Swarm intelligence algorithms are taking the limelight in the field of map optimisation. In this research our attending centres on uniting the Particle Swarm Optimization ( PSO ) algorithm with nutrient scrounging behaviour of honey bees. The ensuing algorithm ( called HBF-PSO ) and its discrepancies are suited for work outing multimodal and dynamic optimisation jobs. We focus on the niching and speciation capablenesss of these algorithms which allow them to turn up and track multiple extremums in environments which are multimodal and dynamic in nature. The HBF-PSO algorithm performs a corporate forage for fittingness in assuring vicinities in combination with single reconnoitering hunts in other countries. The strength of the algorithm lies in its uninterrupted monitoring of the whole exploratory survey and scrounging procedure with dynamic resettlement of the bees ( solution/particles ) if more promising parts are found. We besides propose discrepancies of the algorithm in which each bee has a different place update equation and we utilize familial scheduling ( GP ) for uninterrupted development of these place update equations. This procedure ensures adaptability and diverseness in the drove which leads to faster convergence and helps to avoid premature convergence. We besides explore the usage of opposite Numberss in our algorithm and integrate resistance based low-level formatting, resistance based coevals jumping and resistance based speed computation. The proposed algorithm and its discrepancies are tested on a suite of benchmark optimisation jobs. In the concluding part of our work we report our experiments on the preparation of feedforward nervous webs using our proposed algorithms.
2 Honey Bee Foraging Behavior Inspired PSO Algorithm
In this chapter we describe our Honey Bee Foraging Particle Swarm Optimization ( HBF-PSO ) algorithm. This algorithm is modeled after the nutrient scrounging behaviour of the honey bees and performs a corporate forage for fittingness in assuring vicinities in combination with single reconnoitering hunts in other countries. The strength of the algorithm lies in its uninterrupted monitoring of the whole exploratory survey and scrounging procedure with dynamic resettlement of the bees if more promising parts are found. The algorithm has the possible to be utile for optimisation jobs of multimodal and dynamic nature.
Swarm based optimisation algorithms have witnessed considerable attending recently. Optimization algorithms based on the behaviour of birds ( PSO ) and emmets ( ACO ) are good known and widely used [ Engelbrecht, 2006 ] . PSO is known to be suited for map optimisation jobs where the end is to seek for the optimal value of a map. Multimodal map optimisation and dynamic map optimisation are two discrepancies of the general optimisation job. Research workers have attempted these two categories of jobs utilizing drove based attacks and have shown that discrepancies of canonical PSO algorithm are able to work out these jobs. Some of the chief attempts of drove based solutions for multimodal jobs include plants by Brits et Al. [ Brits, 2002b ] [ Brits, 2002c ] [ Brits, 2003 ] , Kennedy [ Kennedy, 1997 ] , Parrott and Li [ Parrot, 2006 ] , Parsopoulus and Vrahatis [ Parsopoulus, 2001b ] [ Parsopoulus, 2004 ] , and for dynamic jobs include plants by
Blackwell and Branke [ Blackwell, 2006 ] , Carlisle and Dozier [ Carlisle, 2000 ] , Eberhart and Shi [ Eberhart, 2001 ] , Hu and Eberhart [ Hu, 2002 ] , and Jin and Branke [ Jin, 2005 ] . Over the past old ages PSO has undergone many betterments and sweetenings. Majority of these alterations have been made to increase the diverseness of drove in order to better the convergence of the PSO. The alterations made to PSO include the debut of an inactiveness weight, speed clamping, speed bottlenecks, different ways of finding the personal best and planetary best places, and different speed theoretical accounts. In add-on to the alterations made to basic PSO algorithm, a assortment of PSO fluctuations have besides been developed. These include the sub-swarm based PSO algorithms and PSO with niching capablenesss. Sub-swarm based PSO algorithms include all such PSO algorithms that incorporate some signifier of sub-swarm scheme. These sub droves can either be used in cooperation with each other or in an independent mode.
