HIGH-SPEED RAILWAY BASED ON GENETIC ALGORITHM

FOR PREDICTION OF TRAVEL CHOICE

Long Chen-xu, Li Jing and Gao Yue

School of Economics and Management, Information Management and Information System, Beijing Jiaotong University

Yuen Estate, No. 3, 100044, Haidian District, Beijing, China

Keywords: Genetic algorithm, High-speed railway, Forecast analysis, Modal split.

Abstract: Genetic algorithm is a new optimizing searching method based on biology evolutionary theory. Just as

evolution deals in populations of individuals, genetic algorithms mimic nature by evolving huge churning

populations of code, all processing and mutating at once. With the frequency of passenger travel speeding

up and passenger's demand to the quality of life higher and higher, passengers have higher and higher

demands to the travel. Especially in environment, comfort and quick aspect, different passenger focuses on

different aspects, therefore, before the research, we must classify the passenger. This paper applies British

Sheffield university GA toolbox, with the application of matlab, finally makes a forecast analysis. Because

the forecast analysis is based on the questionnaire during the Spring Festival, so the emphasize on the travel

choice made by the passengers in this certain circumstances is necessary. In addition, the forecast analysis

will be more or less different with other forecast analysis normally.

1 INTRODUCTION

Genetic algorithm (GA) is a random search method.

It decreases the effect of original values greatly

through crossover and mutation operations, and it

can easily find out the global optimal results.

WU Qun-qi and XU Xing (2007), thought that

passengers choose transportation mainly rely on

time, economy and feeling these three factors. The

constituent of the value of trip time of the specific

travel subject has something to do with the travel

interests: full economic relevance, totally not related

to the economy, some economic-related. According

to the value of travel time and the correlation of

travel interests, it suggests the mechanism of

passengers' travel choice. Simple genetic algorithm

uses binary encoding, this approach is simple, easy

to implement crossover and mutation operations, in

line with the principle of the minimum character set

encoding. As we all know, the simple genetic

algorithm is poor in the local search capabilities and

has a slower speed in optimizing the global,

therefore, it is necessary to establish the crossover

operator which can simultaneously search both in

feasible and infeasible solution space, the mutation

operator which has a capability of fast search in the

earlier stage and the capability of maintaining the

optimal solution in the later stage, and the selection

operator which has the capability of maintaining the

"elite". MA Yong-jie, MA Yi-de, JIANG Zhao-

yuan, SUN Qi-guo (2009) propose to take use of the

solutions which has been searched for to avoid the

"throwback" of the offspring and the degradation is

obligatory. GONG Gu, ZHAO Xiang-jun, HAO

Guo-sheng, CHEN Long-gao (2009) propose to

divide the search space, we can inherit the excellent

allele to the next generation. While using the taboo

domain and the active domain can quickly improve

the algorithm for performance.

The core problem of this paper is to use the

collected initial data from the questionnaire based on

high-speed railway passenger travel, the influencing

factors theory related to high-speed railway

passengers travel choice, and the knowledge about

genetic algorithm to analyse the factors influencing

passengers choosing transportation and make

predictions. The objective function is maximizing

the passengers' travel utility. However, different

types of passengers have different preferences and

different emphasis on economy, time and feeling.

Therefore, each type of passenger takes different

aspects into consideration and makes travel choice.

Obviously, it is very important to establish a

Chen-xu L., Jing L. and Yue G..

HIGH-SPEED RAILWAY BASED ON GENETIC ALGORITHM FOR PREDICTION OF TRAVEL CHOICE.

DOI: 10.5220/0003433400260031

In Proceedings of the 13th International Conference on Enterprise Information Systems (ICEIS-2011), pages 26-31

ISBN: 978-989-8425-54-6

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

simulation model.

This paper used Matlab to solve the genetic

algorithm problem, with binary encoding, and made

predictions for passengers' travel choice.

2 CONCEPTUAL FRAMEWORK

Passengers choose transportation mainly rely on

time, economy and feeling these three factors.

Dividing these three dimensions, we classify the

factors influencing passengers making travel

choices.

Passengers make travel choices will be

influenced by some objective, potential factors.

Through analysing the passengers' characteristics

and different factors' effects on passengers, we got a

flow chart of passenger travel.

Figure 1: Flowchart passenger travel mode choice.

