You are on page 1of 6

Space Allocation using

Intelligent Optimization Techniques


Rafael E. García Q. Christian G. Quintero M., Ph. D.
Department of Electrical and Electronics Engineering Department of Electrical and Electronics Engineering
Universidad del Norte Universidad del Norte
Barranquilla, Colombia Barranquilla, Colombia
quesadar@uninorte.edu.co christianq@uninorte.edu.co

Abstract— The space allocation problem has taken an chosen due to it is a suitable optimization technique proper
important role in different fields. For instance, placing items in to satisfy the restrictions of the objects to organize in
a warehouse where it is essential to take advantage of the complex environments [16], [12], [7]. On the other hand, the
available space and meet the production requirements.
Similarly, the parking slot allocation for automobiles in a car HC technique was also selected because it is an intelligent
parking in which there are occasions where a lack of proper optimization technique with good results in several
distribution of the automobiles exist. In the same sense, the exploration and search problems [8].
space allocation for crops in a land for cultivation in which it is
important to take into account factors and features such as II. PROPOSED APPROACH
humidity and pH. In this paper, an intelligent organizer of Basically, it is required a system able to search in the
objects is presented using Genetic Algorithms (GA) and Hill
Climbing (HC) to solve the space allocation problem in a
available space a suitable place for each object to organize
Warehouse, a Car Parking and a Land for Cultivation. Also, a based on the features of the objects and of the available
technique that organizes objects randomly in the space was space. The system must provide each final object´s location
implemented to compare results with the intelligent techniques. and suitable metrics that measure the satisfaction level of
Several tests were performed to check the proper system each organization. The available environments where the
operation and then performance tests under different system can work are a Warehouse, a Car Parking and a Land
conditions are shown comparing the results between the
intelligent techniques and the random technique. Finally,
for Cultivation.
advantages and disadvantages of intelligent techniques to solve For each environment have been selected some features or
the space allocation problem are presented. restrictions to satisfy. For instance, in the Warehouse
environment case the selected features or restrictions to
Keywords - Computational intelligence; Genetic Algorithms; satisfy are: “Weight”, “Temperature”, “Frequency”,
Optimization methods; Gray codes; Logistics “Access” and “Distance”. The “Weight” parameter is
referred to the difference between the object`s weight and
I. INTRODUCTION the weight that a determined region of the available space

T HE space allocation problem has taken an important role


in different fields [9], [10]. For instance, the location of
items in a warehouse where it is essential to take advantage
can hold. In the same way, the “Temperature” refers to the
difference between the ideal object`s temperature and the
space`s temperature. “Frequency” is the difference between
of the available space and meet the production requirements the number of times per week that the object´s conditions
[1], [2], [3], [16]. Similarly, the parking slot allocation for have to be checked and the number of times per week that is
automobiles in a car parking in which there are occasions desired to check objects’ conditions in a determined place.
where a lack of proper distribution of the automobiles exist. “Access” is related to the difference between the number of
[4], [5]. In the same sense, the space allocation for crops in a times per week that an object is required and the number of
land for cultivation in which it is important to take into times per week that is desired to access a determined place.
account factors and features such as humidity and pH [6], An obstacle is a region in the space that cannot be occupied
[15]. Due to the importance of the optimization for the use by an object to organize. Thereby, the “Distance” feature
of these spaces, this paper presents a system that allows refers to the distance from a region of the available space in
simulating the organization processes in a warehouse, a car study in which can be placed an object, until the nearest
parking and a land for cultivation. The system allows to set obstacle. The “Distance” parameter was implemented in
several configurations for each environment and the objects order to take advantage of the available space on the
to organize. These configurations may vary on each test. In Warehouse and on the Land for Cultivation, so it is desirable
this sense, several organization techniques can be tested to minimize this “Distance”. The lower the “Distance” for
under different conditions, taking into account restrictions of each placed object, the higher the number of objects that can
the space and the objects. The intelligent technique GA was be placed.

This work was supported by the Universidad del Norte, Barranquilla,


Colombia.

