Poisson Process
Most often applied to occurrences of events in time
Examples
Arrivals of passengers at a stop of a high-frequency bus route
Arrivals of vehicles at an isolated intersection
Arrivals of customers at a bank teller machine
Arrivals of telephone calls at a switchboard
etc.
Thereafter, the term “Arrivals” will be used to refer to occurrences
of events in a Poisson Process
34
Postulates of the Poisson Process
If we have a process involving a number of arrivals occurring
between t1 and t2 and is the arrival rate (i.e. number of arrivals
per unit time between t1 and t2). This process is called a Poisson
Process if the following conditions are met:
P{at least one arrival occurring in a short time period t} . t
The number of arrivals occurring at t2-t1 is not dependent on t1 or on
the number of arrivals prior to t1
| x xxxxx xx x|
t1 t2
E.g. the number of people arriving at a particular subway station between
3pm and 6pm is dependent only on the 3-hour period (3-6pm) and does not
depend on the starting time (3pm) or the # of arrivals before 3pm
35
Postulates of the Poisson Process
The number of arrivals occurring in disjoint time intervals are mutually
independent RVs
| x x x x |
t1 t1 t2 t3 t4 t2
E.g.If # of arrivals between t1 & t2 : X
# of arrivals between t3 & t4 : Y
Then
X & Y are mutually independent
Given that one arrival occurs at a particular time, the
conditional probability that another occurs at exactly the same
time is zero
this means no more than one arrival should occur at the
same time
36
Properties of the Poisson Process
For a Poisson Process occurring over time
interval t and having an arrival rate of
The number of arrivals in t, N(t), is Poisson
distributed with mean
k t
and variance t
( t ) e
P{ N (t ) k } for k 0 ,1, 2 ,....
k!
E { N ( t )} t
Var { N ( t )} t
The elapsed time to the occurrence of the next
arrival (i.e. inter-arrival time) is L and has an
exponential PDF (i.e. L is exponentially distributed)
with mean of 1/ and variance of 1/2
fL(x) = e-x for x0
E[L] = 1/
Var[L] = 1/2 37
Recall the no-memory property of exponential PDF
Properties of the Poisson Process
(cont’d)
The elapsed time to the occurrence of the kth arrival is
Lk and has a kth-order Erlang PDF
k x k 1e x
f Lk ( x ) for x 0 and k 1,2,....
( k 1)!
k
E[ Lk ]
k
Var [ Lk ]
2
For k = 1, obviously L1 is exponentially distributed
Lk can be thought of as the sum of k independent, and
identically distributed exponential RVs each with mean
1/
38
Properties of the Poisson Process
(cont’d)
An important property of the Poisson Process
the unordered arrival times in a Poisson Process are
independently and uniformly distributed over the fixed
time interval of interest
i.e. the probability of each arrival occurring at any time
between 0 & t is equally likely
39
Multiple Independent Poisson
Processes
If we have 2 Poisson Processes operating independently with
arrival rates 1 and 2 (e.g. arrivals of males and arrivals of
females at a particular bus stop)
The combined (i.e. pooled) process is also a Poisson Process with
N(t) = N1(t) + N2(t)
= 1 + 2
Probability that an arrival from Poisson Process 1 occurs before an
arrival from Poisson Process 2
P{x1 < x2} = 1 / (1 + 2)
Since the pooled Poisson Process has no memory, each arrival is
independent of other arrivals.
We can think of each arrival as a Bernoulli trial with a probability of 1 / (1
+ 2) of being type 1 and a probability of 2 / (1 + 2) of being type 2
40
Random Incidence
Bus Stop Example
Suppose that buses arrive successively at a particular bus stop
with the Headway H (defined as the time interval between
successive bus arrivals) either equal to h1 with probability p1 or
h2 with probability p2
Note that this a general point process (not necessarily
Poisson)
41
Random Incidence
Bus Stop Example
Now consider a passenger arriving at the bus stop at a
completely random time (i.e. arrival time of passenger
not influenced by the bus arrival process, which is the
case of high-frequency transit lines)
42
Random Incidence
Bus Stop Example
We are interested in 2 questions
Q1: What is the probability that the passenger arrives
during a headway of h1 and the probability to arrive
during a headway of h2?
Q2:On average, how long will the passenger have to
wait until the next bus arrives?