Passenger travel is based on different purposes,

such as going home, for business, for tourism, so

passengers based on different trip purposes may

make different choices in travel mode. In addition,

the economic capacity of travelers and who will bear

the cost of travel will have a great impact on travel

mode choice.

2.1 The Division of Passenger Types

2.1.1 According to the Travel Purpose

According to passengers' respective trip purpose, all

passenger traffic in the channel can be divided into:

①

business;

②

tourism;

③

work;

④

school;

⑤

home;

⑥

interchange;

⑦

others.

Among them, passengers for business always

focus on convenience and comfort, but have low

sensitivity of the cost; passengers for tourism often

focus on comfort and fare levels, and have high

sensitivity of the cost; work, school and other

commuter passenger traffic often takes fare for the

primary consideration, and has certain requirements

on punctuality; those for home always have low

requirements for comfort and higher sensitivity of

the costs, and take fare levels for main

consideration.

2.1.2 According to the Cost Mode

According to who will bear the cost, passenger

traffic can be divide into:

①

travel at public expense

②

travel at their own

expense.

Travel at public expense means that passenger

does not need to pay for the travel because of the

social production and the need to work, and travel

costs are included in the cost of social production.

Travel at their own expense is to meet passengers'

own needs and travel costs are included in private

consumption. Because of the existence of these two

different ways of bearing the cost of travel, there

will be some differences when the two parts of

passenger traffic choose the travel mode.

Because the travel time is included in the cost of

production, passenger traffic at public expense will

pay more attention to time and look for convenient,

fast and punctuality while they are choosing the

travel mode. So they have higher selection bias of

civil aviation and high-speed railway. Moreover,

they have higher requirements about the frequency

of the mode of transportation, departure time and

arrival time. And they are less sensitive to the travel

cost. However, passenger traffic at their own

expense will have lower requirements for the quality

of transport and higher sensitivity for the travel costs

because they must pay for themselves.

2.1.3 According to Income

According to monthly income, the overall passenger

traffic is divided into:

①

lower-income travelers: 1000 yuan;

②

low-

income travelers: 1000 yuan to 3,000 yuan;

③

middle-income travelers: 3000 yuan to 5000

Element;

④

high-income travelers: 5000 yuan to

10,000 yuan;

⑤

higher-income travelers: more than

10,000 yuan.

Due to a difference in income level, different

types of passengers have different capacity of

bearing the travel costs and have different sensitivity

of costs. The previous data showed that middle-

income and less income stream of passengers have a

preference for the traditional existing rail or road.

HIGH-SPEED RAILWAY BASED ON GENETIC ALGORITHM FOR PREDICTION OF TRAVEL CHOICE

And for the travel costs are the main considerations,

they have a relatively low requirement for the

transport quality such as comfort, convenience and

punctuality. In addition, they have a relatively high

degree of sensitivity for the fare level, so the

fluctuations in fares will cause great changes in the

passenger traffic distribution.

Passengers who have high and higher income

will take comfort, convenience and punctuality into

consideration because they have high abilities to

pay, and they always select high quality

transportation services of transportation, such as

high-speed railway, civil aviation. What's more,

these passengers are less sensitive to the cost, so a

certain range of fluctuations in travel mode will have

a little influence on them.

3 RESEARCH MODEL

The problem was described as: There were m travel

modes and n batches of passengers(category)

waiting to be distributed.

Before the target allocation, the key

considerations of each batch of the target and each

travel mode's weight on each target has been

evaluated and sorted. J-approved visitors' "travel

value" is w

, i-approved travel mode's weight on j-

approved target is

, each travel mode's "trial"

benefit value on each target is

ijjij pwu *=

Among them,

iju

stands for each batch of the

passenger's size of the degree of the effectiveness of

the "trial". The purpose is to meet the basic

principles of the target allocation and pursuit of the

overall effectiveness of the best, that is seeking

).max(

∑

4 RESEARCH METHODS

This paper used binary encoding and the number of

individuals was 40. In addition, the max number of

generations was 50 and the generation gap was 0.9.

This paper used PN instead of passenger

numbers, and TV instead of travel value.

Based on the numerical analysis of

questionnaires, the standard value of the price

dimension is 3.16, the standard value of the time

dimension is 2.69 and the standard value of the

environmental dimension is 4.47.

