1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

57
1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005

Transcript of 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

Page 1: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

11

1-to-Many DistributionVehicle Routing

John H. Vande Vate

Spring, 2005

Page 2: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

22

Shared Transportation Capacity

• Large shipments reduce transportation costs but increase inventory costs

• EOQ trades off these two costs

• Reduce shipment size without increasing transportation costs?

• Combine shipments on one vehicle

Page 3: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

33

TL vs LTL

Inventory

Transport

Transport

Inventory

Page 4: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

44

Shared Loads

Inventory

Transport

Transport?

Inventory

?

Page 5: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

55

Issues

• Design Routes that – Minimize the transportation cost– Respect the capacity of the vehicle

• This may require several routes

– Consider inventory holding costs• This may require more frequent visits

Page 6: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

66

Our Approach

• Minimize Transportation Cost (Distance)– Traveling Salesman Problem

• Respect the capacity of the Vehicle– Multiple Traveling Salesmen

• Consider Inventory Costs– Estimate the Transportation Cost– Estimate the Inventory Cost– Trade off these two costs.

Page 7: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

77

The Traveling Salesman Problem

• Minimize Distance

• s.t. – start at the depot, – visit each customer exactly once,– return to the depot

• A single vehicle no capacity

• Only issue is distance.

Page 8: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

88

IP Formulation• Set Cities;

• param d{Cities, Cities};

• var x{Cities, Cities} binary;

• minimize Distance:sum{f in Cities, t in Cities}d[f,t]x[f,t];

• s.t. DepartEachCity{f in Cities}:sum{t in Cities}x[f,t] = 1;

• s.t. ArriveEachCity{t in Cities}:sum{f in Cities}x[f,t] = 1;

Page 9: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

99

SubTours

3 cities

3 edges

Page 10: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

1010

Eliminating Subtours

• S.t. SubTourElimination

{S subset Cities:

card(S) > 0 and

card(S) < card(Cities)}:

sum{f in S, t in S} x[f,t] <= card(S) - 1;

Page 11: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

1111

An Equivalent Statement

3 cities

No edges out

Page 12: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

1212

An Equivalent Formulation

• S.t. SubTourElimination

{S subset Cities:

card(S) > 0 and

card(S) < card(Cities)}:

sum{f in S, t not in S} x[f,t] >= 1;

Page 13: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

1313

How Many Constraints?

• How many subsets of N items?

• 2N

• Omit 2 subsets:– All N items– The empty set

• 2N - 2

Page 14: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

1414

OK, Half of that...

• If we have an edge out of S, we must have an edge out of N\S.

• Why? The edge out of S is an edge into N\S. But there are exactly |N\S| edges into cities in N\S. Since one of them comes from S, not all the edges from cities in N\S can lead to cities in N\S. At least one must go to S.

Page 15: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

1515

Still Lots!

• How many subsets of N items?

• 2N

• Omit 2 subsets:– All N items– The empty set

• 2N - 2

• Half of that: 2N-1 - 1

Page 16: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

1616

How Many?

10 511100 6.34E+29101 1.27E+30

1000 5.4E+300

N 2N-1 - 1

Too Many!

Page 17: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

1717

Optimization is Possible But...

• It is difficult

• Few 100 cities is the limit

• For more details see www.tsp.gatech.edu

• Is it appropriate?

• Other approaches….

Page 18: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

1818

Heuristics• The Strip Heuristic

– Partition the region into narrow strips– Routing in each strip is easy ~ 1-Dimensional– Paste the routes together

Page 19: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

1919

The Strip Heuristic

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x

Page 20: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

2020

Nearest Neighbor

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x

Page 21: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

2121

Clark-Wright

• Shortcut a “tour” by finding the greatest “savings”

x

x

x

x

x

x

Page 22: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

2222

Clark-Wright

• Shortcut a “tour” by finding the greatest “savings”

x

x

x

x

x

x

Page 23: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

2323

Nearest Insertion

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x

Page 24: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

2424

Nearest Insertion

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x

Page 25: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

2525

Nearest Insertion

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x

Page 26: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

2626

Nearest Insertion

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x

Page 27: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

2727

Nearest Insertion

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x

Page 28: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

2828

Nearest Insertion

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x

Page 29: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

2929

Nearest Insertion

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x

Page 30: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

3030

Nearest Insertion

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x

Page 31: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

3131

Nearest Insertion

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x

Page 32: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

3232

Nearest Insertion

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x

Page 33: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

3333

Nearest Insertion

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x

Page 34: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

3434

Nearest Insertion

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x

Page 35: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

3535

Nearest Insertion

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x

Page 36: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

3636

Nearest Insertion

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x

Page 37: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

3737

Nearest Insertion

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x

Page 38: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

3838

Improvement Heuristics

• 2-Opt

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x

Page 39: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

3939

Local Minima

• No improvement found, but…

• Tour still isn’t good

Page 40: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

4040

Probabilistic Methods

• Simulated Annealing

• With probability that reduces over time, accept an exchange that makes things worse (gets you out of local minima).

