Adaptive Feature Fusion Networks for Origin-Destination Passenger Flow Predic...
ThesisPresentation
1. A platform for enhanced
transportation services
GEORGE GATT
THESIS PRESENTATION FOR THE MSITT PROGRAM
ATHENS INFORMATION TECHNOLOGY
MARCH 2015
THESIS ADVISOR DR SOFOKLIS EFREMIDIS
2. Thesis Objectives
Development of a platform for calculation of the optimal route to a user
specified destination considering the time schedules of various public
transportation means (bus, metro, tram) and the traffic load of the network
Design of an Android application that allows users to set their requirements
and receive the shortest route at current time as a set of directions to follow
Demonstration of the platform with some sample scenarios
4. INTRODUCTION (I)
The problem of finding the optimal time delay route between two points in
an urban environment can have many positive implications (reduce travel cost
and time, reduce traffic jams, improve tourism services)
The user selects an origin address, a destination address, a maximum walking
distance between stations and the types of public transportation means
The server of the service provider receives the request from the user and computes
the shortest path at the time of request by retrieving some dynamic traffic data
from the server of the transport provider
5. INTRODUCTION (II)
A time dependent shortest path (TDSP) algorithm is applied for the
calculation of the optimal route under certain assumptions
The server of the service provider returns its response to the Android client
This platform faces some important computational challenges (limit of
daily requests to the Google Maps API, many client requests at the server
provider, large transportation graphs)
Future improvements are possible by a more reliable real-time estimation
of the traffic load of the network
6. FEATURES OF THE ITRAFFIC APPLICATION
The application gives a user the choice to configure some
parameters and set some preferences for future queries
The application gives a user the choice of setting the origin and
destination addresses either as a string of characters or by clicking
a point on the map
The application displays the optimal route as a list of data and as
a sequence of line segments on the map
8. TOOLS
Android Studio and AVD manager
(http://developer.android.com/sdk/index.html)
Online tool for geocoding (http://mygeoposition.com)
Netbeans v.8 IDE (http://www.netbeans.org)
Glassfish v.4 application server (https://glassfish.java.net/)
MySQL database server (http://www.mysql.com)
10. ARCHITECTURE AND DESIGN
General Architecture of the iTraffic System
Database Model
Transport Provider
Service Provider
Android Client
11. GENERAL ARCHITECTURE
iTraffic
Android
application
(Client)
RESTful Web
Service on
Glassfish
(Service
Provider)
RESTful Web Service for buses
(Transport provider)
RESTful Web Service for
subway (Transport provider)
RESTful Web Service for tram
(Transport provider)
RESTful Web Service for
suburban (Transport provider)
RESTful Web Service for rail
(Transport provider)
MySQL
MySQL
MySQL
MySQL
MySQL
12. DATABASE MODEL
Database model is the same for every type of public transportation
The Velocity values are changed randomly by the transport provider
in order to simulate the current load of the traffic network
13. TRANSPORT PROVIDER
INTERACTION OF COMPONENTS ON GLASSFISH
Glassfish App Server
ORM
(JPA) MySQL
Database
Business
Layer (EJB)
REST
Interface
(JSON,
XML)
Web Interface (JSF)
Web Browser
Service
Provider
14. SERVICE PROVIDER
INTERACTION OF COMPONENTS ON GLASSFISH
Glassfish App Server
REST
Interface
(JSON)
Shortest
Path POJO
Initialization Servlet
REST
Methods
Transport
ProviderAndroid
Client
15. TIME DEPENDENT GRAPHS
Graphs with time-varying edge weight functions
First-in first-out (FIFO) and non-FIFO graphs
Examples: road networks (FIFO) for urban trip planning or
hazardous material routing, communication networks (non-FIFO)
for flow control of packet transmission
Definition of TDSP from node 1 to node 2: earliest arrival time
path at node 2 starting from node 1 at a given time instance t
17. MOST IMPORTANT TDSP ALGORITHMS
A. Orda and R. Rom, "Shortest-path and minimum-delay algorithms in networks
with time-dependent edge-length", in J. ACM, 37(3), pp. 607-625, 1990
K. Sung, M. G. Bell, M. Seong, and S. Park, "Shortest paths in a network with time-
dependent flow speeds", in European Journal of Operational Research, 121(12),
pp. 32–39, 2000
B. Ding, J,X. Yu, and L. Qin, "Finding Time-Dependent Shortest Paths over Large
Graphs", in Proceedings of the 11th Int. Conference on EDT: Advances in Database
Technology, EDBT' 08, March 25-30, pp. 205-216, 2008, Nantes, France.
K. Androutsopoulos, and K. Zografos, "Solving the k-shortest path problem with
time windows in a time varying network", in Operations Research Letters, 36,
pp. 692-695, 2008
M. Omran, "Path Problems in Geographic Information Systems", PhD Thesis,
Carleton University, Ottawa, Ontario, 2014
18. TDSP ALGORITHM AT SERVICE PROVIDER
Initialization: (Step 1)
Xs=ts, fs=NIL; for every k≠s Yk=∞, Xk=NULL, fk=NIL; j=s;
Relaxation: (Step 2)
For all neighbors k of j for which Xk=NULL, do:
a. Yk = min{ Yk, Xj + Djk(Xj)}
b. If Yk changed in Step 2(a), then set fk=j.
If all nodes have non-null X-value, then stop. (Step 3)
Otherwise, let l be a node for which Xl=NULL and such that Yl ≤ Yk for
every k for which Xk=NULL . Set Xl=Yl , j=l , and proceed with Step 2.
This is the TDSP algorithm of Orda & Rom
For FIFO graphs the functions Djk(t) are the edge weight functions
The simplest but not the most efficient TDSP algorithm (O( IVI²) )
Returns the shortest paths spanning tree: j fj
Time delay for shortest path from s to j at time ts: Xj-ts
19. OUR GRAPH MODEL FOR PUBLIC TRANSPORTATION
Is a directed graph
Is not fully connected
At every client request its structure changes
The weight function between 2 nodes depends on the daily time
schedule of the line and on the traffic load (leg velocities) at request time
The vehicles’ waiting time at the stations and the client’s walking
velocity are considered constant
The graph is implemented using the adjacency list mapping for
neighboring nodes and neighboring edge weights
The values of leg velocities for future time instances that are needed
for Step 2(a) are the same with the values of leg velocities at request time
20. ANDROID CLIENT
The Android application provides a user interface for the clients to state
their travel requirements: origin address, destination address, maximum
desired walking distance, types of public transportation
For the last 2 parameters the client can configure the application
The origin and destination addresses can be set as a character string
or by clicking on the map
The application invokes the REST interface of the service provider using
the above parameters
The application parses the response and displays the optimal route on
the map
27. FUTURE WORK
Augment the database with the time schedules during the weekend
Simulate more accurately the variations of the leg velocities
(a) Receive and process real-time location data from drivers and users
(b) Process databases of real-time traffic data from the past
(c) Use trained models for reliable traffic load representation
(d) Combinations of the above
Decrease computation time and memory requirements
(a) Implement more efficient TDSP algorithms
(b) Store locally on the service provider graph data for the road network
(c) Use a cloud infrastructure
(d) Use the Neo4j graph DBMS in embedded form
Support more complex criteria for the route query
(a) Return the first k-shortest path routes
(b) Enable the use of car driving as a transportation mean
(c) Enable the setting of the optimal route request in the scope of a time window
(d) Enable dynamic updates of the optimal route from the service provider after
each client request