Cognitive Urban Transport

Cognitive Computing : Technical Leadership Exchange
© 2016
IBM Internal Use Only - Not for Client Distribution
Cognitive Urban Transport
Sasha Lazarevic, 2017
Presented for the 1st time in Zürich, 2016 November 16

© 20162
Sasha Lazarevic
Current Public Transport
• Living in a city requires daily commuting. But people waste
a lot of time changing from one transport to another.
• Public transport has for many years been organized along
the main axes, but they are not aligned with the
passangers’ itineraries
• The total waste goes up to 50 or 60% of the commute time
• As the result, many commuters prefer taking cars, which
produces congestion, pollution and loss of productivity
The existing public transport is not optimized
to reduce the time and energy waste

© 20163
Sasha Lazarevic
New Concepts Are Not Addressing the Problem
• Car-free zones in the cities reduce the pollution but reduce the connectivity of
these zones with other urban areas.
• Electrical cars reduce pollution, but do nothing to reduce traffic jams.
• Autonomous vehicles allow during traffic jams to use our time productively, but
they don’t reduce the time to arrive from point A to point B.
• Electrical bycicles offer more flexibility. But their range is limited and in some
cities present security risks.
• Car sharing helps to better use cars, but the drop-off sites are far away from our
destinations.
• Ride sharing among colleagues is not an organized way of moving masses of
people and has only limited effect on the overall problem
The only solution that can resolve the traffic congestion
problem, eliminate the time and energy waste and in this
way unlock a tremendous human potential is

© 20164
Sasha Lazarevic
Cognitive Solution
• You are at home and decide to go to the point B within your urban area.
• With the mobile app you select the departure (point A), and destination (point B).
• Thousands of people around you order their bus ride the same way.
• Optimization algorithm with machine-learning capabilities determines the route
that will take the most optimal itinerary to satisfy the maximum number of
passangers sharing the same itinerary.
• You receive the confirmation of the bus arrival time at the nearest bus stop, and
the bus ID.
• You walk to the nearest bus stop to take your bus.
• When you enter the bus, your app will automatically check in.
• A fleet of electrically powered autonomous buses is used to drive passangers.
• Itinerary is visualized on your app, so you can track the distance to the point B.
• If it rains or is cold outside, you relax and enjoy sitting - there is no need to change
the transport.
• After the bus has finished the route, it receives a new itinerary, which always
depends on the passangers’ requests and optimization algorithms.

© 20165
Sasha Lazarevic
CUT Application
Smartphone application for :
 Ordering
 Bus check-in
 Ticket payments

© 20166
Sasha Lazarevic
CUT Algorithm
Simplified case :
• Passanger 1 wants to go from A to B
• Passanger 2 wants to go from C to D
• Passanger 3 wants to go from E to F
Itinerary will be determined and optimized
to take into account all their requests
A
B
C
D
E
F
Problem is when thousands or people want to
move at the same time in different directions

© 20167
Sasha Lazarevic
Technical Solution : Metaheuristics + Neural Networks
Optimization stream 1
DataGathering
Logic
Learning
Neural Networks
External Systems
- weather, road works, police,
technical bus data, traffic lights,
train and airplane schedule
Front End Passangers
Payments
Security
Performance
Ordering Check-in moduleUser Experience
Mobile Apps
Interface devices
Feedback module
FleetManagementOthercognitiveservices
Optimization algorithms will be based on metaheuristics dial-a-ride route optimization methods. These algorithms will use as input the pre-
collected data and will produce the training set of data for the neural networks. Neural networks are used to handle the real-time requests.
Real-time data from the overall system will be used by both metaheuristics and neural networks to tune the optimization results.

© 20168
Sasha Lazarevic
Metaheuristics Algorithm
Static constraints:
- Number of vehicles
- Capacity of vehicles
Dynamic constraints (functions):
- Remaining capacity of vehicles
- Position of vehicles, which determines the k0 number of available vehicles at time t0
- Cost of a ride through a street, which depends on the road and traffic conditions
Let us define the variables and some basic functions:
k0 is number of available vehicles at time t0
U0, as set of unclassified requests (r0,1,..m) at time t0 , where ri is defined as (starting point, drop-off point)
Z(U0) /* Classification function, which classifies the requests per geograhical zones, where the result is geographical zone
classification matrix Z[ ] with k0 rows
R /* matrix with initial solutions with k rows, where each row will contain a set of requests
S /* matrix of solutions optimized through a local search function
C(ri) /* Cost function of one request ri
C(Rm) or C(Sm) /* Cost function of all requests assigned to one vehicle
G(R) or G(S) /* Global cost function of all rides for all available vehicles

© 20169
Sasha Lazarevic
Initial Solution
Initial solution is constructed based on the segmentation of requests into separate geographical zones. The number
of zones is equal to the number of available buses. Constraints of capacity of the buses and exact timing of requests
will determine what requests are admitted to which solution.
Z[ ] := Z(U0)
Repeat
n := n+1 /* n starting from 0
Rn = Initial solution construction algorithm (Zn) /* it creates k initial solutions
Until n = k0
G(R) /* calculate the first global cost baseline
R is a matrix with initial solutions with k rows, where each row will contain a set of requests

