HIGH-SPEED RAILWAY PASSENGERS’ CHOICES OF TRAVEL
FORECAST BASED ON MATLAB NEURAL NETWORK
Gao Yue, Li Jing
School of Economics and Management, Beijing Jiaotong University
Shangyuan Road 3, Haidian District, 100044, Beijing, China
Fan Yuhang
School of Economics and Management, Beijing Jiaotong University
Shangyuan Road 3, Haidian District, 100044, Beijing, China
Keywords: BP Neural network, High-speed railway, Passenger travel, Environmental factors.
Abstract: As a newly developing mode of transportation, high-speed railway is expanding its influences on national
economy and social life. At present, domestic research on high-speed railway mostly focus on tech level,
no systematic and comprehensive research have been done to the aspect of passenger travel. This study
taking uses of Matlab 6.6 concentrates on environmental factors’ effects on travel choices of High-speed
railway passengers, building up a forecast model based on BP Artificial Neural network. Through the
comparison and analysis of predicted and real data, effectiveness of this method is proved.
1 INSTRUCTION
Nowadays, the high-speed railway system is
considered as one of the most impressive
achievement in the railway-high-tech area around
the world. It is prevailing in the currency of the
railway developing projects in lots of countries, not
only because of its huge transport capacity, fast-
speed, and high security confidence and also due to
that it owns perfect punctuality, convenience and
comfort. What’s more, it consumes less fuel and is
friendlier to the environment. As an emerging way
of transport with a fascinating prospect, high speed
railway system is foisting his influence on regime
economy growth, air transport and other aspects in
our country’s development in socio-economy level.
Most of today’s researches on high-speed railway
in our country are focusing on tech level, and lost a
system and overall discussing on the factors
imposing on travelers. Travelers, as the clients of
the high-speed railway , their trips should be
covered and analyzed as a important issue in order
to provide constructive advises, which may be
significant in the future development of high-speed
railway system.
High-speed railway system in our country transfers
an enormous passenger flow every year. Unbalance
is the passenger demand in both aspects of area and
time that it is pretty difficult for people to predict
the passenger travel trend. The ticket supply is
often inadequate to meet the demand at certain
times of a year. Besides, high-speed railway
system’s ability to transfer passengers is restricted
to some limits like the volume of freight traffic.
Due to the multifarious and complicated factors
involved, in which most one are none-linear,
apparent limitations may exist if traditional
aggression methods were used. Nowadays, our
economy’s rapid development has brought
significant improvement to people’s life, as a
consequence that Citizens have put more emphasis
on degree of comfort and environmental factors
rather than traditional factors like price, distance
and time. If the high-speed railway system in our
nation wants to serve better and accomplish its
historical task of providing more conveniences to
the public, it has to put sufficient attention on its
environment construction. To reiterate, a forecast
model on high-speed railway passengers’ travel
was built focused on the environmental factors in
traveling.
254
Jing L., Yue G. and Yuhang F..
HIGH-SPEED RAILWAY PASSENGERS’ CHOICES OF TRAVEL FORECAST BASED ON MATLAB NEURAL NETWORK.
DOI: 10.5220/0003478302540258
In Proceedings of the 13th International Conference on Enterprise Information Systems (ICEIS-2011), pages 254-258
ISBN: 978-989-8425-54-6
Copyright
c
2011 SCITEPRESS (Science and Technology Publications, Lda.)
The artificial neural network grew up in the
1980s, developing through several periods
including initial growth, upsurge and climax,
troughs and the second rapid growth. The artificial
neural network imitates the real biological neural
network, consisting of quantities of neurons that
have none-linear mapping ability and connecting
each other with weight coefficients. The massive
parallel structure helps the network to acquire
wonderful characteristics like Automatic
Knowledge Acquisition, robustness, fault-
tolerance, learning ability and adaptability, and to
show superiority in setting up models for
complicated systems than traditional model
analysis models.
Because of its specific structure and methods in
dealing information, the artificial neural network is
being made great use of in numerous actual
applications, such as automatic control, image
processing, pattern recognition, signal processing,
robot control, welding, geographical analysis, data
mining, military affairs, transportation application,
mining industry, agriculture research, and
meteorological analysis and so on. Among a few
frequently-used neural networks, the BP neural
network is used most extensively, and its practical
utility is best embodied in forecasting time series
like national economy and population
development.
As a none-linear simulation technology with
fault-tolerance, learning ability and adaptability,
the artificial neural network provides a feasible and
practicable solution to solving problems concerning
high-speed railway passengers’ travel choices.