978-1-4244-6742-6/10/$26.00 ©2010 IEEE


“Distance” feature is also taken into account for the case to perform the organization. These techniques differ in their
of Land for Cultivation. Instead, this concept does not have ability to find optima in the search space. In this work, the
significance in Car Parking environment because the search space can be the Warehouse’s available space, the
maximum number of cars that can be placed is given by the Car Parking’s available space or the Land for Cultivation’s
available parking places and this number cannot be available space [2]. GA has some advantages versus other
improved during the organization process. optimization techniques, for instance GA is less likely to get
trapped in a local optimum while looking for a global
optimal [7]. However, HC has shown important results in
some case studies [8] and also by its simplicity it is a good
point of reference to evaluate the performance of any other
intelligent technique such as GAs.
In Fig. 1 is shown how GA works. On the other hand, in
Fig. 2 is shown how HC works.
III. IMPLEMENTATION
The environment’s space (Warehouse, Car Parking or
Land for Cultivation) is represented by a matrix, as have
been done in similar case studies [12], so it is divided into
cells. There are also four corresponding matrices, each one
containing the information about an environment´s property.
For example, in the case of Warehouse´s environment, a
matrix contains the weight that each region in the space can
hold, another matrix contains the temperature of each region
on the space. In total, there are four property matrices and
one for the environment´s space. Clearly, all the matrices
have the same size. Besides, it is also desired to prioritize the
features or restrictions in order to assign more importance to
Figure 1. Flowchart of the proposed Genetic Algorithm. one parameter over the others in a determined organization.
The space is evaluated using the fitness function shown in
The three environments have different properties in terms (1), similar to the ones used in [13] and [15], where i and j
of available space. In Land for cultivation environment all are the coordinates of the region to evaluate. In fact, the cell
the space is available to place crops. In the Warehouse (i, j) is the top-left cell of the region to evaluate. Mk is the k-
environment there are some obstacles such as walls that th environment´s property matrix (e. g., in the case of the
make the available space discontinuous. Finally, in the Car Warehouse´s “Weight”, “Temperature”, “Frequency” and
parking environment just the parking slots are available “Access”), N is the number of rows that the object occupies,
making the space discrete. and M is the number of columns. Imk is the importance of the
k-th feature. Pk is the k-th property of the object to place. For
instance, if Im2 is the importance of the feature
“Temperature”, then P2 is the ideal “Temperature” of the
object to place and M2,n,m is the “Temperature” of the region
to evaluate.
The parameter Im5 is the importance of the “Distance”
restriction and D is the distance of the nearest obstacle to the
evaluated region.

4
1
f (i, j ) = Im5 × D + ∑ Imk × ∑ ∑ M ( k ,n,m) − Pk (1)
N +i M+ j

k =1 N ×M n=i m= j

Moreover, it is important to say that in the regions to


evaluate is always possible to place an object, since the
available regions were previously selected from the
environment´s space.
The fitness function is independent of the technique used,
so each intelligent technique (GA and HC) uses the fitness
Figure 2. Flowchart of the proposed Hill Climbing Algorithm. function of (1) in its operation.
This work proposed to implement at first GA and then HC The implemented GA uses the following parameters which
had good results in other fields of study shown in literature them (if the current cell does not meet the requirements or
[11]: Initial Population Size: 16 individuals, Encoding: Gray the neighbors of current cell have not a worst fitness than
Code, Selection technique: Roulete Weel using (2) to invert the current cell). This modification of HC resides in the
Fitness Function and (3) to Scale Fitness Function, Selection “Evaluate neighbor regions” block of Fig. 2.
Rate: 50%, Mating: Single Point Crossover Mutation Rate:
15%.
Equation (2) inverts the function so that the t-th individuals
with lower fitness value get a higher inverted fitness value V
than the ones which have higher fitness. Equation (3) is then
applied to scale the value V for the t-th individual, making
GA explores a wide region in the first iterations and then,
once the standard deviation σ is reduced, the system
concentrates in the best located regions [11].

ft
Vt = 1 − (2)
∑r =1 f r
16 Figure 3. Moving direction of HC space exploration.

