22 日
臺鐵局鐵路路線入侵事故之發生頻次；第二階段則根據前一階段的 研究結果，進一步以全省各縣市別為單位，納入地區性特性，並以 地理加權迴歸模式探討影響各縣市發生鐵路事故案件之因果關係。
研究結果顯示，區域迴歸模型的模式可以考量不同區位的地理特性 與資料異質性的本質，因此模式配適程度較全域模型佳。此外，在 西部幹線的中南部路段，鐵路正線區段長度與鐵路路線穿越事件呈 現顯著的正相關，顯示路段長度曝光量與發生鐵路正線穿越事件頻 次有因果關係。最後，相對於過去類似文獻，本研究另外發現鐵路 沿線的學校數量與平交道數量，與鐵路路線穿越事件的頻次呈現顯 著正相關。本研究初步的結果，可以提供臺鐵局與相關政府部門
，在未來研擬鐵路路線入侵事故相關防範措施之參考。
中 文 關 鍵 詞 ： 風險管理、鐵路路線入侵、卜瓦松迴歸模式、地理加權迴歸模式 英 文 摘 要 ： This study investigates the risk factors of rail line
trespassing accidents using both global and local
regression models for the Taiwan Railways Administration (TRA) system. A two-stage approach is proposed to
investigate the causal relationships between rail line trespassing accidents and a set of covariates using a 10- year (2005-2014) crash dataset provided by the TRA. In the first stage of this study, the Poisson regression model is used to analyze rail trespassing accident frequency data.
In the second stage, local traffic and environmental
characteristics are incorporated into the modeling process.
Geographically weighted regression model is developed to evaluate the effects of local characteristics on the
occurrence of rail line trespassing accidents. According to the empirical study results, the local model has a better goodness of fit and is able to explore the heterogeneity in different investigated areas. Specifically, a significant positive correlation between rail segment distance and trespassing accidents is found primarily in the southwest region of Taiwan. In addition, a significant positive
correlation exists between the number of highway-rail grade crossings and rail trespassing accidents in northern
Taiwan. Finally, a significant positive correlation is found between the number of schools and rail trespassing accidents in the western main rail corridor. Based on the findings in the empirical study, this study concludes with the policy implications and directions for the government offices in preparing a desirable safety improvement plan for rail trespassing incidents.
英 文 關 鍵 詞 ： Risk management, rail line trespassing, Poisson regression model, geographically weighted regression model
科技部補助專題研究計畫成果報告
（□期中進度報告/期末報告）
鐵路路線入侵事故之分析—以臺鐵為例
計畫類別：個別型計畫 □整合型計畫
計畫編號：MOST 105-2221-E-006-054
執行期間：105 年 8 月 1 日至 106 年 7 月 31 日
執行機構及系所：成功大學交通管理科學系
計畫主持人：胡守任
共同主持人：
計畫參與人員：黃品芸、陳思予、蔣佳蓉
本計畫除繳交成果報告外，另含下列出國報告，共壹份：
□執行國際合作與移地研究心得報告
出席國際學術會議心得報告
□出國參訪及考察心得報告
中 華 民 國 106 年 10 月 22 日
237
Identification of Risk Factors for Rail Line Trespassing Accidents Using Global and Local Regression Models
Shou-Ren Hua* and Pin-Yun Huangb
a, bDepartment of Transportation and Communication Management Science, National Cheng Kung University,
No. 1, University Road, Tainan City, 70101, Taiwan
*Corresponding Author: [email protected] ABSTRACT
This study investigates the risk factors of rail line trespassing accidents using both global and local regression models for the Taiwan Railways Administration (TRA) system. A two-stage approach is proposed to investigate the causal relationships between rail line trespassing accidents and a set of covariates using a 10-year (2005- 2014) crash dataset provided by the TRA. In the first stage of this study, the Poisson regression model is used to analyze rail trespassing accident frequency data. In the second stage, local traffic and environmental characteristics are incorporated into the modeling process. Geographically weighted regression model is developed to evaluate the effects of local characteristics on the occurrence of rail line trespassing accidents.
According to the empirical study results, the local model has a better goodness of fit and is able to explore the heterogeneity in different investigated areas. Specifically, a significant positive correlation between rail segment distance and trespassing accidents is found primarily in the southwest region of Taiwan. In addition, a significant positive correlation exists between the number of highway-rail grade crossings and rail trespassing accidents in northern Taiwan. Finally, a significant positive correlation is found between the number of schools and rail trespassing accidents in the western main rail corridor. Based on the findings in the empirical study, this study concludes with the policy implications and directions for the government offices in preparing a desirable safety improvement plan for rail trespassing incidents.
Keywords: Risk management, rail line trespassing, Poisson regression model, geographically weighted regression model
1. INTRODUCTION
According to the Taiwan Railways Administration (TRA, 2015), the annual volume of served passengers is up to 232,826,496 and the average daily volume is 637,881 in 2014. There was an increase of 4.99% compared to that of 2013. Therefore, railway transport plays a vital role for intercity transport services in Taiwan. In terms of traffic accidents for the TRA system, it is relatively few compared to that of the highway system. Nevertheless, in case of a railway accident, it usually causes significant number of casualties and different degrees of personal and societal impacts, even induces some external costs, such as train delay, locomotive engineer’s psychological impact, and injury recovery and facility maintenance costs, etc.
Past studies concerning rail accidents mainly focused on highway-rail grade crossing accidents (Huang, 2005; Huang, 2009; Hu and Ji, 2014). Investigation of rail trespassing incidents is relatively rare. As shown in Figure 1, the numbers of illegal rail trespassing accidents were more than those of the highway-railway crossing accidents from 2005 to 2014 in Taiwan. In addition, Figure 2 shows that in the investigated years (2005-2014), each year the number of fatalities is always higher than that of injuries.
238
Therefore, the illegal rail line trespassing problem is important and more research endeavors should be spent to find the risk factors of rail line trespassing accidents.
