Artificial Bee Colony Intelligence Algorithm - Research Paper Example

Abstract

Swarm intelligence (SI) is a type of biological artificial intelligence. It is defined as a collective behavior of decentralized and self-organized swarms. Well-known examples for swarms include birds, fish schools and colonies of insects such as ants, bees and termites. In the period of 1990s, two methods of swarm intelligence based on fish schooling and ant colony were introduced. Since then, the area has attracted increased interest from many scholars. A new intelligence approach based on swarm intelligence is referred to as artificial bee colony. Starting from the year 2000, researchers have increasing become interested in the behavior of swarm systems as a means of describing new intelligent approaches. This has led to the development of many algorithms. The artificial bee colony (ABC) is the one that has been widely studied and applied to solve real world problems. The current analysis presents a comprehensive review of the developments in ABC and its usage. It is hoped that the current study will be beneficial to future studies on swarm intelligence, especially ABC algorithm.

Keywords: Swarm intelligence; Artificial bee colony algorithm (ABC); Bee swarm intelligence

1.0 Introduction to Problem Area

Nowadays, swarm intelligence is a prominent new field in artificial biological intelligence. It is inspired by the collective behavior of hefty individuals that collaborate to achieve a common goal. SI is seen in natural structures such as that of flocking birds, ant colonies and fish schooling. Artificial Intelligence (AI) has been in existence for many years. There are diverse definitions of AI in literature. The most used definition is that of John McCarthy who defined it as “the science and engineering of making intelligent machines” (McCarty, 2007). Computational intelligence (CI) is a relatively new research field that is ordinarily referred to as AI, too. CI is defined as the study of the intelligence design of intelligent agents (i.e. systems that perceive their environment and then take measures to capitalize on their chances of success). While CI procedures are counted as AI, there is a clear difference between the two. CI consist of a set of nature-inspired computational methodologies and tactics to address intricate problems in the real world. Subjects in CI include neural networks, fuzzy systems evolutionary computation (EC). Some of the eminent evolutionary algorithms include genetic algorithms (GA), evolution strategies (ES), genetic programming, differential evolution (DE), and evolution programming and differential evolution (DE).

SI is becoming ever more important as an area of study for computer scientists, economists, operational researchers, engineers, bioinformaticians and other experts. This is due to the fact that natural intelligent swarms can solve many problems relating to labor division, finding food and other real world problems. In 1990, Dorigo et al. (1991) proposed two crucial approaches. These approaches were based on ant colony and were referred to as ant colony optimization (ACO). Kennedy and Eberhart (1995) also proposed the particle swarm optimization (PSO) approach in 1995. A number of researchers have proposed new versions of these approaches as they seek to apply them in solving real world problems. Several papers related to the applications of ABC as well as numerous surveys have been presented. The current review seeks to have an objective review of some of the applications of ABC and certain advances in the field.

2.0 Problem Statement

The area of artificial bee colony (ABC) is growing at a very high rate. Many researchers are trying to devise their own versions of ABC algorithm with the aim of solving problems. It is becoming difficult identifying which ABC version is more appropriate in their action to solve real world problems. As such, there is a need for a study that will assess the development and application of ABC algorithm. This study should inspire and inform future studies on swarm intelligence, especially ABC algorithm.

3.0 Artificial Bee Colony: ABC Approach

3.1 Characteristics of Intelligent Swarms

Bonabeau et al. (1999) provide certain features of ABC. They include:

• Fluctuations: presence of random walks, errors and task switching among h are vital for creativity. Randomness is every so often significant for emergent structures since it enables the discovery of new solutions.
• Positive feedback: encouraging the creation of convenient structures. Recruitment and as well as reinforcement such as trail laying and following in some ant species is an example of positive feedback.
• Multiple interactions: agents in the swarm will utilize information coming from the other agents so that the information spreads throughout the network.
• Negative feedback: Counterbalancing positive feedback and having plans to help stabilize the collective pattern. To evade the saturation which be might occur in terms of available victims that need help, a negative feedback mechanism is necessary.

3.2 The Algorithmic Structure of ABC

Just like in the minimal model for forage selection for real honeybees, the colony of the artificial bees in ABC comprises of three groups of bees namely employed bees, onlooker bees, and scout bees. Initially, scout bees discover all food source positions. Subsequently, employed bees and onlooker bees exploit the nectar of food sources, and this constant exploitation makes them to be exhausted. In ABC, the locations of the food source signify possible answers to key problems while nectar amounts parallels the quality of the associated solutions. In its basic form, the sum of the employed bees equals the amount food sources since all employed bees are linked to only one food source. A general algorithmic structure for ABC optimization methodology is given as shown below in various phases:

• Initialization
• REPEAT
• Employed Bees
• Onlooker Bees
• Scout Bees
• Remember best solution
• UNTIL (Maximum Cycle Number)

Within the initialization phase, artificial scout bees initiate the number of food sources or rather solutions thus setting the control parameters. In ABC algorithm, the positioning of a food source embodies a possible solution for an optimized issue while nectar amounts relates with the fitness of the solution. The sum total of employed or onlooker bees equals the number of solutions. An onlooker bee will select a certain food source based on the probability value allied to the food source, pi, which is calculated by the expression shown below (3.1):

(3.1)

Where;

• fiti is the quality value of the solution i assessed by employed bee and is proportionate to the amount of nectar food sources in the position
• SN equals the number of food sources, which is also proportional to the number of employed bees (BN).