Choices for passengers to choose for travel:

①EMUs ②Direct train ③Coach ④Aircraft

⑤MICE.

4.1 According to the Purpose

Passengers for business always focus on

convenience and comfort, but have low sensitivity of

the cost, so the weight for the price dimension is

0,for the time dimension is 0.6 and for the

environment is 0.4, that is, for the purpose of

business, "travel value " is 0.6 * 2.69 +0.4 * 4.47 =

3.402. Passengers for tourism often focus on comfort

and fare levels, and have high sensitivity of the cost,

so the weight for the price dimension is 0.3 and for

the environment is 0.7, that is for the purpose of

tourism, "travel value " is 0.7 * 4.47 +0.3 * 3.16 =

4.077. Work, school and other commuter passenger

traffic often takes fare for the primary consideration,

and has certain requirements on punctuality, so the

"travel value" for the passengers whose purpose is

going to work is 0.6 * 2.69 +0.4 * 3.16 = 2.878,

passengers to school's "travel value " is 0.2 * 2.69

+0.8 * 3.16 = 3.066. Passengers for home have low

requirements for the comfort, often focus on the

price level and have high sensitivity for the costs, so

the "travel value" is 1 * 3.16 = 3.16. Passengers for

transfer have high requirements for time, so the

"travel value" is 1 * 2.69 = 2.69. Other passengers'

weight are similar, that is, its "travel value" is 0.3 *

2.69 +0.3 * 4.47 +0.4 * 3.16 = 3.412.

Table 1: Travel Value - According to the Travel Purpose.

PN 1 2 3 4 5 6 7

TV（

）

3.4

4.0

2.8

3.0

3.1

2.6

3.4

4.2 According to the Cost Mode

Because the travel time is included in the cost of

production, passenger traffic at public expense will

pay more attention to time and look for convenient,

fast and punctuality while they are choosing the

travel mode. So they have higher selection bias of

civil aviation and high-speed railway. Moreover,

they have higher requirements about the frequency

of the mode of transportation, departure time and

arrival time. And they are less sensitive to the travel

cost. So the weight for the time dimension is 0.42

and for the environment dimension is 0.58, that is,

the "travel value" is 2.69 +4.47 * 0.42 * 0.58 =

3.7224. However, passenger traffic at their own

expense will have lower requirements for the quality

of transport and higher sensitivity for the travel costs

ICEIS 2011 - 13th International Conference on Enterprise Information Systems

Table 2: Weights - According to the Travel Purpose.

i-j(

ijp

)

1 2 3 4 5

1 0.44 0.01 0.01 0.53 0.01

2 0.19 0.23 0.31 0.22 0.05

3 0.23 0.36 0.18 0.11 0.12

4 0.15 0.42 0.26 0.12 0.05

5 0.15 0.52 0.19 0.13 0.01

6 0.31 0.41 0.26 0.01 0.01

7 0.25 0.25 0.25 0.25 0.25

Table 3: Travel Value - According to the Cost Mode.

PN 1 2

（

）

3.7224 3.422

Table 4: Weights - According to the Cost Mode.

i-j(

ijp

)

1 2 3 4 5

1 0.28 0.39 0.05 0.27 0.01

2 0.12 0.43 0.11 0.17 0.17

because they must pay for themselves. So the "travel

value" is 0.8 +0.2 * 3.16 * 4.47 = 3.422.

4.3 According to Income

The previous data showed that middle-income

and less

income stream of passengers have a preference for

the traditional existing rail or road. And for the

travel costs are the main considerations, they have a

relatively low requirement for the transport quality

such as comfort, convenience and punctuality. In

addition, they have a relatively high degree of

sensitivity for the fare level, so the fluctuations in

fares will cause great changes in the passenger

traffic distribution.So the "travel value" for the

lower-income is 0.9 * 3.16 +0.09 2.69 +0.01 * 4.47

= 3.1308. And the "travel value" for the low-income

passenger is 0.9 * 3.16 +0.082 * 2.69 +0.02 * 4.47 =

3.1486.

The "travel value" for the middle-income

passenger is 0.5 * 3.16 +0.25 * 2.69 +0.25 * 4.47 =

3.37.

