The Travelling Salesman Problem

A travelling salesman living in Chicago must make stops in these 4 other cities: LA, Denver,
Boston, and Dallas. He must start and finish in his home city of Chicago. He must select the
order of customers to visit that will minimize the total length of the trip.

Below is the specific information about the distances between each of the 5 cities:

(Click Image To See an

Enlarged Version)

Solver Problem Solving Steps

Step 1 – Determine the Objective

In this case, the objective is to minimize the total distance traveled when traveling between all 5
cities. The total distance traveled by the salesman is the Objective to be minimized.

Step 2 – Determine the Decision Variables

We are trying to select the order of cities to visit that minimizes the total distance travelled. The
cities are designated in the Excel model not by their names but by the row that they appear in the
distance chart just shown.

Boston appears in the 1st row of the distance chart. Boston is therefore designated with a “1.”
Chicago appears in the 2nd row in the chart and is therefore assigned a designation of “2.” Dallas
appears in the 3rd row and is designated “3.” Denver appears in the 4th row and is designated “4.”
LA appears in the 5th row and is designated “5.”

We need to determine the order of cities to visit to minimize the total miles travelled. In other
words, we are minimizing the sum of distances travelled between consecutive cities.

Step 3 – Build the Excel Equations That Combine the Objective With
All Decision Variables
 (Click Image To See an
Enlarged Version)

Column C – Decision Variables

The Decision Variables are the arrangement of the cities to visit. In the preceding diagram, the
Decision Variables are the row designations of each city. These Decision Variables are in cells C10
to C14. The order of the Decision Variables shown above (1, 2, 3, 4, 5) indicate that the cities will
be visited in this order: Boston (row 1 in the Distance Chart) to Chicago (row 2 in the Distance
Chart) to Dallas (row 3 in the Distance Chart) to Denver (row 4 in the Distance Chart) to LA (row 5
in the Distance Chart) and back to Boston.

This set of 5 Decision Variables are collectively subject to the Alldifferent Constraint. As a result,
each 1 of these 5 Decision Variables will be assigned an integer between 1 and 5. None of the 5
Decision Variables in this set can be assigned the same number.

Column D – Listing The Cities Attached To Decision Variables

The Excel Index function is used in column D to list the city which corresponds to the Distance
Chart row number that appears in column C. For example, cell D10 contains the Excel formula:

=INDEX($B$3:$B$7,$C10,1)

The Index function has the following syntax:

=INDEX(range, row number, column number)

For the above formula, the range is the cells from B3 to B7. This is a range of cells that has only 1
column.

The row number referenced in this Index function is in cell C10. The contents of Cell C10 = 1. The
row referenced by this Index function is the 1st row.

The column number referenced in this Index function is 1. This would have to be the case since
only 1 column exists within the given range (B3 to B7).
The Index function displays the contents of the cell in the given row (1strow) and column
(1st column) of the given range. This Index function will display the contents of the cell in the
1st row and the 1st column of the given range (B3 to B7). The cell in the 1st row and the 1st column
of cell range (B3 to B7) is cell B3. Cell B3 contains the word Boston, which is displayed in cell D10.

Column E – Calculating Distances Between Each City and the

Previously Visited City

The distances between each city and the previous city visited are shown to the right in column E.
For example, the distance between Boston and the previous city of LA is 3036 miles. The distance
between Chicago and the previous city of Boston is 983 miles. The distance between Dallas and the
previous city of Chicago is 1205 miles. The distance between Denver and the previous city of Dallas
is 801 miles. The distance between LA and the previous city of Denver is 1174.

Distances between each city and its previous city are found by using the Index function. This
function looks up in the Distances chart and locates and displays the distance between a city and
its previously visited city.

The distances shown in Column E are as follows:

The distance between Boston and the previous city of LA is 3036 miles.

The distance between Chicago and the previous city of Boston is 983 miles.

The distance between Dallas and the previous city of Chicago is 1205 miles.

The distance between Denver and the previous city of Dallas is 801 miles.

The distance between LA and the previous city of Denver is 1174.

The Excel Index functions which generated each of these distances are shown to the right of each
distance. An explanation of this use of the Index function is as follows:

The Index function has the following syntax:

=INDEX(range, row number, column number)

For the above formula, the range is the cells from C3 to G7. This cell range holds the distances in
the Distance Chart.

The row number corresponds to the previous city visited. This row number is the previous city’s
row number in the Distance chart.

The column number corresponds to the current city. This column number is the current city’s
column number in the Distance chart.

All of the distances between cities are listed in cells C3 to G7. The Index function locates the cell
that holds the distance between each city and the previous city. Each distance will be found in the
row number of the that previous city and column number of the current city.

