Parametric calibration of speeddensity relationships in mesoscopic traffic simulator with data minin

1
Parametric calibration of speeddensity
relationships in mesoscopic traffic simulator
with data mining

• Adviser Yu-Chiang Li
• Speaker Gung-Shian Lin
• Date2009/10/20
• Information Sciences, vol.179, no.12, pp.
  2002-2013, 2009

2
Outline
Introduction
1
Literature review
2
Data mining
3
Experiments and results
4
Conclusions
5
3
1.Introduction

• Calibrating speeddensity relationship parameters
  using data mining techniques, and proposes a
  novel hierarchical clustering algorithm based on
  K-means clustering
• Mesoscopic simulators aim to model either a
  single vehicle or a group of vehicles in order to
  depict any responsive actions of different
  vehicles to the traffic information.

4
2.Literature review

• In the mesoscopic models which are used in DTA
  systems

0
5
3.Data mining

• LWR(Locally weighted regression)

• Step 1 Take x (densities or both densities and
  flows make up the x) as a center to
  form a space. The width of the space isdescribed
  by the
• q fn

• Step 2 Define the weights of all points in
  specific sectors. The weight of any point
  is the height of a weight function. The common
  weight function is selected

The weight for the observation (xi, yi) is
Step 3 Fit a polynomial for each point in an
independent variable space by using the weighted
least square algorithm
Step 4 Acquire the value of yi.
6
3.Data mining
x
qfn
p(x,xi)ltd(x)?W(u)(1-u3)3
p(x,xi) ?d(x)?W(u)0
7
3.Data mining

• Agglomerative hierarchical clustering algorithm
  based on K-means
• The proposed algorithm can be summarized as
  follows

Step 1 Use K-means to cluster the sensor data
which is taken as training instances, and these
k clusters are named as constraint- clusters.
Densities, flows and speeds contain abundant
information about the traffic status, so they
are chosen to cluster.
8
3.Data mining

• K-means is executed in the following steps
• 1. Randomly select k clustering centers from n
  training cases.
• 2. Find the nearest clustering center to each xi
  (density or both density and flow), then put xi
  in it.
• 3. Compute the objective function E. If the value
  of E is unchanged, we should consider that the
  results of the clustering are also unchanged.
  Then the iteration stops.
• 4. Otherwise, it will return to 2.

9
3.Data mining
k3
10
3.Data mining
11
3.Data mining

• Step 2For each constraint-cluster, use the
  agglomerative hierarchical clustering to
  build a clustering tree.
• The basic steps of the complete-link algorithm
  are

1. Place each instance in its own cluster. Then,
compute the distances between these points.
2. Step thorough the sorted list of distances,
forming for each distinct threshold value dk a
graph of the samples where pairs of samples
closer than dk are connected into a new cluster
by a graph edge. If all the samples are members
of a connects graph, stop. Otherwise, repeat this
step.
3. The output of the algorithm is a nested
hierarchy of graphs, which can be cut at the
desired dissimilarity level forming a partition
(clusters) identified by simple connected
components in the corresponding subgraph.
12
3.Data mining
13
3.Data mining
??1
??3
??2
??6
??7
??5
??4
14
3.Data mining

• Step 3 These k clustering trees are combined as
  an integrated clustering tree by using the
  complete-link algorithm. After all samples are
  clustered, a separate local regression will be
  run for the observation in each cluster.
• Step 4 The new densities and flows are
  classified to the most appropriate cluster by
  using k-nearest neighbors. The k- nearest
  neighbor sorter uses Euclidean distance to
  search k densities and flows samples completed
  clustering.

15
4. Experiments and results

• The sensor data are preprocessed to eliminate
  erroneous data and repair missing ones.
• Step1 Define data in some cycles as data it is
  in some phase and scan the sending time of data
  one by one to find out the missing ones. Check it
  is erroneous or not according to the criteria in
  Table.

• Step 2 Repair the missing data and the erroneous
  data. The average value in the neighboring phase
  is used to repair these data.

16
4. Experiments and results
17
4. Experiments and results
18
4. Experiments and results
Estimated speed by the classical speeddensity
relationship
19
4. Experiments and results
20

• Table presents the RMSPE obtained by each
  approach.

21
5. Conclusions

• The proposed method overcomes the limitations of
  classic models of speeddensity relationships.
• KHCA obtained the highest precision in capturing
  traffic dynamics compared to other existing
  clustering algorithms.

22
Thank You !