Passengers who have high and higher income

will take comfort, convenience and punctuality into

consideration because they have high abilities to

pay, and they always select high quality

transportation services of transportation, such as

high-speed railway, civil aviation. What's more,

these passengers are less sensitive to the cost, so a

certain range of fluctuations in travel mode will have

a little influence on them. So the "travel value" for

those high-income is 0.21 * 3.16 +0.39 * 2.69 +0.4 *

4.47 = 3.5007, and the "travel value" for the higher-

income is 0.15 * 3.16 +0.36 * 2.69 +0.49 * 4.47 =

3.6327.

Table 5: Travel Value - According to Income.

PN 1 2 3 4 5

TV（

）

3.1308 3.1486 3.37 3.5007 3.6327

Table 6: Weights - According to Income.

i-j(

ijp

)

1 2 3 4 5

1 0.01 0.37 0.11 0.01 0.01

2 0.01 0.57 0.31 0.02 0.01

3 0.01 0.03 0.41 0.05 0.01

4 0.68 0.02 0.07 0.42 0.43

5 0.28 0.01 0.09 0.50 0.54

5 FORECAST ANALYSIS

Firstly,because the background is under the spring

festival, some predictions may not match exactly

with the normal. During the spring festival, different

passenger may have different considerations with the

usual, such as the price, the time and the

environmental dimension. Many people may take

"as long as arrive the destination" into main

consideration in the pessimistic circumstance, but

not for the other factors.

5.1 According to the Purpose

According to Table 7, business travelers will choose

plane for the next trip,passengers to tourism and

work will choose EMUs for the next trip,passengers

to school will choose long-distance bus and

passengers to home and transfer will choose direct

trains for the next trips. While the others will choose

EMUs.

Because the passengers for tourism have higher

requirements for the time and environment,

moreover, most of them are free trips, their

sensitivity to the price level is relatively lower. The

predicted result showed they would choose plane for

their next trips, this is more objective. Passengers for

home and transfer have higher requirements for

HIGH-SPEED RAILWAY BASED ON GENETIC ALGORITHM FOR PREDICTION OF TRAVEL CHOICE

Table 7: Predictions - According to the Travel Purpose.

PN 1 2 3 4 5 6 7

Predictions 4 1 1 3 2 2 1

Figure 2: Predictions - According to the Travel Purpose.

time, so the predicted results are more objective.

Figure 2 is the application made by matlab. It is a

change tracking map of the total effective value and

the mean of the population.

5.2 According to the Cost Mode

According to table 8, passengers who travel at public

expense will choose plane next trips and the others

who travel at their own expense will take direct

trains next trips.

Table 8: Predictions - According to the Cost Mode.

PN 1 2

Predictions 4 2

Passengers who travel at their own expense have

low sensitivity for the price, because they do not pay

the fees, they will take time or comfort into

consideration. So the predictions are more objective.

However, the train is less expensive compared with

other mode of travel, the passengers who travel at

their own expense choosing the direct train is more

realistic.

Figure 3 is the application made by matlab. It is a

change tracking map of the total effective value and

the mean of the population.

Figure 3: Predictions - According to the Cost Mode.

5.3 According to Income

According to Table 9, low and lower income

passengers will take direct trains as their next trips,

middle income passengers will choose long-distance

bus, high-income passengers will choose EMU and

higher-income passengers will choose planes.

Table 9: Predictions - According to Income.

PN 1 2 3 4 5

Predictions 2 2 3 1 4

In real life, most low-income passengers will

choose direct trains as their travel mode for the price

is lower, and this is very beneficial to those low-

income passengers. However, middle-income

passengers will be on a more balanced consideration

of all aspects, so the predictions are more realistic.

High and higher income passengers are less

sensitivitive to the price, what they care about are

time, comfort, so the predictions is also close to

reality.

Figure 4 is the application made by matlab. It is a

change tracking map of the total effective value and

the mean of the population.

Figure 4: Predictions - According to Income.