IV. EXPERIMENTATION
The scaled fitness of the t-th individual is given by St in Several tests were done to establish initially if the system
each iteration. Therefore, best individuals have the higher is able to organize objects in the available space for a
probability to survive, so the best schemas have the higher Warehouse environment, a Car Parking environment and a
probability to be inherited than the worst ones. Land for Cultivation environment, and then to see how much
that organization satisfies the requirements of the objects
⎧ Vt − V and the environment.
⎪1 + ⇐ σ ≠ 0 ∧ V − Vt ≤ 1.8σ As examples in this paper are going only to be shown the
⎪ 2σ operation tests made for the Warehouse environment using
St = ⎨ 1⇐σ = 0 (3)
GA. The results of the operation test for HC in the
⎪ 0.1 ⇐ σ ≠ 0 ∧ V − Vt 〉1.8σ Warehouse environment and the Land for Cultivation
⎪ environment are very similar to the results of GA, but are

very poor in the Car Parking environment due to the
The coss-over consist in the combination of the Gray
discontinuity of the space, it places the cars randomly. The
Codes of each pair of individuals selected between them by
shown test just configures one property in the environment,
a random point, as it is shown in the example of Table I. The
the Warehouse Weight, the setup of this property is shown in
next step is to replace the 8 worst individuals with the new
Fig. 4 and the object´s configuration is shown in Table II.
offprings.
The importances of the features are shown in Table III. The
TABLE I. CROSS-OVER PROCEDURE. environment has some obstacles symbolized by blue empty
rectangles in Fig. 5. The results are shown in the Fig. 6 and
Gray Code A 1 1 0 1 1 1 Table IV.
Gray Code B 1 0 0 0 1 0
TABLE II. OBJECTS FEATURES FOR OPERATION TEST.

Offpring 1 1 0 0 1 0

In the HC case, it is important to note that HC can stall in a


determined region if it falls on a route with a constant
fitness. To avoid this problem, the system chooses a
direction in which the space is going to be explored. The
direction selection is done in the first iteration, the system
chooses the best neighbor and moves to that position, but for
the next iteration it is only going to explore tree cells in the
direction it has moved in the first iteration, e. g., as can be
seen in Fig. 3, the best is neighbor 5, so the system moves to
that cell, in the next iteration, neighbor 5 becomes the
current cell and the system only explores neighbors A, B and
C because these cells are in the direction of the first
movement and it is going to move to the best one between
Figure 4. Setup Warehouse “Weight”
TABLE III. FEATURE´S IMPORTANCES FOR OPERATION TEST.

Figure 5. Warehouse environment with obstacles


Figure 7. Environment´s Setup for Performance Test I
Table IV contains the metrics Efficience and Cost Mean
which are the ratio between the placed objects and the A. Test I
objects to place, and the mean of the evaluation function The following setup was used to develop the first test:
shown in (1) for all the organized objects respectively.
In the test developed the GA technique could place the • The features of the environment are defined in Fig.
100% of the objects, it can be seen in the “Efficience” in 7.
Table IV. By the way, the Cost Mean was 0, the reason is • The importances of the features were kept constant
that the selected spaces meets completely the objects and equal to 1as it is shown in Table VI.
restrictions.
• For the setup of the objects, the feature values
varied in a pseudorandom way as is shown in Table
V.

TABLE V. OBJECTS SETUP FOR TEST I.

Figure 6. Result for the Operation Test

TABLE IV. TEST´S METRICS.

Now that the operation of GA and HC have been tested, TABLE VI. FEATURE´S IMPORTANCES FOR TEST I, TEST II AND TEST
III.
comparative tests can be done between the two intelligent
systems and the Random technique to evaluate the
performance each other, measuring the Efficience and the
Cost Mean in an interval of 500 simulations, averaging the
results. All these tests were performed in the Warehouse In general, it can be seen in Table IX that the Cost Mean
environment because it is very representative of the system´s for the GA results lower than the Cost Mean for HC, which
operation due to the balance between available space and means that the GA could satisfy in a higher level than HC
obstacles. The results are shown in Table IX. the restrictions of the objects and of the environment.
However, it can be seen that the Efficience was higher for
HC that for GA, which means that HC could place more
objects than GA. HC can organize a bigger number of
objects than GA, but in return HC satisfies in a lower level
the restrictions. In both cases Random technique had the
worst performance.

