Advances in Transportation Studies an international Journal 2015 Special Issue, Vol. 1
the bus is not comfortable and convenient than private car, most travelers are still like the means
of transport as buses. Meanwhile, an important measure of a city's traffic situation is the degree of
development of public transport. Study of bus arrival time prediction system help residents ease
anxiety of mind, develop a tolerable travel plans and meet modern city’s breath [4-6].
2. Current research status of the bus arrival time prediction
In recent years, most countries are conscious of the precise real-time transit vehicle arrival
time prediction occupies an important position in intelligent transportation technologies or solve
traffic problems [7-10]. So many countries use the advanced equipment and bus information
collection tools, integrated multiple transportation factors to the study buses arrival time
prediction technology. Aimed at predicting the arrival time and release real bus arrive time for
reducing the waiting time and help travellers choose travel routes rational, reasonable
arrangements [11-14]. Bus arrival time needs amount of data and easily to be disturbed by other
factors. This is its characteristic. Establish a reliable image of the city, promoting urban public
transport and intelligent transport application and development.
3. The SUPPORT VECTOR MACHINE and Kalman filter algorithm
3.1. SUPPORT VECTOR MACHINE algorithm
SVM is a static algorithm and need amount training data. We can get a result from the training
model. KF is a dynamic algorithm that can fix the trained data from SVM. The real-time data is
needed, like previous arrival time, time difference between two stations and so on.
SVM is a data mining methods [15-16], which can be very successful in handling regression
problems (time series analysis) and pattern recognition (classification, discriminate analysis) and
many other issues. It can be utilized in a variety of disciplines such as science, engineering and
management. Currently international SVM in theoretical research and practical application are
both in rapid development. SVM and KF can not only overcome dependence on the amount of
training data, but also has a strong anti-jamming capability. Bus system has strong regularity; high
stability. The application of the integrated forecasting model can achieve high precisely real-time
predictions. That’s why we choose these two methods. It is also commonly used in statistical
classification and regression analysis. Vector will be mapped to a higher dimensional vector
space, and establish a maximum margin hyperplane in this space. There has two mutually parallel
sides hyperplane in the hyperplane which separating the data, maximize the distance between the
two parallel hyperplane. Mathematical form of hyperplane can be written as:
(1)
where is the point on the hyperplane, is the vector perpendicular to hyperplane.
According to the geometric knowledge, we know that w vector perpendicular to the
hyperplane. The purpose of adding the displacement of b is to increase the interval. If have not b,
then it will have a hyperplane passing through the origin, limits the flexibility of this approach.
Since we require the maximum interval, so we need to know a parallel to support vector (with the
best hyper-plane) and the nearest hyperplane to support vector. We can see that these parallel
hyperplanes can denote as the equation:
(2)
(3)