Figure 1. Rail Line Trespassing and Level Crossing Accidents (Source: TRA, 2015)
Figure 2. Injury and Fatality of Rail Line Trespassing Accidents (Source: TRA, 2015)
For rail line trespassing accidents in the international rail community, more than 800 trespassing accidents were happened in the USA per year (FRA, 2012). In Taiwan, only 34 trespassing accidents in 2012. See Figure 3 for the comparison of the selected countries (FRA, 2012; MLIT, 2016; RSSB, 2012; TRA, 2015; TSB, 2016). However, the proportion of trespassing accidents in all railway incidents in Taiwan was higher than those of the other compared countries (see Figure 4). According to the TRA’s 2015 statistics, it was found that the fatality rate in trespassing is 130 times higher than that of the highway system. The accident rate in terms of number of accidents per million train-kilometer in railway is about 263 times higher than the number of accidents per million vehicle-kilometer of the highway system. These comparative results indicate that the rail trespassing incidents are associated with high-risk levels, and should be paid special attention for the safety improvements of the traveling public.
37
50
37 39 38
32 35 34
31 19 21
14 12
4
15 6
14 16 16
13
0 10 20 30 40 50 60
2 0 0 5 2 0 0 6 2 0 0 7 2 0 0 8 2 0 0 9 2 0 1 0 2 0 1 1 2 0 1 2 2 0 1 3 2 0 1 4
Rail line trespassing Level crossing
28
39
23
36
28
24 25 27
15
9 11 14 11
3
10 8 10 9
0
4 0
5 10 15 20 25 30 35 40 45
2 0 0 5 2 0 0 6 2 0 0 7 2 0 0 8 2 0 0 9 2 0 1 0 2 0 1 1 2 0 1 2 2 0 1 3 2 0 1 4
Fatality Injury
239
Figure 3. Number of Trespassing Accidents in 2012 among the Compared Countries
Figure 4. Proportion of Trespassing Accidents in 2012 among the Compared countries
2. PRIOR RESEARCH
Illegal trespassing is often considered more convenient, shorter path than overpass or underpass crossing to the other side of the tracks, is accounted for the majority proportion of railway accidents (Lobb, 2006). In order to clarify the casualties between the occurrence of a rail accident, Stanchak and DaSilva (2014) used the data provided by the Volpe National Transportation Systems Center for rail trespassing analysis. The main characteristics causing the illegal trespassing accidents are disregard for grade crossing warning signs, trespasser intoxication, use distracting electronic devices, and right-of-way proximity to stations, bridges, and rail yards. Illegal trespassing cases in America are often occurred in the summer and evening times. Most trespassers are young to middle-aged males. Savage (2007) analyzed illegal trespassing incidents on the United States railroads by regression models (e.g., Negative Binomial, Prais- Winsten AR(1)). The annual casualty count has remained relatively stable in recent decades because growing affluence, which tends to reduce risk-taking behavior, has been balanced by increases in railroad activity and the size of the population. This study found that the risks of injury and death are particularly acute for males in their 20s and 30s. In Finland, Silla and Luoma (2012a) found that both rail suicides and accidents occurred most often at the end of the week but no specific peak for the time of a year
820
34 61 67 74
0 100 200 300 400 500 600 700 800 900
USA Taiwan England Japan Canada
46%
52%
11% 8.08% 7%
0%
10%
20%
30%
40%
50%
60%
USA Taiwan England Japan Canada
240
was found. Suicides occurred most frequently from afternoon to night and rail accidents occurred during the rush hours, and most train–pedestrian fatalities happened in densely populated areas. In conclusion, the effective prevention of railway suicides and accidents calls for a systems approach involving effective measures introduced by authorities responsible for urban planning, railways, education and public health.
In addition to those statistical methods to explore the casualties in railway accidents, Silla and Luoma (2012b) used questionnaires to survey opinions on railway trespassing of people living close to a railway line in Finland. The respondents supported countermeasures such as building an underpass or fencing off the tracks. In addition, education in schools on the dangers of trespassing was suggested. These results allow practitioners and researchers to see the problem from a local perspective and thus develop a better understanding. Some railway agencies identify risks from potential accident precursors. By lowering precursor frequency, the probability of more serious accidents may be reduced (Kyriakidis et al., 2012). To assist in reducing risks, the safety maturity model (SMM) aims to address not only the behavioral and/or attitudinal culture, but also technical, operational and methodological elements and actual achievements in terms of safety outcomes. Statistical analysis indicates a positive correlation between injuries and top events as well as between injuries and precursors.
Wang et al. (2016) used global model (i.e., the logistic regression model) and local model (e.g. geographically weighted logistic regression model) to investigate the non- crossing rail-trespassing crash data provided by Federal Railroad Administration (FRA).
The independent variables include personal attributes of individuals, environmental and location attributes, time of crash and pre-crash behaviors. The results showed that pre- crash behaviors were found to be key factors showing significant associations with the probability of rail-trespassing injury, especially lying or sleeping on or near tracks.
Fundamentally, the basic assumption of spatial stationarity in traditional regression models does not fully hold in the situation explored. Therefore, regional considerations in some specific situations should guide the implementation of treatments and policies.
The following means are commonly used to prevent illegal trespassing strategies in the past, including setting fences and barriers, cooperation between related government offices, environmental design, setting monitoring equipment, education (Havârneanu et al., 2015). Studies found that fencing and landscaping are more significantly utility than prohibitive signs and education (e.g., Lobb et al., 2001; Silla and Luoma, 2011). Harrison (2015) mentioned that more than 500 trespassing fatalities and nearly as many injuries occur each year in the U.S. The related government offices could try setting signage, crossing guards, and clearinghouse, held the workshop to exchange experiences and so on. Harrison (2015) also provided other advices, such as developing rail media guideline and encouraging responsible reporting, discouraging sensational reporting, what make fewer copycat of illegal trespassing.
Recently, many studies conduct before and after studies of illegal trespassing preventing measures; relatively few spatial analysis of railway safety issues in the past studies (e.g. Wang et al., 2016), which may implement the wrong strategy in some regions. Therefore, this research used the GWR to solve the data heterogeneity problem.
