Ant Colonies As Logistic Processes Optimizers

34
Ant Colonies As Logistic Processes Optimizers

description

Ant Colonies As Logistic Processes Optimizers. Outline. Abstract Introduction The Logistic Process Scheduling Using Ant Colony Stimulation Results and Analysis Real work Example Conclusion References. Abstract. The optimization of logistic processes using ant colonies. - PowerPoint PPT Presentation

Transcript of Ant Colonies As Logistic Processes Optimizers

Page 1: Ant Colonies As Logistic Processes Optimizers

Ant Colonies As Logistic Processes Optimizers

Page 2: Ant Colonies As Logistic Processes Optimizers

Outline• Abstract• Introduction• The Logistic Process• Scheduling Using Ant Colony• Stimulation Results and Analysis• Real work Example• Conclusion• References

Page 3: Ant Colonies As Logistic Processes Optimizers

Abstract

The optimization of logistic processes using ant colonies.

The analysis of the algorithm parameters is done in a simulation.

It was applied to a real logistic process at Fujitsu-Siemens Computers.

The results show that the ant colonies provide a better solution to logistic processes.

Page 4: Ant Colonies As Logistic Processes Optimizers

Introduction

• What is Logistics? – Planning, handling, and control of the

storage of goods between the manufacturing point and the consumption point.

• cross-docking centers instead of stocks.•  The key issue is to deliver the goods in time by

minimizing the stocks. •  The scheduling algorithm has to decide which

goods are delivered to which customers.•  Centralized static scheduling vs. dynamic

distributed scheduling

Page 5: Ant Colonies As Logistic Processes Optimizers

The Logistic Process

Fig. 1. General representation of the logistic process

Page 6: Ant Colonies As Logistic Processes Optimizers

The Logistic Process

The birth process (arrival of new orders). Poison distribution of the birth process:

x : the random variable number of orders

lambdaT : the probability of this event occur on a certain time T.

Page 7: Ant Colonies As Logistic Processes Optimizers

The Logistic Process

The death process (delivery of orders) is modeled by the exponential distribution.

T: the random variable μ: the death rate.

Page 8: Ant Colonies As Logistic Processes Optimizers

Process Description

• Order arrival. - order is a set of components ci

and contain a desired delivery date• Component request.

- Each component has quantity.• Component arrival.

- supplier delay: time to be delivered to the logistic system.

Page 9: Ant Colonies As Logistic Processes Optimizers

Process Description

• Component assignment. The focus of this paper

-A component stock contains the available components and their quantity.

-A order stock is waiting list• Order delivery

-delay d is difference between the delivery date and the desired date.

Page 10: Ant Colonies As Logistic Processes Optimizers

Scheduling Policies

• Pre-assignment vs. dynamic decentralized approach

Pre-assignment (p.a.). Components are assigned to specific orders. Not efficiently

Distributed approach. The agents associated with orders and components exchange information between each other. More flexible than pre-assignment

Page 11: Ant Colonies As Logistic Processes Optimizers

Scheduling Using Ant Colonies

• The optimization of the scheduling process is a NP-hard problem. 

• The problem’s information can be translated into the pheromones, and used by all the interacting agents in order to achieve a good global solution.

Page 12: Ant Colonies As Logistic Processes Optimizers

Scheduling Using Ant Colonies

• Two different set of entities:component ≈ food source order ≈ nest.• m ants, one per food source, distribute

the food to the n nests. • In every iteration t of the algorithm, the

ants have to choose with some probability p which is the nest to visit first. Then, they deposit a pheromone τ in the path from the food source to the nest.

Page 13: Ant Colonies As Logistic Processes Optimizers

Scheduling Using Ant Colonies

• Each ant delivers an amount qij

from the total amount qi of component i {1, . . . , m} to an order j {1, . . . , n}. Since there are several nests to visit, the ant k chooses the path to a particular nest with a probability p

Page 14: Ant Colonies As Logistic Processes Optimizers

Scheduling Using Ant Colonies

ij is the amount of pheromone connecting i to j,

nij is a visibility function

Tk is the tabu list of the kth ant. This list contains all of the visited orders + the orders don’t need component type.

α and β express the relative importance of trail pheromone (experience) with visibility (knowledge)

Page 15: Ant Colonies As Logistic Processes Optimizers

Scheduling Using Ant Colonies

• dj is the delay of order j • We try to have nij = 1.• The local update of the pheromone

concentration is then given by

where c is small constant.

Page 16: Ant Colonies As Logistic Processes Optimizers

Scheduling Using Ant Colonies

• At the end of a complete tour the change of pheromones in all paths is given by

the solution can be evaluated using a performance measure

n is the number of orders and dj is the delay of order j.