© 201610
Sasha Lazarevic
Local Search Optimization
Local search optimization goes through a matrix with k rows and for each row identifies the optimal order of requests.
n := 0
Repeat
n := n+1
C (Rn) /* calculate the cost function Rn
For each Rn do /* Rn is a request set from the initial solution for the bus n
S’ (Rn) /* Taboo search optimization metaheuristics, using local requests moves
C’ (Rn) /* calculate the cost function of the new solution, after requests moves
if C’ < C, then Sn = S’ /* construct matrix S with best local solutions with minimized cost function
End
Until n = k
G(S) /* calculate the global cost function for locally optimized requests in S solution matrix

© 201611
Sasha Lazarevic
Global Optimization
After local search has improved the initial set of solutions, go beyond that and performing request moves
between solutions (relocation, exchange, 2-opt moves)
Repeat
n := n+1 /* starting from n=0
For each Sn do
S’’ (S1,2..k) /* request move algorithm, with exclusion of Sn
if G(S’’) < G(S) then keep S’’ as S
End
Until n = k
G’’ will be the optimized cost function for globally optimized routes

© 201612
Sasha Lazarevic
• Input layer is used for the incoming requests in real time
• Number of neurons in the input layer := number of incoming requests
• Numer of neurons in the output layer := number of available buses
• Use supervised learning, where the correct outputs are represented by
associations of the requests to the selected bus
Neural Networks Pipeline

© 201613
Sasha Lazarevic
Key Benefits
• The economic progress depends on the transportation. Resolution of
problems like traffic congestion, polution, time and energy waste, will unlock
tremendous human potential, and help us achieve sustainable development.
• Convenient use of public transport significantly reduce the use of private cars
• Shorter bus rides with no changing, walking or waiting connecting fares
• Optimized routes that avoid producing congestions
This is a huge market opportunity: billions are spent
on the public transport and a lot can be saved with
an investment in this cognitive technology.

© 201614
Sasha Lazarevic
Not Only That..
• Multiple optimization algorithms can be used for different use periods (peak times,
weekends, night time, public holidays etc). This will allow for better precision and
real-time predictions
• Taking into account external events, like weather data, road works etc
• Integration of cognitive bus system with other transport systems like subways,
airports, train stations, traffic lights
• Learning and improving itinerary selection based on the real-time driving data,
customer feedback and overall resource utilisation
• Autonomous vehicles reduce cost for those buses which are in the waiting state
for the requests
• Smart bus transport system is ideal for those countries that don’t have developed
railway or urban subway networks

© 201615
Sasha Lazarevic
Challenges
• Cities and governments will need to understand how to manage the transition
to the new intelligent and dynamic system
• This is a big change – cities will need larger fleet of smaller vehicles, but not
all of them will be used all the time
• Integration with other transport systems like subways and city trains
• Data security to protect privacy of passangers’ trips and other personal data
• Deal with passangers without smartphone
• Guarantee the ride SLAs – arrival times
• Passangers who take ride later in the itinerary (closer to the final destination)
will later order the bus ride. For them the cognitive system will have to rely
more on the prediction capabilites
• Ideally this program should be followed by government stimulus to give up the
use of personal cars

© 201616
Sasha Lazarevic
How to implement?
• Cognitive Urban Transport is currently a concept, but it will definitely
be implemented during the next several years. Our competitors are
working on that (Uber, Google)
• The best business model would be:
i) development of a prototype system, together with few selected partners
in the domain of self-driving vehicles and fleet operation, with
involvement of our IBM research centers,
ii) pilot project in a smaller city and then
iii) replication on mass scale. New versions and additional modules can be
distributed from the central development team
• Revenues can be achieved from the business consulting services,
algorithms, infrastructure platform, implementation and integration
services, managed services and support

© 201617
Sasha Lazarevic
Business Development Plan
• Determine internal IBM governance structure, and identify the executive sponsor
• Define the joint go-to-market, innovation and delivery frameworks with selected partners
• Do the business case analysis (feasibility study), obtain management approvals and budget
• Inventory of reusable assets and synergies with existing cognitive and IoT programs
• Gather the business and engineering team, nominate technical and sales leaders
• Identify the initial roadmap and list of «quick wins» projects
• Detailed solution architectural design of the core elements: algorithms, machine learning
logic and application functionalities
• Technical infrastructure, coding
• Create demostration package and room for marketing purposes
• Architecture design for the additional, value-adding cognitive services
• Alignment with the partners in producing the first joint product
• Integration, tests
• Work with the existing and new customers to present the solution and obtain funding of the
pilot projects
• Adjustments in the technical solution based on the test and pilot results
• Training packages for the sales reps and architects
• 2nd roadmap, detailed planning and budgeting

Cognitive Urban Transport

More Related Content

What's hot (20)

Similar to Cognitive Urban Transport (20)

More from Sasha Lazarevic (11)

Recently uploaded (20)

Cognitive Urban Transport