The present research also incorporate variables about the unique aspects in Taiwan, such us farmland proportion in each rail segment, number of schools near the railway line, number of highway-rail grade crossings in each rail segment. We are not only applying traditional statistical models for the current research topic, but also using the GWR to investigate the data heterogeneity. Finally, we present the experimental study results by ArcGIS to give a visualized illustration of the spatial distributions of illegal rail line trespassing accidents.
241
3. METHOGOLOGY
Global models are applied based on the accident data of the entire TRA system, the significant variables can only analyze the general characteristics of accident nature of the TRA system. Alternatively, local models (e.g. the GWR) can handle data heterogeneity problems, showing different significant variables and giving various modeling results for different regions. The following briefly describes the methodological aspects used in this study.
3.1. Poisson Regression Model (PRM)
PRM can deal with the accident frequency of the count data in a period of time.
We will first introduce the PRM as follows.
In equation (1), is the number of illegal trespassing accidents occurred in the i- th segment, and is a non-negative integer, which follows the Poisson distribution.
The following is the Probability Density Function (PDF) of :
f (Y | ) = Pr (Y = )= ! ， 0,1,2, … (1)
where,
Pr (Y = )：probability of accidents occurred in the i-th segment;
：number of accidents occurred in the i-th segment;
：average number of accident occurred in the i-th segment.
In PRM, is the expected value, the expected value of a Poisson variate is the same as the variance and it is represented by the following equations.
E[ ]= (2)
E[ ]=Var[ ]= (3)
In order to ensure is positive, it is assumed that is correlated with a set of independent variable by an exponential relationship as shown in equation (4):
= or ln ( ) = (4)
where,
：vector of independent variable coefficients;
：vector of independent variables during a period of observation in each segment, for example, train frequency, type of railway line, etc.
Combine equations (2) and (4), we have equation (5):
E[ ]= = (5)
Combine equations (1) and (5), we have equation (6):
（! ） (6)
242
Parameter β is estimated by the maximum likelihood estimation (MLE) method, and the following is the likelihood function:
∏ （! ） (7)
Let ln = , then
∑ ln ! (8)
Maximize to get the maximum likelihood estimator of , , ... , , and “n”
is the number of parameters.
PRM use to evaluate the goodness of fit, and the value of is between 0 and 1.
There is better goodness of fit when is closer to 1; and vice versa (Washington et al., 2003).
1 (9)
where,
：The log-likelihood function value when the parameters vector is β;
：The log-likelihood function value when the parameters vector is 0.
3.2. Geographically Weighted Regression (GWR) Model
GWR estimates the parameter for each location ( , ) using a weighted least squares method, as shown in the following equation (Fotheringham et al., 1998;
Fotheringham et al., 2000; Fotheringham et al., 2003; Hadayeghi et al., 2010; Chiou et al., 2015).
, ∑ , (10)
The weight matrix is a diagonal matrix, and each element on the primary diagonal line represents a function of the location i. The weight matrix is expressed as W(i), and according to the principle of weighted least squares method, the original parameter at
location i is estimated by … as： .
where,
⋱ diag … (11)
1 …
1 …
⋮ 1
⋮ ⋱ ⋮
…
， ⋮ (12)
243
The weighted function estimated by using the weighted least square method is the spatial analysis function for different conditions at each data point, where i is the regression point for the parameters at the i-th location. When the data point j is closer to i, the impact is enhanced. For example, in the global regression model, if the distance between data point j and regression point i is within a given radius d, the weight function = 1 (i.e., in this model, the data point weights are identical); otherwise, = 0.
However, this will generate a non-continuous spatial weights problem. Based on the above principles, there are several weight function options, such as Gaussian and bi- square, are commonly adopted, which are respectively expressed as follows.
， j = 1,2,…, n, (13)
where is the distance between regression point i and data point j, and b is the bandwidth. When b is fixed and the data point j is distal to regression point i, the weight is closer to 0.
1 , , 1,2, … ,
0, , (14)
For the bi-square function, if the distance between two locations is larger than the preset bandwidth, the weight is 0.
The following describes the method for the bandwidth calculation.
AIC 2 2π , (15) where n is sample size, is standard deviation, is the predicted value matrix to estimate the bandwidth function.
4. DATA
This study aims to investigate the risk factors of illegal rail line trespassing accidents based on the TRA’s accident dataset from 2005 to 2014. The scope covers all the service routes of the TRA system. Recorded to the end of 2014, the operation mileage of the TRA system is 1061.3 kilometers, including a total of 91 kilometers of the branch lines. Along the operation lines, there are only 466 grade crossings with different control means (mainly active control by quadrant-type gate) and the entire railway operational line is divided into 225 segments, with each segment connecting two consecutive stations. Table 4.1 shows the description of the collected variables, in which variables of farmland proportion, No. of schools and No. of highway-rail grade crossings were not investigated in the past rail trespassing related studies.
In Taiwan, there are many schools near the railway lines, because the TRA system is an important commuter transportation mode. Some students, who live near the station, prefer to illegally cross the tracks to go home rather than walking on the overpass or underpass facilities, because they think it is convenient to reach the destinations.
Thereby, this research investigates whether a school near the tracks within 365.7 meters
244
from the railway (Smith, 2006) is one of the influence factors of rail trespassing accidents. In addition, there are some farmlands on both sides of the tracks; farmers need to cross to the other side of the tracks for planting. Therefore, a train might hit farmers when crossing the track or farmland can separate the highway and railway and further to prevent trespassing accidents from happening. Thereby, this study also plans to explore whether a farmland near the tracks is one of the impact factors of rail trespassing accidents, and we use quartile division to calculate the proportion of a farmland. Finally, this research also investigates whether more accidents are occurred because of fewer legal trespassing locations, such as highway-rail grade crossings (or level crossings). Therefore, the number of highway-rail grade crossings within a rail line segment was also considered in this study.