The order of the cities (actually, the order of the city row numbers in column C) are arranged so
that sum of the distances between each city and the previous city is minimized.
Step 4 – List All Constraints

This problem provides an excellent opportunity to showcase the Alldifferent Constraint. We are
visiting each city only once so we need each city (actually the city’s row number in the Distance
Chart) to be listed only one time without repeating.

Each city’s unique row number in the Distance Chart must be assigned to only 1 of the 5 Decision
Variable cells. We must therefore apply the Alldifferent Constraint to all of the Decision Variable
cells (cells C10 to C14) simultaneously as a group. As a result of the Alldifferent Constraint, these
5 cells will hold the integers 1 to 5. No 2 cells in this group will be assigned the same number. This
ensures that each city will be visited only once and that all cities will be visited.

The following Solver Dialogue box shows the set of 5 Decision Variables (Cells C10 to C14) subject
collectively to the Alldifferent Constraint.

(Click Image To See an

Enlarged Version)

Step 5 – Test the Excel Spreadsheet

This spreadsheet can be very easily tested by varying the integer values in the Decision Variable
cells (cells C10 to C14). The city names and distances between cities should correctly change to
match the new Decision Variable integer values corresponding to different rows.
Step 6 – Insert All Data into the
Solver Dialogue Box

Once again here is the completed Solver dialogue box:

(Click Image To See an

Enlarged Version)

The Travelling Salesman Problem provides an excellent opportunity to demonstrate the use of the
Evolutionary method. The Evolutionary method must be used if the Mathematical Path to the
Objective contains any cells holding non-smooth or discontinuous formulas.

Common non-smooth Excel functions are MIN, MAX, and ABS.

Common discontinuous Excel functions are INDEX, HLOOKUP, VLOOKUP, LOOKUP, INT, ROUND,
COUNT, CEILING, FLOOR, IF, CHOOSE, NOT AND, OR, GREATER THAN, LESS THAN, and EQUAL TO.

The INDEX function in this problem appears in cells that are part of the Clear Mathematical Path to
the Objective. We therefore must select the Evolutionary method to solve this problem.

After we have selected the Evolutionary method, we hit Solve and solution shown in several pages
is reached.

This solution could be interpreted as follows:

The salesman stars in his home town of Chicago. He then visits Denver, LA, Dallas, Boston, and
finally back to Chicago in that order. The total miles travelled on this route are 6,447 miles. This is
the shortest route that will cover all 5 cities starting and ending in Chicago.

Note that no special provision has to be made to ensure that the starting point is Chicago.
Following are views of the Excel model before solving and then after solving with Excel Solver’s
Evolutionary method:

Before Running Solver

(Click Image To See an

Enlarged Version)

After Running Solver

(Click Image To See an

Enlarged Version)

Solver Answer Report

Part 1

Note:

- The Solver Result

- How long Solver took to solve the problem (especially important in this case – This time could
be reduced by several of the Options settings.)

- The Solver Engine that was used and the Solver Options settings

- Where the Objective Cell was labeled in the Excel model for its name to appear as it does in Part
1 of the Answer Report

(Click Image To See an Enlarged Version)

Note the solution time of 164 seconds. This could have been reduced by limiting the maximum
allowable run time, iterations, or subproblems using the Options menu.

Part 2 – Variable Cells

- Note that the Variable Cells contain the Decision Variables

- Note where the labels for each Decision Variable are placed in the Excel model so that the
Decision Variable’s name will appear here in Part 2 of the Answer Report as it does

- Note the type of variable - Either Continuous or Integer (Integer, Binary, or Alldifferent)

- Note the Before and After values of each Decision Variable

Part 3 - Constraints

- Note how each Constraint is labeled in the Excel model in order for the Constraint’s name to
appear here in Part 3 of the Answer Report as it does.
- Note which Constraints are binding (had their limits hit) and which aren’t.

- Note how much slack is still available in any Constraint that has not had its limit hit.

- Note any Integer Constraints (Integer, Binary, Alldifferent)

(Click Image To See an Enlarged Version)

Note that Cells C10 to C14 were set to AllDiff simultaneously as a group.

Solver Population Report

The Population Report is made available when the Evolutionary method is used. The Population
Report provides indication about whether and how you can make improvements to the model or to
the Evolutionary method Options.

The Population Report can be valuable if successive runs of the Evolutionary method produce
different answers for the Objective. The setting of the Mean Value and Standard Deviation during
successive runs provide insight into whether your successive solutions are getting closer to the
most optimal solution. See the section on a detail description of this report for more information
about this topic.