Since in this test the system is “transparent” to the


features of the environment, only is being taken into account
Figure 8. “Temperature” regions for Performance Test II. the “Distance” feature. It can be seen then in Table IX that
B. Test II the Cost Mean was lower for GA than for HC and unlike the
The following setup was used for the second test: two previous tests, the Efficience of GA was better than the
one of HC. It shows that HC is able to place a higher
• Features of the environments are defined in Fig. 8. quantity of objects than GA while are being taken into
• The importances are shown in Table VI. account many features in the organization process, between
them “Distance”, but when only is being taken into account
• For the setup of the objects some features varied in a “Distance”, the GA has a better performance because in this
pseudorandom way and others were kept constant as case the Efficience strongly depends on the Cost Mean (The
is shown in Table VII. lower the “Distance” the higher the quantity of objects that
In this test, the results were similar to the ones of Test I, can be placed) and as GA is better than HC according to this
due to the HC Efficience was higher than the one of GA, but measure, GA gets better results.
the Cost Mean of GA was better than the one of HC. It
TABLE IX. “EFFICIENCE” AND “COST MEAN” FOR TEST I, TEST II AND
corroborates that GA satisfies better the restrictions than HC, TEST III.
but in return HC organizes more objects. The Efficience of
the Random technique was lower than the others, but its
Cost Mean was close to the Cost Mean of HC.

TABLE VII. OBJECTS SETUP FOR TEST II.