Table 1. Variable Description
Variable Variable description References No._ accidents
(dependent variable-PRM)
Number of trespassing accidents occurred in each segment
Chi, 1997; Huang, 2005; Savage, 2007;
Huang, 2009;
Hadayeghi et al., 2010; Hu and Ji, 2014 Daily trains Daily trains in each segment Huang, 2005; Savage,
2007; Huang, 2009;
Hu and Ji, 2014 Distance of
segment
Distance between two stations Huang, 2009 farmland_
proportion The proportion of farmland in each segment by quartile division (0%, 25%, 50%, 75%, 100%)
N/A
No._ schools Number of schools within 365.7 meters in each segment
Smith, 2006
No._ highway- rail grade crossings
Number of highway-rail grade crossings
in each segment N/A
Figure 5 shows that most illegal trespassing accidents occurred in the western and northern main rail corridors. This phenomenon is exhibited at the regions with high population density and high frequency of trains in the western railway corridor, but fewer illegal rail line trespassing accidents in the eastern main rail corridor, which is because of the mountainous landscape.
245
Figure 5. No. of Rail Line Trespassing Accidents in Taiwan
5. RESULTS AND DISCUSSION
In this section, empirical study results of applying both the global and local models to the rail trespassing accident data are provided. Additionally, policy implications on risk management of rail line trespassing accidents are discussed.
5.1. Global Model
According to model estimation results shown in Table 2, the significant variables for the accident frequency of rail trespassing accidents are: 1) Distance of segment, 2) Daily trains, 3) No. of schools, and 4) No. of highway-rail grade crossings. The first two significant variables are traffic exposure related, meaning that the longer of a segment and/or the more number of daily trains, the higher probability of a rail line trespassing accident. Besides, the more the number of schools and/or grade crossings at a specific rail line segment, the more the rail line trespassing accidents. Based on above results, education and grade separation are important strategies in the segments with a large number of schools and/or highway-rail grade crossings.
246
Table 2. Estimation Results of the PRM
Variables β p-value
Constant -0.775 0.003
Distance of segment 0.066 0.034**
Daily trains 0.004 0.000***
Farmland_proportion -0.088 0.693
No._schools 0.099 0.001***
No._grade crossings 0.161 0.000***
Summary Statistics Sample size
Log likelihood function LL(β) Restricted log likelihood LL(0) McFadden Pseudo R-squared
225 -322.29 -424.34 0.241 Note：“**” indicates p-value < 0.05
“***” indicates p-value < 0.01 5.2. Local Model
This study further applies the GWR model to the ten-year crash dataset, and obtains the following results.
i. As shown in Figure 6, significant positive correlations between segment distance and trespassing accidents are primarily in the southwest of Taiwan.
Therefore, the longer segments could increase the number of trespassing accidents. TRA could set the priority to set fencing in the southwest of Taiwan to prevent trespassing incidents and/or accidents.
ii. As revealed in Figure 7, significant negative correlations between the farmland proportion and trespassing accidents are primarily in northern Taiwan. Therefore, the proportion of agriculture could be increased to decrease trespassing accidents. Based on the empirical evidence in the spatial model, farmland could separate the railway and roadway systems, and further reduce the number of accidents. Moreover, a special case in northeastern Taiwan, the farmland can be grown is narrow. Therefore, farmers would make agricultural land use extremely, which is quite close to the railway’s right of way. With the poor sight distance of the railway line, trespassers are even more likely to be killed than other areas, given a crash.
iii. As can be seen in Figure 8, significant positive correlations between the frequency of daily trains and trespassing accidents are found in most segments, primarily in the eastern coast rail corridor and western main rail corridor. Trespassing accidents tend to increase as traffic exposure in terms of the number of daily trains increases. Therefore, the higher frequency of daily trains could increase trespassing incidents and/or accidents.
iv. As shown in Figure 9, limited by the mountainous terrain, the t-value in
247
southern Taiwan are relatively low, indicating limited explanatory power for trespassing data. However, the higher the latitude is, the higher the t- value is. There are significant positive correlations between the number of highway-rail grade crossings and trespassing accidents in northern Taiwan.
Moreover, the more the highway-rail grade crossings are, the higher probability of a trespassing accident is observed. However, trespassers are not necessarily crossing the railway at a highway-rail grade crossing, but illegally trespass the tracks because of convenience. Therefore, making grade separation for the segments with a large number of highway-rail grade crossings is a priority policy direction to reduce traffic accidents.
v. As indicated in Figure 10, significant positive correlations between No. of schools and trespassing accidents are primarily in the western main rail corridor. This may partly because that the transportation network in the western region is more complete than that in the eastern region. Many schools near the tracks, due to the TRA system is an important commuter transport mode, resulting in students illegally trespassing the railway to go home or elsewhere. Therefore, education for the long-term plan is a priority countermeasure to reduce the probability of rail trespassing accidents and/or incidents.
vi. Finally, as shown in Figure 11, the goodness of fit in northern and southern is bettern than that of the other regions. It indicates that the applied spatially disaggregated regression model (i.e. GWR model) provides a good remedy for global models and uncovers local spatial variation since it allows the regression parameters to vary across spatial locations.
Figure 6. Distance of the Segment
248
Figure 7. Farmland Proportion
Figure 8. Daily Trains
249
Figure 9. No. of Highway-rail Grade Crossings
Figure 10 No. of Schools near a Railway Line
250 Figure 11. Local
6. CONCLUSIONS
This study investigates the risk factors of rail line trespassing accidents in Taiwan.
Using a ten-year rail line trespassing accident dataset, the empirical study reveals a few crucial findings. The following summaizes the conclusions and recommendations.
i. The GWR model is an effective approach for railway safety analysis, which has a good explanatory power to explain the illegal trespassing data and makes people more comprehensively understand the relationship between trespassing accident frequeicy and regional characteristics across different regions.
ii. A few new variables were found to be significantly associated with trespassing accidents, such as the number of schools, farmland proportion, and number of highway-rail grade crossings, which are rarely reported in the past studies.
iii. Education and outreach programs for the students could be a desirable policy direction to reduce fatalities and injuries due to illegal trespassing accidents.
iv. The related government offices are suggested to make evaluation of
251
farmland used in the northeastern area of Taiwan, because the farmland is quite close to the railway’s right of way and is associated with poor sight distance, which would cause a rail line trespassing accident.
v. Finally, the TRA is recommended to make grade separation first in the segments with a large number of highway-rail grade crossings to prevent train and vehicle crashes caused by illegal trespassing behaviors.