2 PRINCIPLES AND THE
STRUCTURE OF THE BP
NEURAL NETWORK MODEL
FOR FORECASTING
PASSENGERS’ TRAVEL
CHOICES OUT OF
ENVIRONMENTAL FACTORS
The study attempted to build a forecast model
using BP neural network to research on
environmental factors that affect high-speed
railway passengers’ travel choices based on a
survey conducted from January 1 to January 30 in
2010 with 1270 questionnaires that were
distributed at random to high-speed railway
passengers. Environmental factors that affect
passengers’ travel choice are definitely of various
kinds, the questionnaires used in the research
picked two main aspects including comfort level
and security level. Each of the aspects was set to
separate into eight degrees from “extraordinary
bad” to “extraordinary good”, from which
respondents could choose according to their own
experiences and feelings. Considering that different
means of conveyances would have various
influences on passengers’ selections, a third impact
factor was taken into account, namely “means of
conveyance”. The options set in this factor were:
1.hard-seat, 2.soft-seat, 3.hard sleeper, 4.soft
sleeper, 5.standing-room ticket, 6.unknown. For
the reason that some other factors besides
environmental factors would also have impacts on
passengers’ travel choices, all the others factors
were concluded into a forth aspect falling under
passengers’ degree of satisfaction as a whole in
order not to jeopardize the research on
environmental factors. The options set in the forth
aspect were also separated into eight degrees from
“extraordinary satisfied” to “extraordinary
unsatisfied” for respondents to choose from. In
view of the prediction of passenger’s next choice
on various traffic modes, a fifth question was set as
“The next traffic mode you would choose:”, and
the options were as follows: 1. the Multiple Units,
2. The Through Train, 3. The inter-city bus, 4.
Aircrafts, 5. Self-driving. The ultimate result of the
survey consisting of 1270 eligible questionnaire,
from which 503 valid records were screened based
on Rough Set methods.
The BP Neural network usually consists of
three layers including the input layer, the hidden
layer and the output layer. The numbers of nodes in
the input and output layer are decided by sample
dimensions constructed by data processor, and the
number of nodes in the hidden layer is determined
by both of the numbers of the input and output
layer. The work process of the Neural network is
composed of two stages, one is the work period, in
which the weights connected each nodes are fixed
and calculating each unit’s state change in order to
keep stability; the other is the learning period, in
which each calculation unit’s state is fixed while
the weights connecting them are changeable and
modified. A neural network is train for the purpose
that a group of input vectors could produce a group
of expecting and efficient output vectors. The
training process is realized through certain process
(Learning Algorithm) in which the network’s
weights are modified based on a series of training
samples.
HIGH-SPEED RAILWAY PASSENGERS' CHOICES OF TRAVEL FORECAST BASED ON MATLAB NEURAL
NETWORK
255
Figure 1: Structure of BP Neural network.
The BP Neural network model for forecasting
passengers’ choices of travel out of environmental
factors consists of two parts: data processing unit
and the BP Neural network. The real data of
passengers’ travel was processed in the first part
and formed the sample for experiment. The input
layer of the BP neural network is set to include four
nodes according to the four influence factors in the
investigation data, each of which separately
represents the degree of comfort, the security
standard, the pattern of embarkation and the
general satisfication degree. The output layer has
one node representing the choices of travel for the
next time. The number o nodes for hidden layer
could refer to the formula below:
L<n-1 (1)
L<(m+n)1/2+a (2)
L<㏒ 2n
(3)
In the formulae above, n represents the number of
nodes in the input layer; l represents the number of
nodes in the hidden layer; m represents the number
o nodes in the output layer; a could be any constant
between 0 and 10.
This study picks 5 as the number of the hidden-
layer’s nodes based on experiences.
3 METHODS FOR
FORECASTING PASSENGERS’
CHOICES OF TRAVEL OUT OF
ENVIRONMENTAL FACTORS
BASED ON BP NEURAL
NETWORK
When using the BP neural network model to
forecast the passengers’ behaviors, two main steps
are built including data processing and the neural
network establishment. The concrete algorithm is
as follows:
Figure 2: Algorithm process of the BP neural network
forecast model.
Step 1: 400 sets of data for training are randomly
selected from 503 sets of original data that are
effective. The other 103 sets of effective data are
used for forecasting. Input represents the input data
and output represents the output data.
k=rand(1,503);
[m,n]=sort(k);
input_train=input(n(1:400),:)';
output_train=output(n(1:400),:)';
input_test=input(n(401:503),:)';
output_test=output(n(401:503),:)';
Step 2: Normalization of data for training. To
normalize the training data can largely improve the
Network’s training speed and accuracy. Function
mapminmax in Matlab is chosen to normalize the
training data.
[inputn,inputps]=mapminmax(input_tra
in);
[outputn,outputps]=mapminmax(output_
train);
Step 3: Build up the Neural network.
net=newff(inputn,outputn,5);
Step 4: Establishment of neural network parameters.
With data normalized, less iteration times are needed
when the number of experiment data is not
enormous. The learning rate usually is set between
0.01~0.1 according to various models. So do the
choice for the goal of accuracy .
The learning rate is chose at 0.05 in this study and
iteration times is set at 1000 based on experience.
