OBCHS: AN EFFECTIVE HARMONY SEARCH ALGORITHM WITH OPPOSITIONBASED CHAOS-ENHANCED INITIALIZATION FOR SOLVING UNCAPACITATED FACILITY LOCATION PROBLEMS

In this paper, a continuous harmony search (HS) approach is investigated for tackling the Uncapacitated Facility Location (UFL) task. This article proposes an efficient modified HS-based optimizer to improve the performance of HS on complex spatial tasks like UFL problems. For this aim, opposition-based learning (OBL) and chaotic patterns are utilized. The proposed technique is examined against several UFL benchmark challenges in specialized literature. Then, the modified HS is substantiated in detail and compared to the basic HS and some other methods. The results showed that new opposition-based chaotic HS (OBCHS) algorithm not only can exploit better solutions competently but it is able to outperform HS in solving UFL problems. * Corresponding author


INTRODUCTION
The HS optimizer thought up by (Geem et al., 2001) is one of the latest robust heuristics approaches that inspire the creativeness of musicians. Because of HS simplicity and easy carrying out, it has involved considerable attention and has been effectively employed for tackling problematic tasks (Degertekin, 2008). This algorithm is incapable to create a blameless sense of balance amongst exploration and exploitation (Saka, 2009). To relieve this problem, this article proposes an efficient modified HS-based optimizer to improve the performance of HS on complex spatial tasks in this research. For this aim, opposition-based learning (OBL) (Wang et al., 2011) and chaotic patterns (Tokuda et al., 1998) are utilized.
The key thought in OBL is to simultaneously consider an approximation and its matching opposite to recognize a better value for the existing solution (Wang et al., 2011). In many works, the scheme of opposite numbers is implemented to increase the convergence ratio of optimizers (Xu et al., 2014). Population initialization can be considered as a significant phase in HS since it can affect the fineness of the outcomes. Hence, chaos theory can also enhance the efficiency of the HS by improving the diversity of the members during iterations (Hong, 2009).
For this purpose, the proposed technique is examined against several UFL benchmark challenges in specialized literature. Several exact procedures developed before to handle the UFL task such as Lagrangean relaxation (Barcelo et al., 1990), branch and bound (Klose, 1998), dual approach (DUALLOC) (Erlenkotter, 1978), the primal-dual approaches (Korkel, 1989) and linear programming. It is recognizable that as this problem remains NP-hard (Sevkli et al., 2006); precise processes are not capable to tackle large applied problems competently. In the other hand, several attempts are offered to crack UFL task by utilizing metaheuristics. Some of main works are simulated annealing (SA) (Aydin et al., 2004), tabu search (TS) (Laurent et al., 2004) and genetic algorithms (GA) (Jaramillo et al., 2002). In this paper, the modified HS has been utilized and substantiated in detail and compared to the basic HS optimizer, GA and particle swarm optimization (PSO).

HARMONY SEARCH ALGORITHM
Harmony search (HS) algorithm can be regarded as a relatively well-established, robust meta-heuristic that has been implemented to tackle several complex tasks (Pan et al., 2011). This algorithm stimulates the improvisation of melody performers looking for an appropriate state of harmony. The HS optimizer reveals several benefits over the old-fashioned calculus-based problem solvers. The latter usually require some mathematical characteristics like differentiability and/or convexity, but HS is founded on a stochastic random exploration. It generates an updated candidate solution by examining all obtainable solutions.
The HS technique is implemented by the subsequent phases: Step1: Initialize the target problem and required parameters.
Step2: Arrange the harmony memory. The opening harmony memory (HM) is randomly created as       where j i D indicates the ith variable in the jth solution matrix and HMS denotes the size of harmony memory.
Step 3: Improvisation of a firsthand harmony based on the HM. Improvisation can be defined as the creation of a novel harmony. Then, the reorganized vector xnew is generated through these procedures: First, random selection, second, HM consideration, and then pitch adjustment.
Generally, these procedures are observed as the principal terms of HS that can be represented via () 1 .
. (1  ) ( ( 1)) . . , demonstrates the rate to pick out a appropriate value for Di by using random selection process, ( ( ( )) / is entirely produced, it should be compared to the rest of the saved vectors in HM. If it is superior to the worst value in HM, it shall be reorganized.  The conventional HS is talented to reveal a superior performance compared to former optimizers in tackling constrained tasks. Nonetheless, premature convergence to regional bests can still be perceived in handling spatial optimization problems. In solving location problems, HS is usually incompetent to attain a correct balance among exploration and exploitation capacities. To relieve this problem, this paper suggests an effective improved HS-based technique to boost the performance of HS on complicated geospatial tasks like this research. For this goal, oppositionbased learning (OBL) and chaotic motions can be used. In the next sections, the OBL is introduced to be used in HS initialization.