REFERENCES
Chiou, Y. C., Jou, R. C., & Yang, C. H. (2015). Factors affecting public transportation usage rate: Geographically weighted regression. Transportation Research Part A:
Policy and Practice, 78, 161-177.
Federal Railway Administration, Office of Safety Analysis (2012). Retrieved from http://safetydata.fra.dot.gov/OfficeofSafety/publicsite/query/AccidentByRegionS tateCounty.aspx.
Fotheringham, S., Charlton, M., & Brunsdon, C. (1998). Geographically weighted regression: A natural evolution of the expansion method for spatial data analysis.
Environment and planning A, 30(11), 1905-1927.
Fotheringham, A. S., Brunsdon, C., & Charlton, M. (2000). Quantitative geography:
perspectives on spatial data analysis. Sage.
Fotheringham, A. S., Brunsdon, C., & Charlton, M. (2003). Geographically weighted regression: The analysis of spatially varying relationships. John Wiley & Sons.
Hadayeghi, A., Shalaby, A. S., & Persaud, B. N. (2010). Development of planning level transportation safety tools using geographically weighted Poisson regression.
Accident Analysis and Prevention, 42(2), 676-688.
Havârneanu, G. M., Burkhardt, J. M., & Paran, F. (2015). A systematic review of the literature on safety measures to prevent railway suicides and trespassing accidents.
Accident Analysis and Prevention, 81, 30-50.
Harrison, J. A. (2015). Research results. The 2015 ROW fatality & trespass prevention workshop. 15-42.
Huang. W. S. (2005). Exploring the factors affecting the level crossing accidents of TRA.
(Master’s thesis). National Chiao Tung University, Taiwan, in Mandarin Chinese.
Huang, C. F. (2009). The analysis of dangerous degrees for railway level crossing accidents. (Master’s thesis). National Central University, Taiwan, in Mandarin Chinese.
Hu, S. R., & Ji, J. L. (2014). Investigating risk factors of traffic accidents at highway- railroad grade crossings using counted data statistical modeling: A case study of the TRA system. Transportation Planning Journal, 43(2), 115-142, in Mandarin Chinese.
Kyriakidis, M., Hirsch, R., & Majumdar, A. (2012). Metro railway safety: An analysis of accident precursors. Safety Science, 50(7), 1535-1548.
Lobb, B., Harre, N., & Suddendorf, T. (2001). An evaluation of a suburban railway pedestrian crossing safety programme. Accident Analysis and Prevention, 33(2), 157-165.
Lobb, B. (2006). Trespassing on the tracks: A review of railway pedestrian safety research. Journal of Safety Research, 37(4), 359-365.
Ministry of Land, Infrastructure, Transport and Tourism (2016). Retrieved from http://www.mlit.go.jp/toukeijouhou/chojou/stat-e.htm.
Savage, I. (2007). Trespassing on the railroad. Research in Transportation Economics, 20, 199-224.
Silla, A., & Luoma, J. (2011). Effect of three countermeasures against the illegal
252
crossing of railway tracks. Accident Analysis and Prevention, 43(3), 1089-1094.
Silla, A., & Luoma, J. (2012a). Main characteristics of train–pedestrian fatalities on Finnish railroads. Accident Analysis and Prevention, 45, 61-66.
Silla, A., & Luoma, J. (2012b). Opinions on railway trespassing of people living close to a railway line. Safety Science, 50(1), 62-67.
Smith, A. (2006). Level crossing safety performance report. Rail Safety and Standards Board, London.
Rail Safety and Standards Board. (2012). Annual safety performance report. RSSB, London.
Stanchak, K., & daSilva, M. (2014). Trespass event risk factors (No. DOT/FRA/ORD- 14/32). Federal Railroad Administration. Retrieved from http://www.fra.dot.gov/eLib/details/L16047#p1_z50_gD_lRE_lRO_lRT_ktrespa ssing.
Taiwan Railways Administration (2015). Retrieved from http://www.railway.gov.tw/tw/index.html
Transportation Safety Board of Canada (2016). Retrieved from http://www.tsb.gc.ca/eng/rail/index.asp
Washington, S. P., Karlaftis, M. G., & Mannering, F. (2003). Statistical and econometric methods for transportation data analysis. CRC Press.
Wang, X., Liu, J., Khattak, A. J., & Clarke, D. (2016). Non-crossing rail-trespassing crashes in the past decade: A spatial approach to analyzing injury severity. Safety Science, 82, 44-55.
ACKNOWLEDGMENTS
The authors would like to thank TRA, which provides the historical crash data of trespassing accidents and inventory data of the TRA system in the empirical study. This study was partially supported by grant MOST-105-2221-E-006-054 from Ministry of Science and Technology, Taiwan.
ABOUT THE AUTHORS
Shou-Ren Hu received the B.S. degree in Transportation and Communication Management Science from National Cheng Kung University, Taiwan, in 1989; M.S.
degree in Civil Engineering from National Taiwan University, Taiwan, in 1991, and the Ph. D. degree in Civil Engineering from Purdue University, U.S.A., in 1996. He is currently a Professor of Department of Transportation and Communication Management Science, National Cheng Kung University. His research interests include intelligent transportation systems (ITS), transportation network modeling, highway safety, and risk analysis at railway line and level crossings. Dr. Hu is a registered Traffic Engineer in Taiwan.
Pin-Yun Huang received her B.S. degree in Transportation Engineering and Management from Feng Chia University, Taiwan, in 2014; master degree in Transportation and Communication Management Science from National Cheng Kung University, Taiwan, in 2016. Her research interests include risk analysis at railway line and level crossings, transportation planning and operations. She is currently working at Evergreen Marine Corporation.