The goal of the training is initially set at 1 e-10.
net.trainParam.lr = 0.05;
net.trainParam.epochs = 1000;
net.trainParam.goal = 1e-10;
Step 5: Training for BP neural network. The neural
network training's performance depends on choices of
transfer function, training function and many other
parameters. Here the most commonly used functions
in the matlab neural toolbox are taken as the tool for
training the BP Neural network.
net=train(net,inputn,outputn);
Step 6: normalization of the predicted data and the
counter-normalization of the BP neural network’s
forecast outputs. Because the training data has been
normalizd before, the same process should be taken to
the data or forecasting. Use the trained Neural
network to forecast output data and counter-
normalize the output
data.
inputn_test=mapminmax('apply',input_te
st,inputps);
an=sim(net,inputn_test);
BPoutput=mapminmax('reverse',an,output
ps);
Step 7: Analyze the forecast result and error.
ICEIS 2011 - 13th International Conference on Enterprise Information Systems
256
4 APPLICATION AND THE
FORECASTING RESULT
In order to evaluate the forecast result, the survey data
was used to conduct a simulation research based on
Matlab 6.6. A BP Neural network consisting of four
input nodes, one output node and five nodes in the
hidden layer was built and trained to forecast the
environmental factors’ impacts on High-speed railway
passengers’ travel choices, and the result is shown in
figure3. Figure 4 represents the forecast error in the
training and forecasting process with the neural
network. Figure 5 shows the training performance of
the network.
Figure 3: The BP neural network’s forecast result on
passengers’ travel choices.
Figure 4: The forecast error of the BP neural network
model.
Figure 5: The training performance of the BP neural
network.
After 1000 times of learning process, the BP
Neural network achieved the best performance with
a minimum error of 2.2884e-009, and the network
came to a convergence ending the training process.
The forecast result concluded in this research is
scientific and accurate, demonstrating the validity of
using BP neural network in analyzing environmental
factors’ effects on High-speed railway passengers’
travel choices.
5 CONCLUSIONS
Based on the establishment of BP neural network,
we take the influences that the complicated and non-
linear environmental factors have on High-speed
railway travel into the extent that it can be predicted
and analyzed. The neural network built in this
experiment is of high accuracy. In practical
application, only with a few basic information
including travel mode, comfort degree, security level
and degree of satisfaction could researchers use this
model to forecast what kind of mode passengers will
take for travel in their next trip. Through controlling
variables, we could also study what will change in
passengers’ travel choices in the event of various
environmental factors. Thus, it will aid in decision-
making of forecasting traffic flow. At the mean time,
our government could control or improve those
factors that have enormous implications in affecting
people ‘s choices according to our model’s result, so
as to better macro-control our traffic systems and
improve citizen’s travel quality.
The specific characteristics of neural network that it
has a good simulation of nonlinear operation and can
deal with self-organizing and self-learning problems
HIGH-SPEED RAILWAY PASSENGERS' CHOICES OF TRAVEL FORECAST BASED ON MATLAB NEURAL
NETWORK
257
make it unnecessary for us to analyze in detail what
distinctive futures like safety and comfort degrees
and how big the influence on choosing traveling
means each factor has, but simply to choose formats
of study samples data and some parameters of
learning process for neural network, we can make
the neural network learn by itself, thus avoiding the
condition that the man-made model that was built
according to the traditional methods does not match
with the actual situation, which leads to inaccuracy.
But, in terms of the environmental factors affecting
traveling, it is certain that the factors we choose
might be incomplete, and we can’t eliminate the
influences of other factors such as price and time
besides the environmental factors on the choices of
travel modes. What is more, a better solution of
choosing parameters for the established BP neural
network may exist. How to establish a more
effective neural network model remains a valuable
subject. The research is based on the popular topics
on environmental factors in recent years,
establishing a BP neural network model with a small
error and high precision, which could be a beneficial
and promising explore in this area. To reiterate, a
further experiment and research is expected.
REFERENCES
Yao Ru-qing, 2010.The regional economy comes into the
era of high-speed railway. Zhejiang.Economy. 2010,5
Ma Yan-xia, 2007. Opinions on the development of
artificial neural network. Journal of Jilin Radio and
Television University. 2007, 5
Jin Xing-ji, Jia Wei-wei ,2008. Review of researches on
artificial neural network. Journal of Forest Technoloy.
40(1), 65-66.
Ren Ying-wei, Song Ling-ling, Wu Shou-ren, 2009.
Application of BP neural network in traffic
engineering. Traffic Engineering. 2009,20,136-137.
Xu Jin-qiang, Zhang Yu-qing, 2006. Urban traffic volume
prediction based on BP neural network. Automotive
Electronics. 2006, 5, 23.
Qiu Shi-qi, Wang Qi, 2009. Freeway traffic incident
detection based on BP neural network. China
Measurement & Test. 2009, 35（2), 48-52.
ICEIS 2011 - 13th International Conference on Enterprise Information Systems
258