V. CONCLUSIONS AND FUTURE WORK


Some studies focus the development of a space allocation
system in a specific case study such as in [1], [12] and [15],
making difficult to implement the same system in other case
studies. In this work, the proposed intelligent space
allocation system is flexible and has been successfully
C. Test III
implemented in a Warehouse environment, a Car Parking
The following is the setup made to develop the third test: environment and a Land for Cultivation environment
• The features of the environment were unvaried and providing a suitable organization taking into account
they had no significance in this test. features and restrictions of the objects, the cars or the crops
and of the available space, also giving the possibility to set
• The importances are shown in Table VI. up priorities in those features. The proposed system manages
• The setup of the objects was made following Table complex problems in which many factors need to be
VIII, varying many features in a pseudorandom considered and a large amount of data handled.
way and keeping others constant. It has been shown that the intelligent techniques used in
this work improve the performance of the organizer in most
TABLE VIII. OBJECTS SETUP FOR TEST III. of the cases, comparing their performance with the Random
technique results, similar to results found in [12]. REFERENCES
As with other combinatorial optimization problems, in the [1] M. Bazzazi, N. Safaei and N. Javadian, “A genetic algorithm to solve
real instances of the space allocation problem it is usually the storage space allocation problem in a container terminal,”
Computers and Industrial Engineering, vol. 56, Issue 1, February
desirable to present a set of high quality solutions so that a 2009, pp. 44-52.
human administrator can decide which allocation will be
[2] R. D'Andrea and P. Wurman, “Future challenges of coordinating
finally implemented [13]. The developed system proposes a hundreds of autonomous vehicles in distribution facilities”
different high quality solution each time it runs, specially Technologies for Practical Robot Applications, IEEE International
while using GA. To find several solutions we suggest Conference on November 2008, pp. 80 – 83.
reinitiate the GA or the HC. [3] Q. Dahua, C. Guoquan and W. Zhuan, “The study of optimal goods
distribution of automated storage/retrieval system,” Information
HC tends to satisfy the “Distance” feature, placing more Management, Innovation Management and Industrial Engineering,
objects than the other techniques in most of the cases, but International Conference on 2009. vol. 2, pp. 461 – 465.
setting aside the search for the suitable place for each object. [4] S. Chou, S. Lin and C. Li, “Dynamic parking negotiation and
The reason of this behavior is one of the stopping criterion guidance using an agent-based platform,” Expert Systems with
used on HC technique that makes the algorithm stops its Applications: An International Journal, vol. 35, Issue 3, October 2008,
pp. 805-817.
iterations and place the object once the current region cannot
[5] K. Ganesan, and K. Vignesh, “Automated parking slot allocation
keep moving in the selected direction and one of the reasons using RFID technology,” Signal Processing and Its Applications,
to stop moving is to find an obstacle, so many times the ISSPA. 9th International Symposium on, February. 2007, pp. 1 – 4.
algorithm stops the search for a suitable place and put the [6] B. Bossyns, H. Eerens and J. Van Orshoven, "Crop area assessment
object close to an obstacle, improving the Efficience. using sub-pixel classification with a neural network trained for a
Similarities between HC and Random technique have been reference year,” Analysis of Multi-temporal Remote Sensing Images,
International Workshop on July 2007, pp. 1 – 8.
also corroborated, in Test II the Cost Mean of HC and
[7] Z. Si-ru, “Application of improved genetic algorithm in optimization
Random technique were close. The starting point of HC is to computation,” Natural computation, ICNC, Fifth International
select a random point in the available space, then is expected Conference on, vol. 4, August 2009, pp. 144 – 148.
this point starts moving until it finds a suitable region to [8] R. Rocha, F. Ferreira and J. Dias, "Multi-robot complete exploration
place the object, but this second process can be affected by using hill climbing and topological recovery,” Intelligent Robots and
the presence of obstacles. Systems, IEEE/RSJ International Conference on September. 2008, pp.
1884 – 1889.
It has been shown that GA was the best of techniques used
[9] B.B. Pal, D. Chakraborti and P. Biswas, “A genetic algorithm based
in the system. In all the tests GA had the lower Cost Mean hybrid goal programming approach to land allocation problem for
and when it was requested to improve the “Distance” feature optimal cropping plan in agricultural system,” Industrial and
(Test III) it also had the higher Efficience. Results were Information Systems International Conference on 2009, pp. 181 – 186.
similar to research results found in other fields of study such [10] J. Xu, A. Lim, C. Shen and H. Li, “A heuristic method for online
as Project Planning Optimization Problem were GA warehouse storage assignment problem,” Service Operations and
Logistics, and Informatics IEEE International Conference on 2008,
significantly outperforms HC and Random technique [14]. vol. 2. pp. 1897 – 1902.
The system proposed in [15] implements Simulated [11] A. A. Hopgood, Intelligent systems for engineers and scientists,
Annealing technique for the rural land-use allocation. It can Second Edition Published by CRC Press 2001, ch. 7.6.
be seen as a more complex version of the allocation of crops [12] M. Li, X. Chen and C. Liu, “Pareto and niche genetic algorithm for
in Land for Cultivation environment. The GA technique storage location assignment optimization problem,” Innovative
implemented in this work can be a suitable alternative to Computing Information and Control, 3rd International Conference on
2008. pp. 465 – 465.
develop the space allocation in [15] due to the similarities
[13] E. K. Burke, P. Cowling and J. D. Landa, “Hybrid population-based
between the fitness functions used, the features evaluated metaheuristic approaches for the space allocation problem,”
and the problem approach of both works. Evolutionary Computation, Proceedings of the Congress on 2001, vol.
To keep doing a research work on this area the following 1, pp. 232 - 239
recommendations are shown: As have been done in other [14] G. Antoniol, M. Di Penta and M. Harman, “Search-based techniques
application fields [13] combine GA with HC to implement a applied to optimization of project planning for a massive maintenance
project,” Proc. of the 21st IEEE International Conference on Software
hybrid technique aiming at improving the performance of Maintenance, September 2005, pp. 240 – 249.
the organization process, focusing on reducing [15] I. Santé, R. Crecente and D. Miranda, “GIS-based planning support
computational burden and finding a good balance between system for rural land-use allocation,” Computers and Electronics in
the number of allocated objects and the satisfaction level of Agriculture, vol. 63, Issue 2, October 2008, pp. 257-273.
restrictions, in other words a balance between Efficience and [16] Y. Chen and F. He, “Research on particle swarm optimization in
Cost Mean. Develop a system for a 3D space allocation, location assignment optimization,” Intelligent Control and
Automation 7th World Congress on 2008, pp. 111 – 116.
combining the Warehouse environment with a shelf model
such as the used in [16] to bring the system closer to more
complex conditions and to automatic storage and retrieval
systems (AS/RS) such as the shown in [2]. Find the best
setup for the GA´s parameters (e. g. Initial population, cross
over technique) to enhance the performance of the system.

You might also like