1
日期：106 年 7 月 21 日
一、 參加會議經過
(一) 會議時間西元二0 一七年七月四日至六日
(二) 會議地點
日本廣島市/ 廣島國際會議場 (International Conference Center Hiroshima) (三) 參與國家與人數
約十五個國家、兩百餘名代表，發表二百四十六篇文章。
(四) 重要活動日程 詳細活動內容：
第一天（7/4）：啟程。
開幕式（Opening Ceremony）
Host: Ushio Sumita, Keio University Welcome Remarks
開幕演講（Keynote Speech）
Keynote Speaker: Fernando A. F. Ferreira, ISCTE University Institute of Lisbon Topic: Multiple Criteria Decision Analysis (MCDA): An Evolving Field of Research
第二天（7/5）
參與技術場次研討 Session E1: E-learning
計畫編號 MOST 105-2221-E-006-054
計畫名稱 鐵路路線入侵事故之分析–以臺鐵為例
出國人員
姓名 胡守任 服務機構
及職稱 成功大學交通管理科學系/ 教授
會議時間 106 年 7 月 4 日至
106 年 7 月 6 日 會議地點 日本廣島市
會議名稱
(中文) 2017 年商業與資訊國際研討會
(英文) The 2017 International Conference on Business and Information
(BAI 2017)
發表題目
(中文) 以全域與區域迴歸模型探討鐵路路線入侵事故之風險因子
(英文) Identification of Risk Factors for Rail Line Trespassing Accidents
Using Global and Local Regression Models
2
Chair: Dr. Jay Rajasekera, International University of Japan, Japan
Presenter: Tien-Chi Huang, National Taichung University of Science and Technology, Taiwan Presentation title: A Highway Driving Safety VR Training Framework Based on Scaffolding Theory
第三天（7/6）
技術場次主持與報告（Technical Session Presiding and Presentation）
Session A6: Multidisciplinary Articles Time: 09:15 – 10:30 A.M.
Venue: Meeting Room A
Chair: Dr. Shou-Ren Hu, National Cheng Kung University, Taiwan
Presentation title: Identification of Risk Factors for Rail Line Trespassing Accidents Using Global and Local Regression Models
日本廣島市與關西地區公共運輸系統考察。
二、 與會心得
本次參加「2017 年商業與資訊國際研討會」（The 2017 International Conference on Business and Information, BAI 2017）相關學術活動，總共參與以下學術活動：1) 參加專題演講場次、2) 進行技
術場次論文報告與主持工作；以及3) 參觀當地公共運輸系統並蒐集相關資料。藉由本次會議的參
與，彙整相關與會心得說明如后。
1. 2017 年的 BAI 國際研討會，係由國立臺北大學與國際商學策進會（International Business Academics Consortium, iBAC）合辦，大會主席為國立臺北大學企業管理學系方文昌教授。該國 際研討會涵蓋的主題，屬於廣義的管理議題（topics related to management in general），包括：
企業管理與策略、財務金融與管理、經濟與會計、資訊管理、人事管理、組織行為、行銷與流 通、管理研究方法、創新與科技、企業責任等等。今年報名參與的人員來自十五個國家、兩百 名以上的各界專業人員，在三天的會議過程中，進行兩百餘篇現場口頭報告與海報展示，其中
不乏學術水準高、具政策意涵的高水準文章，有機會進一步被推薦到Contemporary Management
Research（CMR）與 International Journal of Business and Information（IJBI）等管理學門相關的 期刊審查與刊登。因此，參與該國際研討會不僅可以汲取管理相關議題的新知，同時也有機會 被推薦至相關領域的期刊審議，藉以提升國內的學術研究水準，並與國際學術平台接軌。
2. 藉由參加本次 BAI 2017 國際研討會的機會，除了報告個人投稿的技術論文之外，同時獲邀擔任 同一場次（A6）的主持人工作；該分組場次主題為跨領域的管理議題（Multidisciplinary topics），
到場報告的除了本人報告鐵路交通安全與管理的課題之外，尚包括：非營利組織的營運策略與 績效評估，以及企業籌資的影響因素等兩篇文章，內容多元且具政策意涵，同時也呼應該場次 的多領域及跨領域的訴求，讓不同專業領域的學術研究人員，可以交換彼此的研究心得，同時 吸收不同領域的專業知識。
3. 本次 BAI 2017 的主辦城市所在地日本廣島市，為第二次世界大戰期間第一個原彈爆發地，戰後 日本政府重新規劃該城市，除了進行完整的都市計劃之外，在交通運輸系統的安排上，也別具 用心；尤其在都市道路交通系統的規劃上，地面電車或輕軌系統發達，廣島市素有「電車博物 館」的美稱，主要是各種新舊不同系統的地面電車交陳，不僅有日本其他都市贈與該市的地面
電車（例如：京都電車系統），尚有外觀新穎、功能完備的國外進口系統（例如：德國西門子
3