43
Pedestrian Crossing Problem
Auto Traffic
Pedestrian Traffic Pedestrian Traffic
L R
Pedestrian
Crossing Signal 44
Pedestrian Crossing Problem
It is desired to design a pedestrian light at a crossing
location, and 3 control rules are considered
Description
Pedestrians approach from the left/right side of the crossing in
a Poisson process over an extended period of time, with
average arrival rates of L / R pedestrians per minute
The left and right Poisson processes are mutually independent
Under any rule, pedestrians would wait at the crossing light
until it flashes, at which time pedestrians would cross the
road, then the light would instantly turn red again
Each time the light flashes is referred to as a “release”
45
Pedestrian Crossing Problem
The 3 considered rules are: You are required to analyze
each rule according to
Rule A: release every T 1. NL = expected number of peds
minutes crossing L to R on any release
Rule B: release whenever 2. = probability of 0 peds
the total number of waiting crossing L to R on any release
pedestrians (from right
plus left) equal No 3. W = expected wait time until
crossing for a randomly
Rule C: release whenever arriving ped (from L or R)
the first pedestrian (from R
or L) to arrive after the 4. V = expected wait time until
previous release has crossing for a randomly
waited To minutes arriving observer
46
Four Steps for Any Probabilistic
Modelling Experiment
Step 1
Define the random variables of interest
Step 2
Identify the joint sample space
Step 3
Determine the probability law over the sample space
Step 4
Work in the sample space to answer any question of
interest
47
Four Steps for Any Probabilistic
Modelling Experiment
Step 1
Recall in the stick cutting problem, we defined the 2
RVs X1 and X2 for the locations of the 1st and 2nd
marks, respectively
Step 2, Identify the joint sample space
X2
Joint sample space
X1
1 48
Four Steps for Any Probabilistic
Modelling Experiment
Step 3, Determine the probability law over the sample
space
X2
1
f(x1,x2)
X1
1
Step 4, Work in the sample space to answer any 49
question of interest
Extension of the 4-Step Approach
for Functions of RVs
In some problems, we may have one (or more) RV (say
X) of known probability law (i.e. PDF of X is known)
and we may be interested in deriving the probability law
of another RV (say Y) that is expressed as a function of X
This class of problems is called “Functions of RVs” or “Derived
Distributions”
In applying the 4-step approach to derived distribution
problems, we detail step 4 as follows:
Step 4a
Derive the CDF of the desired RV (i.e. Y), working in the sample
space of X
Step 4b
Take the derivative to obtain the desired PDF of Y
50
Example 1: Response Distance of an
Emergency Vehicle
Consider a highway of a specific length where an
emergency vehicle (EV) travels back and forth between
the two ends of a highway segment. If an incident (e.g.
crash, vehicle breakdown, etc.) occurs at any point along
the segment, the EV is dispatched immediately from its
current location to the location of the incident
| | | |
EV I
51
Example 1: Response Distance of an
Emergency Vehicle
Now, we are interested in asking questions about this
distance which the EV has to travel to reach the incident
(e.g. what is the expected distance the EV has to travel to
respond to incidents)
As such, we need to determine the probability law (i.e.
PDF) of the travel distance (Y) for the EV to reach a
random incident
Solution
fY(y) = 2(1-y)
52
Important Relationships in
Probability Theory
If X is a continuous RV with a known fX(x)>0 and Y is a
strictly increasing or decreasing function of X (i.e. Y =
g(X)), then
fY(y) = fX(x) . |dX/dY|, with X=g-1(Y)
If X is a RV and Y = g(X), and suppose we are
interested in E[Y]
E[Y ] E[g(X )] g(x i ).pX (x i ) if X is discrete
i
E[Y ] E[g(X )] g(X ). f X ( x)dx if X is continuous
53
Important Relationships in
Probability Theory
If X is a RV and “a” is a constant
E[aX] = a.E[X]
E[a] = a
Var[aX] = a2Var[X]
Var[a] = 0
54
Applications of the Previous
Relationships to “Scaling”
We often select a convenient scale for our problems, and
after we obtain the answer we apply scaling factors
Suppose we derived the probability law of the RV W,
and we wish to rescale it as V = aW + b, where “a” and
“b” are constants (scaling factors)
dW
fV (v ) fW ( w).
dV
V b dW 1
W
a dV a
v b 1
fV ( v ) fW ( ). for any a 0 55
a a
Applications of the Previous
Relationships to “Scaling”
The above equation could also be derived
using the 4 steps
Ifyou are only interested in the mean
value and variance of V
E[V] = a E[W] + b
Var[V] = a2Var[W]
56
Example 1: Revisited
Let’s reconsider Example 1 but with the
following changes:
Consider a rectangular response area of dimensions
Xo and Yo which has a grid street network
The location of a random incident is (X1,Y1) and the
location of a patrolling EV at the moment of the
incident is (X2,Y2). The 2 locations are independent of
one another, and each is uniformly distributed over
the response area
57
Example 1: Revisited
Y
grid street network
yo
(x2,y2)
(x1,y1)
X
xo
58
Example 1: Revisited
Denote the response distance as D1
D1 = |X1 – X2| + |Y1 – Y2|
To get E[D1], we can follow the 4-step approach (too
complicated with 4-d sample space!)
Alternatively, we can use the results we obtained
before
Here D1 can take on values between 0 and (Xo+Yo),
and since (Xo+Yo) is a constant, then
D1 = (Xo+Yo) Y
E[D1] = (Xo+Yo).1/3
59
of the example of unit length highway
Example 2: Ratio of Manhattan to
Airline Distance
Consider a grid street network (like Toronto’s)
We are interested in the following question
How much longer on average do we have to travel
from a random point (x1,y1) to a random point (x2,y2)
through the network compared to the straight line
distance (i.e. airline distance or Euclidean distance)
between the 2 points?
Stated differently, what is the expected penalty
incurred by travellers because of travelling through
the grid street network?
60
Example 2: Ratio of Manhattan to
Airline Distance
Solution
If R is the ratio of the 2 distances, we can use the 4-
step approach to show that
4 1
f R (r ) for 1 r 2
2r 2
61
Perturbation to Random Variables
In dealing with probabilistic problems in the
urban context, we often encounter complicated
settings which render the RV of interest hard to
study
One approach to address this problem is to
define the RV of interest as the sum of an “easy
to analyze” RV (i.e. removing complications)
plus one or more “perturbation” RVs
62
Perturbation to Random Variables
Example: Barrier to Travel
Consider the rectangular response area of dimensions
Xo and Yo for an Emergency Vehicle
If a barrier to travel (e.g. wall) of length “a” was
erected as shown, what is the expected travel distance
between the EV and a random incident?
Y
Yo
I a II
Xo X 63