4.0 Studies on ABC optimization

Karaboga is one of the researchers that has done extensive research on ABC algorithm. He invented the ABC algorithm in 2005 and published the first conference paper on the algorithm in 2006 (Basturk & Karaboga, 2006). Karaboga and Basturk (2007) presented the first journal article discussing ABC and appraising its performance. In their article, they compared the performance of ABC to the PSO, GA and particle swarm-inspired evolutionary algorithms. In the following year, Karaboga and Basturk (2008) presented a performance appraisal of ABC. Currently, there exists several codes on ABC that are written in various programming languages. ABC is quite simple algorithm, the reason why its application is highly straightforward and practical. Its computations costs are also very low. After Karaboga and his colleagues invented ABC algorithm, several authors have carried out studies on the algorithm, trying to come up with different versions aimed at improving its applicability. Others have just resorted to comparing the various forms of ABC and their modifications, while others have focused on its application.

4.1 Comparisons and Modifications

In the beginning, ABC optimization was suggested to solve numerical problems. As a result, the Initial studies aimed at evaluating how the algorithm worked on a comprehensive set of numerical test functions thus comparing it with conventional algorithms such as PSO, ACO, GA, and DE. In 2007, Karaboga and Basturk compared ABC with PSO and GA for in their multivariable functions. Karaboga and Akay (2009) the offered a comparative study of the ABC algorithm in optimizing massive sets of numerical functions and compared the results with those of the PSO, ES and DE algorithms. Their study was closely followed by that of Mala et al. (2009) who applied ABC in test suite optimization, comparing ACO with ABC. Mal et al. concluded that ABC possesses many advantages when compared to ABC. The increased success of ABC algorithm as an optimizer has motivated scholars to extend the use of the algorithm to other platforms. For instance, Akay and Karaboga (2009) decided to apply ABC algorithm to integer programming, concluding that ABC has a capability of handling integer programming competently.

4.2 Hybridization

So as to improve the performance of ABC algorithm, it has been combined with conventional and revolutionary algorithms leading to what is known as hybridized ABC. For example, a hybrid simplex ABC algorithm combines the Nelder-Mead simplex method with the ABC approach to advance search efficiencies in computation (Kang et al., 2009). A novice hybrid algorithm based on ABC by Marinakis et al. (2009) presented randomized adaptive search procedure with the aim of optimally clustering the n objects into the k clusters. Other studies have combined ABC optimization with quantum evolutionary algorithm while others have integrated the ABC algorithm with greedy heuristic and local search in trying to solve the quadratic knapsack problem. A fresh hybrid approach combining GA and ABC enhanced efficiency, tuning the PI speed controller in a permanent synchronous motor drive of a magnet (Jatoth & Rajasekhar, 2010).

4.3 Applications of ABC

The ABC algorithm has been applied in many fields. One of such area is in the training of neural networks. For example, it has been employed in ABC algorithm to train feedforward neural networks such as in locating the optimal weight set (Karaboga et al., 2007). Akay and Karaboga (2009) also tested the ABC algorithm on artificial neural networks employed in signal processing applications where it worked perfectly. Two researchers, Karaboga and Ozturk (2009) used ABC on feed-forward neural networks, classifying various datasets used in machine learning community. Some researchers, in order to solve the optimization problems, have used ABC electrical engineering.

In 2008, Rao and colleagues established a new method that utilizes the ABC algorithm in ascertaining the sectionalizing switch used in distribution systems to reduce losses. Then followed De Oliveira et al. (2009) studies on the performance of the ABC algorithm when used in PWR nuclear power plants diagnose accidents. Lalitha et al. (2010) utilized the ABC algorithm in DG placement to lower losses in radial distribution systems. Initially, Rao had presented a method that was based on the ABC algorithm that was used in capacitor placement for distribution systems with the aim of reducing power losses and improving voltage profiles.

ABC algorithm has found increased application both civil and mechanical engineering areas. It has been used in modeling and optimizing process parameters in machining wire electrical discharge. ABC has also been used in parameter optimization for multi-pass milling processes with the results compared to those of SA and PSO. Gómez-Iglesias et al. (2010) analyzed a modification of the original ABC algorithm to optimize confined plasma equilibrium in nuclear fusion devices as well as its use in grid computing environments. ABC algorithm has also been introduced in structural optimizations of space and planar trusses under stress, buckling constraints and displacement (Hadidi et al., 2010).