（Hiroshima City Rental Cycle Peacecle），整個都市地區的公共運輸系統發達，綠運具的使用比 率也高，有效改善都市交通擁擠與空氣汙染等問題。
4. 藉由參加本次國際會議的機會，除了上述有關廣島市的公共運輸系統的考察之外，另外也參訪 日本近畿與關西地區的幾個重要城市，包括：岡山、倉敷、高松、神戶、姬路、京都、大阪等 府與縣市，其中有一個共同的特色是，多數縣市開始設立公共自行車系統或自行車共享系統，
而且以樁柱式的系統為主，在日本多數城市的自行車使用相當普遍的情況之下，短期的觀光客 需求或臨時的通勤使用，自行車共享系統仍有其需求。以廣島市為例，該系統分為單日臨時計 畫（One Day Pass Plan）與會員制計畫（Membership Registration Plan），單日的部分，一天的費
用為日幣1,080 圓，可以用現金或信用卡支付。至於會員制的部分，則分單日、單程、包月制
等三種會員資格，收費方面分別為一天日幣1,080 圓、每三十分鐘日幣 108 圓、前三十分鐘免
費，接下來每三十分鐘日幣108 圓；付費方式則只能用信用卡支付。在使用時間方面，每天開
放租借時間為7:00 AM~11:00 PM，同一天必須於 11:59 PM 前歸還。相較於國內的公共自行車
系統的便利性（例如：雙北的U-bike、高雄市的 C-bike），日本的樁柱式公共自行車系統的使
用彈性較為受限，收費也高，但仍有一定的使用族群，扮演公共運輸系統第一哩與最後一哩的 重要接駁運輸功能。以目前國內正在為有、無樁柱式的公共自行車系統的停車秩序與公平正義
等議題爭論不休之際（即o-bike 的議題），日本各大城市鄰近地鐵或火車站附近的路邊停車、
不分車種皆一律收費的情形（包括自行車，通常一天收費日幣100 圓），無樁柱式的公共行車
的停車仍須付費。基於停車秩序與停車付費等考量，目前日本仍以樁柱式的公共自行車系統之 推廣為主要發展方向。
5. 日本的鐵路運輸系統不僅發展的時間很早，且公、私鐵不同鐵路運輸系統相輔相成，提供民眾
高品質的公共運輸服務。以大阪府為例，大阪市區不僅有JR West 區域鐵路與環狀線分別服務
大阪市的城際交通與城內的旅次，都會區內、外尚有大阪市交通局經營的地鐵系統，以及阪急、
近鐵、南海等私鐵系統，加上公車系統，提供大阪地區綿密且完整的公共運輸系統，成為民眾 每天通勤或探親訪友等旅次目的所倚賴的重要交通工具。支撐上述各項公、民營鐵路系統得以 永續經營且提供高品質的運輸服務的關鍵在於本業的票箱收入，以及場站開發與附屬設施等業 外收入，前者主要靠旅客運量為基礎，後者則需要有高度的商業眼光與經營模式，這兩項重要 收入來源缺一不可。以目前國內正在討論的前瞻基礎建設計畫中、占最大經費比例的軌道運輸 建設為例，各縣市在積極爭取中央的經費補助與青睞之際，是否已針對上述兩項議題，進行客 觀且具科學化的質、量分析，同時考量地區特性與未來發展潛能及限制，誠實的面對可能遭遇 的運量與本業收入不足的問題，據以謀求業外收入挹注之道，才是負責任的做法。他山之石可 以攻錯，儘管日本鐵路運輸系統發展成功、可資借鏡，但仍應酑衡我國都市發展的特性，以及 公共運輸扮演社會公益或營利等角度，適當配置政府有限的預算，以投入最關鍵的軌道運輸基 礎建設對象與項目，才能收事半功倍之效。
三、 發表論文全文或摘要
軌道運輸具有專用路權、運量大等特性，因此軌道運輸系統近年來受到各國的廣泛重視與推 廣。根據交通部臺灣鐵路管理局(以下簡稱：臺鐵局)民國 103 年的年運量人數統計，臺鐵局年運量 可高達2 億 3282 萬 6496 人次，且日平均運量達 63 萬 7881 人次，較民國 102 年增加了 4.99%，說 明了臺鐵局在城際運輸中扮演相當重要之角色。相較於公路系統，臺鐵局的事故發生頻率相對較 低；然而，倘若發生事故，往往造成重大生命財產損害，甚至造成許多無形的成本，例如：列車
4
incidents)。目前臺鐵局總營業里程為 1061.3 公里(含支線 91 公里)，包括 466 座不同管制方式的平 交道，而在全長一千餘公里的營業里程範圍內，針對入侵鐵路正線所造成的事故風險與衝擊程度，
更不容小覷。
針對上述研究課題，本研究針對臺鐵局近十年(2005-2014)鐵路路線入侵事故統計資料進行兩 階段的探討，第一階段以對應的全域統計模式探討臺鐵局鐵路路線入侵事故之發生頻次，包括：
卜瓦松迴歸模式、負二項迴歸模式、TOBIT 迴歸模式；第二階段則根據前一階段的研究結果，進 一步以全省各縣市別為單位，納入地區性特性，並以地理加權迴歸模式探討影響各縣市發生鐵路 事故案件之因果關係。研究結果顯示，區域迴歸模型的模式可以考量不同區位的地理特性與資料 異質性的本質，因此模式配適程度較全域模型佳。此外，相對於過去類似文獻，本研究另外發現 鐵路沿線的學校數量與平交道數量，與鐵路路線穿越事件的頻次呈現顯著正相關，本研究初步的 結果，可以提供臺鐵局與相關政府部門，在未來研擬鐵路路線入侵事故相關防範措施之參考。
四、 建議
根據上述有關參與本次「2017 年商業與資訊國際研討會」（The 2017 International Conference on Business and Information, BAI 2017）的相關學術活動與日本關西與近畿地區多個城市的公共運輸系 統考察，初步提出以下的建議事項。
1. 鼓勵教師與研究生參與不同領域的國際研討會議：
近年來由於經濟的快速發展與社會環境的大幅變遷，都會區與城際交通擁擠問題已成為世界 各國所共同面臨的問題，以傳統增加運輸供給的手段來改善交通問題，因為運輸設施用地取得不 易與財源有限，日益不可行；由於資訊與通訊等技術的快速發展，以及電信市場的自由化與成本 降低，此時以先進科技應用於改善交通運輸問題，變得可行且深具發展潛力。BAI 國際研討會即 在於提供世界各國以管理與科技的手段，改善實務管理的相關問題之經驗交換與技術交流的場 合。建議國內交通運輸領域勵相關人員可以考慮投稿至不同管理領域的國際研討會，與跨領域及 不同專業的人員，進行國際交流與經驗分享。
2. 提升研究生的專業能力與國際競爭力：
藉由參與本次技術場次的機會，本人注意到日本與韓國相關大學的研究水準之提升，不僅在 英文簡報技巧與口說皆有水準以上的表現，同時在即席應答上更有明顯的進步；而反觀我國大學 的研究生在國際場合上的表現，即相形見拙，主要原因在於平時缺乏口說英文的練習場合，同時 也鮮少有機會在國際學術舞台上表現，因此在簡報技巧、乃至於即席應答上，皆有改善的空間。