(Click Image To See an

Enlarged Version)

Solución del Problema del Vendedor Viajero

por GEO Tutoriales el 10/03/2015 en Programación de Trabajos, Programación Entera 0

El Problema del Vendedor Viajero(conocido también como Travelling

Salesman Problem o simplemente TSP) consiste en encontrar
el circuito óptimo (en términos del viaje más corto) que deberá seguir un
vendedor en un caso con n ciudades, en el que cada ciudad se visita
exactamente una vez. Básicamente es una adaptación del Problema de
Asignación que considera restricciones adicionales que garantiza la exclusión
de subcircuitos en la solución óptima.
Específicamente en el caso de n ciudades se define las variables de decisión de
la siguiente forma:
PUBLICIDAD

Sea la distancia de la ciudad i a la ciudad j, donde , el modelo del

agente o vendedor viajero corresponde a:

El conjunto de restricciones (1) y (2) definen un modelo de asignación

tradicional. Lamentablemente en general, el problema de asignación producirá
soluciones de subcircuito más que circuitos completos que abarque
las n ciudades.
Para ilustrar los conceptos de circuito y subcircuito en el contexto del
Problema del Vendedor Viajero, consideremos un agente de venta que vive en
la ciudad 1. Miami (Florida) en Estados Unidos y debe visitar a importantes
clientes en las siguientes ciudades: 2. Chicago (Illinois), 3. Houston (Texas), 4.
Las Vegas (Nevada) y 5. San Francisco (California). Para mayor claridad se han
destacado los estados mencionados anteriormente con un color distintivo.
Un circuito factible sería viajar en el siguiente orden: Miami (FL), Chicago (IL),
Houston (TX), Las Vegas (NV), San Francisco (CA), Miami (FL). Es
decir, .
Por otra parte un subcircuito correspondería, por ejemplo, a Miami (FL), San
Francisco (CA), Las Vegas (NV), Miami (FL), junto a Houston (TX), Chicago (IL),
Houston (TX). Es decir, , lo que naturalmente
no es una solución factible para el problema que se busca resolver.
El modelo del vendedor viajero se caracteriza por su versatilidad para
representar otros casos prácticos en optimización. Uno de ellos es el Problema
de Secuenciamiento de la Producción como el que se presenta a
continuación:
El programa de producción diaria de una empresa de pinturas incluye lotes de
color Blanco (B), Amarillo (A), Negro (N) y Rojo (R). Como la empresa
utiliza las mismas instalaciones en las cuatro clases de pintura, es necesario
hacer una limpieza entre los lotes. La siguiente tabla resume el tiempo de
limpieza, en minutos, donde al color de la fila sigue el color de la columna. Por
ejemplo, cuando después de la pintura Blanca sigue la Amarilla, el tiempo de
limpieza en 10 minutos. Como un color no puede seguir a sí mismo, a los
elementos correspondientes se les asigna un tiempo de setup infinito. Se desea
determinar la secuencia óptima para la producción diaria de los cuatro colores,
que minimice el tiempo total de limpieza necesario.
Se puede hacer una analogía con el problema del vendedor viajero, asumiendo
que cada pintura es una “ciudad” y que las “distancias” representan el
tiempo de limpieza necesario para cambiar de un lote de pintura al siguiente.
En consecuencia, el problema se reduce a determinar el circuito más corto que
se inicie en un lote de pintura y pase exactamente una vez por cada uno de los
tres lotes restantes, para regresar al punto de partida.
En este contexto dada la cantidad de pinturas, la secuencia óptima se puede
encontrar por enumeración exhaustiva de los 6 circuitos posibles , es
decir, . En el ejemplo dicha secuencia óptima corresponde a
Blanco, Amarillo, Rojo, Negro, Blanco, con un tiempo total de setup de 98
minutos. Naturalmente esta estrategia no es eficiente y queda limitada a
problemas muy pequeños.

Alternativamente se puede utilizar implementar en Solver el modelo de

asignación presentado anteriormente, haciendo uso de los parámetros descritos
en el ejemplo del secuenciamiento de la producción de pinturas. A continuación
un extracto de los resultados donde se observa que no se alcanza una solución
de circuito.
En la actualidad existen programas computacionales que permiten enfrentar
estas dificultades que establece el problema del vendedor viajero. Uno de ellos
es el software TSPSG (TSP Solver and Generator) que se caracteriza por una
interfaz intuitiva y que a continuación se detalla la implementación de nuestro
problema (recordar que la Ciudad 1 correspondería al color Blanco, y así
sucesivamente).
Una vez ingresado los datos al seleccionar “Solve” el programa se ejecuta
entregando los resultados alcanzados que por cierto coincide con aquellos que
identificamos por enumeración.
En términos algorítmicos los métodos disponibles para resolver el problema del
agente o vendedor viajero tienen su base en las ideas de los algoritmos
generales de ramificación y acotamiento (Branch and Bound) o de Plano de
Corte, en los cuales abordaremos en próximos artículos.