It is crucial to note that the application of ABC algorithms has been minimal in wireless sensor networks. One study by Udgata et al. (2009) analyzed the application of ABC sensor deployment problems, modeled as data clustering. Then followed the study by Mini et al. (2010) in which the researchers applied ABC algorithm to sensor deployment in a 3D terrain. Karaboga and Ozturk (2009) have also described ABC algorithm based the dynamic deployment approach for both mobile and stationary sensor networks geared towards augmenting the coverage area of the networks. Mini et al. (2010) presented a sensor deployment algorithm based on ABC, targeting coverage issue in 3D terrain. The study by Okdem et al. (2011) assessed the efficiency of ABC algorithm on routing operations for wireless sensor networks.

5.0 Summary and Conclusion

The developments in ABC algorithm as a component of swarm intelligence has exceeded most expectations, notwithstanding the fact that ABC is barely 9 years old. An overview of the articles analyzed in the current paper clearly shows that much of the effort has been put towards the algorithmic and applicability aspects of the ABC algorithm. Therefore, many areas of ABC algorithm remain to be explored. Researchers should, therefore, investigate other topics under ABC. Some notable topics include theoretical studies and the self-adaptability of control parameters. A design of ABC without prior fine-tuning of parameters is another viable topic to explore.

It was evidenced that ABC lacks in-depth theoretical studies. Consequently, it will be fascinating if a researcher decided to perform a theoretical study, for instance, on the convergence and run-time properties of ABC. Some other aspects of ABC such as its dynamics and fitness landscapes could offer attractive theoretical research themes on the algorithm. Even though ABC has a high potential, it is clear that some modifications are critical for better performance. In addition, ABC has been very promising as a framework for integrating traditional algorithms.

Another promising area for ABC application is its dynamic optimization environments. To improve the performance of ABC with regard to convergence, novel production mechanisms can be suggested where scout production delivers diversity. Accordingly, improved strategies can be offered in the scout production phase. Furthermore, such a strategy can be activated adaptively depending on search dynamics. There can also be new selection strategies for onlooker distribution to sources to improve the performance of ABC algorithm.

Just like any other optimization approach, ABC has its own shortcomings. For instance, it does not use an operator like crossover, which is employed in DE and GA or DE distribution. This makes ABC’s convergence performance to be slow. Therefore, more research needs to be taken on ways of convergence performance. Symbolic regression is another problem associated with ABC. Extended versions of the ABC algorithm can be used to improve on this problem. It is likely that more studies are going to be carried out on ABC in the coming years.

2.0 Problem Statement

The area of artificial bee colony (ABC) is growing at a very high rate. Many researchers are trying to devise their own versions of ABC algorithm with the aim of solving problems. It is becoming difficult identifying which ABC version is more appropriate in their action to solve real world problems. As such, there is a need for a study that will assess the development and application of ABC algorithm. This study should inspire and inform future studies on swarm intelligence, especially ABC algorithm.

3.0 Artificial Bee Colony: ABC Approach

3.1 Characteristics of Intelligent Swarms

Bonabeau et al. (1999) provide certain features of ABC. They include:

• Fluctuations: presence of random walks, errors and task switching among h are vital for creativity. Randomness is every so often significant for emergent structures since it enables the discovery of new solutions.
• Positive feedback: encouraging the creation of convenient structures. Recruitment and as well as reinforcement such as trail laying and following in some ant species is an example of positive feedback.
• Multiple interactions: agents in the swarm will utilize information coming from the other agents so that the information spreads throughout the network.
• Negative feedback: Counterbalancing positive feedback and having plans to help stabilize the collective pattern. To evade the saturation which be might occur in terms of available victims that need help, a negative feedback mechanism is necessary.

3.2 The Algorithmic Structure of ABC

Just like in the minimal model for forage selection for real honeybees, the colony of the artificial bees in ABC comprises of three groups of bees namely employed bees, onlooker bees, and scout bees. Initially, scout bees discover all food source positions. Subsequently, employed bees and onlooker bees exploit the nectar of food sources, and this constant exploitation makes them to be exhausted. In ABC, the locations of the food source signify possible answers to key problems while nectar amounts parallels the quality of the associated solutions. In its basic form, the sum of the employed bees equals the amount food sources since all employed bees are linked to only one food source. A general algorithmic structure for ABC optimization methodology is given as shown below in various phases:

• Initialization
• REPEAT
• Employed Bees
• Onlooker Bees
• Scout Bees
• Remember best solution
• UNTIL (Maximum Cycle Number)

Within the initialization phase, artificial scout bees initiate the number of food sources or rather solutions thus setting the control parameters. In ABC algorithm, the positioning of a food source embodies a possible solution for an optimized issue while nectar amounts relates with the fitness of the solution. The sum total of employed or onlooker bees equals the number of solutions. An onlooker bee will select a certain food source based on the probability value allied to the food source, pi, which is calculated by the expression shown below (3.1):

(3.1)

Where;

• fiti is the quality value of the solution i assessed by employed bee and is proportionate to the amount of nectar food sources in the position
• SN equals the number of food sources, which is also proportional to the number of employed bees (BN). Read More