建議除了上述提供碩、博研究生參與國際會議的誘因與補助之外，平時在課程設計與研究過程中，
應加強英文的讀寫聽說等能力，以期在國際舞台上與各國同儕競爭與合作。
3. 借助日本發展公共運輸系統之經驗：
如前所述，日本的鐵路、公車、渡輪等公共運輸系統不僅發展的早，且因公、私部門的適當 角色扮演、相輔相成，都會區的公共運輸使用率相當可觀。分析日本公共運輸系統成功的原因，
主要在於由下而上、市場導向的路網規劃與經營管理，亦即日本在交通運輸系統的規劃方面，向 來著重「情報」的蒐集與運用，一項公共建設或重大交通運輸系統的投資計畫，事前的主、客觀 因素蒐集與分析，從可行性評估到財務與建設計畫的擬定，必須有詳實的背景資料與關鍵情報做 基礎，才能依序推動，以收實效。我國目前正處於軌道運輸系統的成長期，除了環島的臺鐵與西 部運輸走廊的高鐵系統，各縣市正在積極爭取的捷運系統或鐵路高架化、電氣化等重大建設項目，
建議借助日本發展鐵路運輸與其他公共運輸系統的經驗，縝密規劃與確實執行，才能有所成效。
5
1. 「2017 年商業與資訊國際研討會」（The 2017 International Conference on Business and Information, BAI 2017）論文集隨身碟
2. 「2017 年商業與資訊國際研討會」（The 2017 International Conference on Business and Information, BAI 2017）大會手冊
3. 日本廣島市及其他都市與城際交通系統簡介（包括：地面電車、公車、地鐵、JR、公共自行車 等等）
六、 附錄：(與會與參訪照片)
圖 1 B A I 2 0 1 7 會 議 場 所 ( 廣 島 國 際 會 議 場 )
6
圖 2 技 術 場 次 A 6 會 場 ( 廣 島 國 際 會 議 場 / M e e t i n g R o o m A )
圖 3 主 持 技 術 場 次 A 6 證 書 ( i s s u e d b y P r o f . W e n c h a n g F a n g , C h a i r )
7
圖 4 廣 島 市 地 面 電 車 ( 德 國 西 門 子 系 統 )
圖 5 廣 島 市 地 面 電 車 ( 傳 統 電 車 系 統 )
8
圖 6 廣 島 市 自 行 車 共 享 系 統 ( H i r o s h i m a C i t y R e n t a l C y c l e P e a c e c l e )
圖 7 廣 島 渡 輪 系 統 ( 宮 島 線 )
9
圖 8 J R 西 日 本 新 幹 線 ( 山 陽 線 _ 岡 山 站 )
圖 9 岡 山 市 區 域 鐵 路 系 統 ( 福 山 線 _ 岡 山 站 )
10
圖 1 0 岡 山 市 地 面 電 車
圖 1 1 岡 山 市 自 行 車 共 享 系 統
11
圖 1 2 高 松 市 J R 區 域 鐵 路 系 統 ( 高 松 站 )
圖 1 3 高 松 至 直 島 渡 輪 系 統
12
圖 1 4 神 戶 市 J R 區 域 鐵 路 ( 三 ˊ 宮 站 )
圖 1 5 神 戶 市 自 行 車 共 享 系 統 ( k o b e l i n )
13
圖 1 6 J R 姬 路 站
圖 1 7 姬 路 市 自 行 車 共 享 系 統
14
圖 1 8 J R 北 近 畿 觀 光 列 車 ( 丹 後 線 _ 城 琦 溫 泉 站 )
圖 1 9 日 本 三 大 名 景 _ 京 都 府 天 橋 立 纜 車 系 統
15
圖 2 0 日 本 三 大 名 景 _ 京 都 府 天 橋 立 吊 籃 系 統
圖 2 1 J R 京 都 站
16
圖 2 2 京 都 府 機 械 式 公 車 動 態 資 訊 站 牌
圖 2 3 大 阪 市 阪 急 電 鐵 ( 堺 筋 線 _ 堺 筋 本 町 站 )
17
圖 2 4 大 阪 市 地 下 鐵 ( 千 日 前 線 _ 難 波 站 )
圖 2 5 大 阪 市 地 下 鐵 ( 御 堂 筋 線 _ 心 齋 橋 站 )
18
圖 2 6 大 阪 府 單 軌 電 車 系 統 ( 彩 都 線 _ 萬 博 紀 念 公 園 站 )
圖 2 7 大 阪 府 南 海 鐵 路 ( 南 海 本 線 _ 難 波 站 )
19
圖 2 8 關 西 國 際 機 場 無 人 駕 駛 單 軌 電 車 系 統
圖 2 9 大 阪 府 關 西 國 際 機 場
計畫主持人：胡守任 計畫編號：105-2221-E-006-054- 計畫名稱：鐵路路線入侵事故之分析– 以臺鐵為例
成果項目 量化 單位
質化
（說明：各成果項目請附佐證資料或細 項說明，如期刊名稱、年份、卷期、起 訖頁數、證號...等）
國 內
學術性論文
期刊論文 0
研討會論文 0 篇
專書 0 本
專書論文 1 章
成功大學交通管理科學系碩士論文：鐵 路路線入侵事故分析-以臺鐵為例 (2016.7)
技術報告 0 篇
其他 0 篇
智慧財產權 及成果
專利權 發明專利 申請中 0
件
已獲得 0
新型/設計專利 0
商標權 0
營業秘密 0
積體電路電路布局權 0
著作權 0
品種權 0
其他 0
技術移轉 件數 0 件
收入 0 千元
國 外
學術性論文
期刊論文 0
研討會論文 1 篇
The 2017 International Conference on Business and Information (BAI 2017)
專書 0 本
專書論文 0 章
技術報告 0 篇
其他 0 篇
智慧財產權 及成果
專利權 發明專利 申請中 0
件
已獲得 0
新型/設計專利 0
商標權 0
營業秘密 0
積體電路電路布局權 0
其他 0
技術移轉 件數 0 件
收入 0 千元
參 與 計 畫 人 力
本國籍
大專生 0
人次
碩士生 3 黃品芸、陳思予、蔣佳蓉
博士生 0
博士後研究員 0
專任助理 0
非本國籍
大專生 0
碩士生 0
博士生 0
博士後研究員 0
專任助理 0
其他成果
（無法以量化表達之成果如辦理學術活動
、獲得獎項、重要國際合作、研究成果國 際影響力及其他協助產業技術發展之具體 效益事項等，請以文字敘述填列。）
