Location Theory II
Logistical and Transportation
Planning Methods
Single Tree Center Problem
G is a tree. e
i
, i=1,2,...,m
are the end vertices.
Single Tree Center Problem
e
N
Step 1: Choose any
and find the end vertex,
say e
N
, farthest from x.
x ∈G
G is a tree. e
i
, i=1,2,...,m
are the end vertices.
x
Single Tree Center Problem
e
N
Step 1: Choose any
and find the end vertex,
say e
N
, farthest from x.
x ∈G
G is a tree. e
i
, i=1,2,...,m
are the end vertices.
Step 2: Find the end vertex,
say e
S
, farthest from e
N
.
x
e
S
Single Tree Center Problem
e
N
x
Step 2: Find the end vertex,
say e
S
, farthest from e
N
.
Step 3: The absolute center
of G is the midpoint
of the path from e
N
to e
S
.
The vertex center is the
vertex closest to
that midpoint.
Step 1: Choose any
and find the end vertex,
say e
N
, farthest from x.
x ∈G
G is a tree. e
i
, i=1,2,...,m
are the end vertices.
e
S
Single Tree Center Problem
e
N
x
Step 2: Find the end vertex,
say e
S
, farthest from e
N
.
Step 3: The absolute center
of G is the midpoint
of the path from e
N
to e
S
.
The vertex center is the
vertex closest to
that midpoint.
Absolute center
Vertex Center
Step 1: Choose any
and find the end vertex,
say e
N
, farthest from x.
x ∈G
G is a tree. e
i
, i=1,2,...,m
are the end vertices.
e
S
Minimum Manhattan Distance
Paths in the Presence of Barriers
Larson, Li, Networks
One of n
Q
O-D points
One of n
B
polygonal barriers
One of n
Q
O-D points
One of n
B
polygonal barriers
Communication:
The name of the game!
? Two points having coordinates (x, y) and
(u, v) COMMUNICATE if there exists at
least one feasible staircase path between
them.
? Such a staircase path has minimum possible
Manhattan length:
? For example, any 2 adjacent vertices of a
barrier communicate
L[(x,y),(u,v)] =| x ? u | + | y ? v |
Vertex Seeking Tree
This tree is from a O-D point
+∞
Vertex Seeking Tree
This tree is from a barrier vertex
+∞
Vertex Seeking Tree
This tree is from a barrier vertex
+∞
Vertex Seeking Tree
This tree is from a barrier vertex
+∞
Vertex Seeking Tree
This tree is from a barrier vertex
Vertex Seeking Tree
This tree is from a barrier vertex
2 Nodes Communicating Simply if
? They are adjacent vertices of a barrier
? One is the root and the other is a terminal
node of a vertex-seeking tree
? If they they have x and y probes that share
at least one common non-nodal point b.
Theorem 1
? Suppose 2 points communicate but not
simply. Then there exists at least one
feasible staircase path between them
containing a sequence of simply
communicating nodes.
? Use 'path push and amalgamation' for proof.
Theorem 2
? Suppose 2 points do not communicate.
Then there exists at least one minimal
distance path between them containing a
sequence of simply communicating vertices.
? All this implies that we have reduced the
problem to a finite network problem and
can use Dijkstra, etc. for analysis.
Median Facility Locations with
Barriers and Manhattan Metric
Median Facility Locations with
Barriers and Manhattan Metric
1/5
1/5
1/5
1/51/5
Median Facility Locations with
Barriers and Manhattan Metric
1/5
1/5
1/5
1/51/5
1-median
Stochastic Queue Median
Stochastic Loss Median
Motivation
? Standard Hakimi deterministic median
? Lack of effects of uncertainty in time and
duration of service request
? Lack of queueing effects or lost customers
? Queueing delay can be far greater than
travel time
The Model
Single server queue garaged at
λh
i
= Poisson rate of service requests from
node i
Travel time = travel distance/speed = d/v
Time required to travel a fraction α of link
(i,j) is (α)d
ij
/v (d
ij
= length of link)
S
i
(x)=service time associated with node i
x ∈G
S
i
(x)=service time associated
with node i
S
i
(x) = βd(x,i)/ v + R
i
+ W
i
R
i
(β-1)d(x,i)/v
W
i
Travel time to the customer
On-scene service time
Follow-up travel time
Off-scene service time
Time
d(x,i)/v
Potential Location of a Service
Facility
x
a
b
x
l
ab
-x
Model 1:
Stochastic Loss Median
? When a customer requests service:
– Assign server to customer, if server is available
– Or, assign customer to a back-up system
(i.e., customer is lost)
Model 2:
Stochastic Queue Median
? Assume infinite capacity queue
? When a customer requests service
– Assign server to customer, if server is available
– Otherwise enter thhe customer into a FCFS
queue
Model 1:
Stochastic Loss Median
ρ(x)=server utilization factor, when server is
garaged at x
Expected cost of travel for a random service
request is
J(x) = (1 ? ρ(x)) h
i
i=1
n
∑
d(x,i)/v + ρ(x)γ
h
i
= fraction of service requests from node i
γ = cost of rejection > 0
Problem: Stochastic Loss Median
Find such that
x* ∈G
J(x*) ≤ J(x) for all x ∈G
Lemma: There exists at least one node of G
that is a stochastic loss median and that node
corresponds to the Hakimi median.
Proof requires that we show that is monotonically
increasing with mean travel time (or distance)
J(x)
Model 2:
Stochastic Queue Median
T
R
(x) ≡ expected response time = W
q
(x) + t(x)
where
W
q
(x) = mean in - queue delay
t(x) = mean travel time
Stochastic Queue Median:
Find x* ∈[a,b], [a,b] ∈L,such that
T
R
(x*) ≤ T
R
(x) for all x ∈(a',b'), (a',b')∈L.
Problem: we have breakpoints within links, as we have
already discussed in class. A breakpoint is a point that is
equidistant paths to a node, exiting the link on either end.
It is the point at which an optimal path to that node changes.
Algorithm
? Can prove convexity of the objective
function between breakpoints
? Implies finite step algorithm for finding x*
x*(λ): Trajectory of the Optimal
Location as a Function of λ
? When λ = 0+, x* is an Hakimi median of G
? When λ approaches λ
max
-, then x* is again
an Hakimi median of G
? What happens in between extreme values of
λ?