Page 41: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

4141

Optimization-Base Heuristics

• Minimum Spanning Tree Heuristic

• Build a minimum spanning tree on the edges between customers

• Double the tree to get a Eulerian Tour (visits everyone perhaps several times and returns to the start)

• Short cut the Eulerian Tour to get a Hamilton Tour (Traveling Salesman Tour)

Page 42: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

4242

The Spanning Tree

• Is Easy to construct– Use the Greedy Algorithm

• Add edges in increasing order of length

• Discard any that create a cycle

• Is a Lower bound on the TSP– Drop one edge from the TSP and you have a

spanning tree– It must be at least as long as the minimum

spanning tree

Page 43: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

4343

The Spanning Tree

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x

Page 44: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

4444

Double the Spanning Tree

• Duplicate each edge in the Spanning Tree

• The resulting graph is connected

• The degree at every node must be even

• That’s an Eulerian Graph (you can start at a city, walk on each edge exactly once and return to where you started)

• It’s no more than twice the length of the shortest TSP

Page 45: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

4545

The Spanning Tree

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x1

2

3

45 6

7

8

910

111213

1415

1617

1819

20

21

22

23

2425

26 27

28

29

30

Page 46: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

4646

Short Cut the Eulerian Tour

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x1

2

3

45 6

7

8

910

111213

1415

1617

1819

20

21

22

23

2425

26 27

28

29

30

Page 47: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

4747

Short Cut the Eulerian TourShort Cut the Eulerian Tour

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x1

2

3

45 6

7

8

910

111213

1415

1617

1819

20

21

22

23

2425

26 27

28

29

30

Page 48: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

4848

Spacefilling Curves

• There are no more points in the unit square than in the interval from 0 to 1!?

Page 49: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

4949

“Proof”• Each point (X,Y) on the map• For illustration let’s consider points in [0, 32]2

• Express X = string of 0’s and 1’s, 5 before the decimal– X = 16.5 = 10000.10 1*24+0*23+0*22+0*21+0*20+1*2-1 +0*2-2

• Express Y = string of 0’s and 1’s5 before the decimal– Y = 9.75 = 01001.11 0*24+1*23+0*22+0*21+1*20+1*2-1 +1*2-2

• Space Filling Number - interleave bits and move the decimal (X,Y) = 10010.000011101

Page 50: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

5050

So,...

• Each pair of points

X = 16.5 = 10000.10

Y = 9.75 = 01001.11

maps to a unique point

(X,Y) = 10010.000011101

Page 51: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

5151

How to Use this?

• A mapping of (X,Y) into the unit interval, i.e.• 18.056640625000 = 10010.000011101

X = 16.5 = 10000.10 Y = 9.75 = 01001.11

• Think of this as the inverse mapping of the unit interval onto the square (our super tour)

• For a given customer (X,Y) is the fraction of the way along the super tour where it lies

• Visit the customers in the order of (X,Y) (short cut the super tour to visit our customers)

Page 52: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

5252

The TSP

• For More on SpaceFilling Curves visithttp://www.isye.gatech.edu/faculty/John_Bartholdi/mow/mow.html

• There are several books on the TSP

• ….

Page 53: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

5353

Our Approach

• Minimize Transportation Cost (Distance)– Traveling Salesman Problem

• Respect the capacity of the Vehicle– Multiple Traveling Salesmen

• Consider Inventory Costs– Estimate the Transportation Cost– Estimate the Inventory Cost– Trade off these two costs.

Page 54: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

5454

Different Approaches

• Route First - Cluster Second– Build a TSP tour– Partition it to meet capacity

• Cluster First - Route Second– Decide who gets served by each route– Then build the routes

Page 55: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

5555

Route First

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x

6

4

55

5

5

3

26

4 6

3

352

Vehicle Cap: 15

Page 56: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

5656

Cluster First• Sweep Heuristic

x

x

x

x

x

xx

x

x

x

x

x x

x

x

x

6

4

55

5

5

3

26

4 6

3

352

Vehicle Cap: 15

Page 57: 1 1 1-to-Many Distribution Vehicle Routing John H. Vande Vate Spring, 2005.

5757

Our Approach

• Minimize Transportation Cost (Distance)– Traveling Salesman Problem

• Respect the capacity of the Vehicle– Multiple Traveling Salesmen

• Consider Inventory Costs– Estimate the Transportation Cost– Estimate the Inventory Cost– Trade off these two costs.