Page 17: Ant Colonies As Logistic Processes Optimizers

Scheduling Using Ant Colonies

• Fig. 2. Example of an ant colony applied to the logistic process with pheromone concentration level on the trails: High (—), Medium (- -) and Low (··)

Page 18: Ant Colonies As Logistic Processes Optimizers

Scheduling Using Ant Colonies

• At each tour N of the algorithm (where each tour has n × m iterations t), a z is computed and stored in the set Z = {z(1), · · · , z(N)}

• If z(N) is higher than the previous z Z, then the actual solution has improved and the used pheromones should be increased. If it is worse, they should be reduced.

Page 19: Ant Colonies As Logistic Processes Optimizers

Scheduling Using Ant Colonies

• This is done by the global pheromone update

Page 20: Ant Colonies As Logistic Processes Optimizers

Algorithm

Fig. 3. Ant colonies optimization algorithm for logistic processes

Page 21: Ant Colonies As Logistic Processes Optimizers

Simulation Results

• Let lambdaT = 10;• each order can have at the most 7 different

types of components ci; • the quantity for each component varies

randomly between 1 and 20;• each type of component has a constant supplier

delay, which are 1, 3, 2, 3, 1, 2, 6 days for components type c1, · · · , c7 respectively.

• For each order a desired date is generated using an exponential distribution with μ = 7.

• The simulation refers to an interval of 6 months.

Page 22: Ant Colonies As Logistic Processes Optimizers

Simulation Results

• The results are presented with the parameters α = 1, β = 10, p = 0.9 and Nmax = 20.

Table 1. Comparison between the scheduling methods in number of orders 

Page 23: Ant Colonies As Logistic Processes Optimizers

Simulation Results

Fig. 4. Histograms of the order delay d. Pre-assignment method (left) vs. the results for the ants (right).  

Ants:higher number of delivery on time (d=0)lower spread between max and min delay

Page 24: Ant Colonies As Logistic Processes Optimizers

Tuning the Parameters

The parameters α and β which are coupled between each other are changed at the same time, while others decoupled parameters remain constant.

Varying α and β, using a fixed value p = 0.9.

Page 25: Ant Colonies As Logistic Processes Optimizers

Tuning the Parameters

Fig. 5. Number of orders delivered for a fixed ρ and varying α and β.

Page 26: Ant Colonies As Logistic Processes Optimizers

Tuning the Parameters

• Number of orders delivered on the correct date is high, if β is small.

• It has an optimal value for α = 1 • We can conclude that the α parameter tunes

the number of orders in the right day• and β controls the spread around that value

Page 27: Ant Colonies As Logistic Processes Optimizers

Tuning the Parameters

Evaporation coefficient (1-p)  p 0: the increment received by the new ants it

will influence greatly the paths of the next antsp 1 the solution can rapidly stagnate.  As it can be seen, the value of evaporation

should be around 0.1 (p = 0.9), in order to achieve a good solution

Fig. 6. Number of orders in the right day, for different sets of fixed α and β and varying p 

Page 28: Ant Colonies As Logistic Processes Optimizers

Tuning the ParametersNumber of colonies per day Nmax.

Fig. 7. Evolution of the solution for different number of coloniestoo few : not have enough iterations to find a

good solution

too many: increased severely the computational cost.

Page 29: Ant Colonies As Logistic Processes Optimizers

Real World Example

The analysis presented an optimized solution with α = 1, β = 0.5, p = 0.9, and Nmax = 20.  

In small data set of the data

Fig. 8. Histograms of the orders delay d for the scheduling methods

Page 30: Ant Colonies As Logistic Processes Optimizers

Real World Example

Ants: More orders are delivered on time. Less orders are delayed

Good alternative to pre-assignment scheduling method !

Page 31: Ant Colonies As Logistic Processes Optimizers

Conclusions

• To apply in the ant colonies optimization algorithm to the optimization of logistic processes.

• It’s explored the correlations between the parameters and their role in the algorithm.

• The results show how the analysis is able to improve the algorithm performance, and explain the reasons for that improvement.

•  Finally, the algorithm was applied to a real data set, and the ant algorithm proved to be a better scheduling method than the pre-assignment. 

Page 32: Ant Colonies As Logistic Processes Optimizers

Conclusions

• Future work: – The use of a different cost function z, – The incorporation in the ants with some sort

of prediction mechanism. – There is still some information in the

process, like priority of the orders, that was not used so far and can influence the scheduling of the logistic process.

Page 33: Ant Colonies As Logistic Processes Optimizers

References 1. Jayashankar M. Swaminathan, S.F.S., Sadeh, N.M.:

Modeling supply chain dynamics: Amultiagent approach. Decision Sciences Journal 29 (1998)

607–6322. McKay, K., Pinedo, M., Webster, S.: A practice-focused

agenda for production schedulingresearch. Production and Operations Management 10

(2001)3. Palm, R., Runkler, T.: Multi-agent control of queuing

processes. In: To appear in ProceedingsofWorld Conference on Automatic Controlo - IFAC’2002,

Barcelona, Spain. (2002)4. Silva, C.A., Runkler, T., Sousa, J.M., Palm, R.:

Optimization of logistic processes using ant

Page 34: Ant Colonies As Logistic Processes Optimizers

The End

• Question?

• Comment?