ICEIS 2011 - 13th International Conference on Enterprise Information Systems

6 RESTRICTIVE

This paper relies on the background of spring, so the

passengers' choices would be some different with the

usual. For the Spring Festival, the biggest feature is

not very easy to purchase, sometimes one even

cannot buy a well-content ticket though he/she has

lined up row for a few days. Even worse, he/she may

buy a standing ticket. Under such circumstances, the

passenger will automatically consider less about the

price, time and environment, at least he/she can

reach his/her destination. So, one could reduce the

sensitivity to price even he/she cares a lot about the

price at ordinary times, one could reduce the

requirements on time even he/she concerns a lot

about the time in peacetime and one could ignore the

comfort of the tool even he/she demands high at

ordinary times. The data obtained from the

questionnaire is limited, this will lead to different

degrees of restriction of the final prediction results,

that is, the universality of the prediction is

restrictive.

7 CONCLUSIONS

The truth "survival of the fittest" has always been

throughout the genetic algorithm, it is applied to

solve the optimization problem,especially for fuzzy

problems and has very good robustness. However,

factors influence high-speed railway passengers

choosing travel mode are complex and diverse, the

most important and elusive is the passenger's

personal preferences. Personal preferences of

passengers are difficult to be precise analysis in any

case because it involves personal changes in mental

activity. When predicting the travel mode, we should

first analyse which factors influence travelers to

choose the travel mode, such price, time and

environment are factors from the external, and

passengers' personal characteristics are partly on

behalf of their personal preference. The basic idea of

the genetic algorithm to predict the travel selection

methods is that the choice of traveling factors will

act on the passengers, whether subjective or

subconscious, when passengers choose their travel

modes, they will maximize their benefits. Thus,

making the fitness function value for the

effectiveness of passenger travel, that is, the bigger

the value of the fitness function, the greater the

benefit of the passenger, and the more satisfied the

passengers. So, under the guidance of this thought,

we made predictions of various different types of

passengers choosing travel modes.

Passengers choice of travel mode can show

whether of such services industry is doing well, and

the predictions at least can provide some good

advice for those services sectors who have fewer

passengers, such as what to improve and how to

improve from price, time and environmental

comfort, etc.

REFERENCES

Zhang Qiang, Wang Bin, Zhang Rui, Xu Chun-xia, 2008.

Genetic Glgorithm-Based design for DNA sequences

sets. Chinese Journal of Computers, 2008, 31 (12)

:2193-2199.

He Wei-hui, Wang Jia-Lin, Hu Long-sheng, 2009. The

improvement and application of real-coded multiple-

population genetic algorithm. Chinese Journal of

Geophysics, 2009,52 (10): 2644 ~ 2651.

Ma Yong-jie, Ma Yi-de, Jiang Zhao-yuan, Sun Qi-guo,

2009. Fast genetic algorithm and its convergence.

Systems Engineering and Electronics. 2009,31 (3)

:714-718.

Ding Jian-li, Chen Zeng-qiang, Yuan Zhu-Zhi, 2003. On

the Combination of Genetic Algorithm and Ant

Algorithm. Journal of Computer Research and

Development. 2003,40 (9) :1351-1356.

Xu Yonghua, Huang Lijun, 2008. Improved Genetic

Algorithm in milk transport vehicles Road King

optimization. Northeast Agricultural University

.2008,39 (11) :111-115.

Gong Gu, Zhao Xiang-jun, HAO Guo-sheng, Chen Long-

gao, 2009. Study and implementation of Genetic

Algorithm Based on improvment of search space

partition. Journal of Henan University Narural

Science. 2009,39 (6) :631-636.

Jiang Yun-cai, Wang Qing-bo, 2004. Application of

MATLAB Genetic Algorithm Toolbox. Journal of

Jiangxi Electric Voactional and Technical College.

2004,17 (3): 42-44.

Qiang Li-xia, Yan Ying, 2006. Analysis on the Difference

of demand for passenger transport of high-speed

railways between home and abroad. Railway

Transport and Economy. 2006,28(9):18-21.

Wu Qun -qi, Xu Xing, 2007. Mechanism research on

travelling choices of passengers. Journal of Chang'an

University(Social Science Edition). 2007,9(2) :13-16.

The Dvelopment of the world's high-speed

railway,2006. Railway Survey and Design. 2006,1: 54-

56.

Chen Zhang-ming, JI Xiao-feng, 2008. Study on features

of railway passengers' travel activities. Railway

Transport and Economy. 2008,30(11): 23-25.

HIGH-SPEED RAILWAY BASED ON GENETIC ALGORITHM FOR PREDICTION OF TRAVEL CHOICE