Mathematical definition of OBL
, then the position Y can be substituted by Y , if not, we will carry on with Y.

Opposition-based HS (OBHS)
The computational time of HS principally will be related to the spatial distribution of the early population. Hence, population initialization can be considered as a significant phase in HS since it can affect the fineness of the outcomes. The key thought in OBL is to simultaneously consider an approximation and its matching opposite to recognize a better value for the existing solution. In many works, the scheme of opposite numbers is implemented to increase the convergence ratio of optimizers.

Opposition-based Chaotic HS (OBCHS)
Analogous to other heuristics, two core stages are noticeable in HS optimizer: initialization and breeding new generations. In this work, the OBL scheme is employed in the initialization phase. Population initialization can be considered as a significant phase in HS since it can affect the fineness of the outcomes. Hence, chaos theory can also enhance the efficiency of the OHS by improving the diversity of the members during iterations (Heidari et al., 2015). In these experiments, the logistic map is utilized in the starting stage of OHS. The logistic map can be formulated as where xk signifies the kth number and k shows iteration number. Obviously, x (0, 1) with opening condition x0 (0,1). In later employment, a = 4 is utilized (Heidari et al., 2015).

UFL PROBLEMS
Location problems have been classified as NP-hard problems (Sevkli et al., 2006). Hence, these problems have been treated broadly by various methodologies. All the qualified optimizers for tackling the UFL problems are either exact or heuristic procedures. Because the UFL is NP-hard, exact procedures are incapable to tackle large scale location tasks efficiently. There are some investigations to treat UFL with metaheuristics such as SA and PSO algorithms. In the following, the improved algorithm is utilized to investigate the UFL problems.
In UFL problems with m sites and n customer, the cost of each site (fci) is fixed. There is a transference cost among both site and customer cij. There is no restriction for the location capacities, but the entire request of every customer should be allocated to one location. The intention is to determine a set of localities with a minimum cost. Here, the exact UFL is described as  (9) is considered to preserve integrity. As it is presumed that no facility has capacity restriction, the request size of every client is omitted, and then (8) verified without any request variable. The structure presented in (Sevkli et al., 2006) is also utilized here to handle the continuous UFL.

EXPERIMENTAL RESULTS
Here, the performance of new version of HS optimizer is investigated in detail. For experimentations of this paper, each algorithm is implemented by using MATLAB R2012a (7.14) on a T6400@4 GHz Intel Core (TM) 2 Duo processor PC with 4 GB RAM. The OCHS algorithm is employed to tackle 15 UFL benchmarks obtained from OR Library (Beasley, 2005 where Hi symbolizes the ith replication value and U is the finest value explored in past attempts and R shows the replications number. Other indexes are "Optimum Rate" denoted by (HR) and running time showed by (RT). In addition, HR represents the ratio of the total runs returned the best to the total trials. The greater HR and MRPE show the better quality of the solutions.  Table 3. Simulation outcomes obtained by OBCHS For more investigation, OBCHS is compared with a GA exposed by (Jaramillo et al., 2002) and ESA suggested by (Aydin et al., 2004). The associated results are tabulated in Table 4 and Table 5.  (Sevkli et al., 2006)) From the outcomes, it is notable that ESA outperforms HS and HS overtakes GA based on standard deviation. The suggested approach executes better compared to the other methods in both indexes. In detail, with respect to CPU time, proposed OBCHS is more efficient than GA, ESA and HS. Particularly for Cap-C, ESA performs with more time than OBCHS. From the results, it can be observed that OBCHS has no deviation in tackling Cap 71, Cap 72, Cap 74, and Cap 101-134. These outcomes affirm that the convergence issue can be alleviated considerably with the proposed mechanism.

CONCLUSIONS
In this article, a new HS-based technique is utilized to tackle UFL tasks. The OBCHS has been developed by using chaos and OBL theme. OBCHS has been examined on some benchmark cases and best solutions are attained after a reasonable time. The comparative outcomes of novel procedure vividly affirm that early convergence issue is mitigated significantly. New HS algorithm not only can exploit better solutions competently but it is able to outperform HS in solving UFL problems. To the best of our knowledge, this research is the first implementation of chaos based HS stated for UFL in the expert literature. For future researches, OBCHS can be employed to realize other spatial optimization tasks.