2.2 Sub-swarms and Niching
2.2.1 Sub-swarm based PSO
Those PSO fluctuations in which grouping of atoms into sub-swarms have been incorporated are called sub-swarm based PSO. These sub-swarms can either be in concerted manner or competitory manner. In concerted manner there exists a type of cooperation between the sub-swarms, by and large in footings of information exchange. In competitory manner the sub-swarms are in competition with each other [ Engelbrecht, 2006 ] . We first expression at some of the concerted sub-swarm based PSO discrepancies. In [ Lovberg, 2001 ] [ Lovberg, 2002 ] , Lovberg et Al. used a sub-swarm based PSO. For each sub-swarm, a gbest PSO algorithm was employed. They used an arithmetic cross-over operator as the genteelness mechanism and allowed non merely the random choice of parents from within the sub-swarm but parents from different sub-swarms were besides allowed to bring forth offspring. They found that this genteelness mechanism worked good for a big figure of droves. In this attack the cooperation between different sub-swarms is achieved by the choice of parents from different sub-swarms therefore leting the exchange of familial stuff between them.
In multi-phase PSO ( MPPSO ) , Al-Kazemi and Mohan [ Al-Kazemi, 2002a ] [ Al-Kazemi, 2002b ] [ Al-Kazemi, 2002c ] besides employed sub-swarms by indiscriminately delegating the atoms to one of two every bit sized sub-swarms. These sub-swarms were either in an attractive force stage, in which atoms of a sub-swarm affected towards the planetary best, or in a repulsive force stage, in which the atoms of the sub-swarm moved off from the planetary best. In add-on, the sub-swarms were allowed to alter their stage after a certain figure of loops had been exceed or a user defined threshold had been crossed or if the atoms of the sub-swarm failed to demo improved fittingness within a certain figure of back-to-back loops. The cooperation here was achieved through the choice of the planetary best atom from the whole population. The life-cycle PSO ( LCPSO ) presented by Krink and Lovberg [ Krink, 2002 ] [ Lovberg, 2002 ] , is yet another illustration of a sub-swarm based PSO. In LCPSO a atom can be in one of three stages. It can either be a regular PSO atom or a GA person or a stochastic hill-climber. An single atom is allowed to alter from one stage to another if its fittingness does non better over a figure of back-to-back loops. The LCPSO can therefore be considered to dwell of three sub-swarms, each consisting of persons in one of the three stages. The cooperation between the sub-swarms consequences when an single alterations its stage and moves to another sub-swarm. Kennedy presented a bunch based PSO by using the construct of pigeonholing to PSO [ Kennedy, 2000 ] . In this attack the atoms are first clustered into a predefined figure of bunchs utilizing a simple bunch algorithm. The atoms so use the norm of personal bests of the bunch as their personal best and the planetary best is defined in the usual mode. Here excessively there are multiple sub-swarms and the cooperation between them is achieved by each sub-swarm utilizing the same planetary best.
Thompson et Al. besides presented a bunch based PSO algorithm [ Thompson, 2003 ] . It differed from the attack of Kennedy in the sense that while updating the speed of the bunch centre, the bunchs best place and the planetary best places are used. Whereas when updating the speed of other atoms, the bunch centre, the atoms personal best and the bunchs best place is used. In this attack the cooperation between the sub-swarms is achieved by the usage of planetary best when updating the speed of bunch centres.
Van lair Bergh and Engelbrecht proposed the concerted split PSO ( CPSO-Sk ) [ van den Bergh, 2000 ] [ new wave lair Bergh, 2002 ] [ new wave lair Bergh, 2004 ] . In this attack they have split each atom into K separate parts of smaller dimension. Each portion is so optimized utilizing a separate sub-swarm. The cooperation between the sub-swarms is maintained by concatenating the planetary best places from the K sub-swarms and utilizing it as a context vector to stand for the complete solution. To cipher the fittingness of each atom, it is swapped into the corresponding place of the context vector, maintaining all other constituents of the vector invariable. The original fittingness map is so used to cipher the fittingness. For an illustration of the competitory sub-swarm based PSO discrepancies we turn our attending to the PSO proposed by Silva et Al. They introduced the predator-prey PSO [ Silva, 2002 ] , in which they used a individual marauder atom and the remainder of the atoms were considered quarry. The marauder invariably moved towards the planetary best prey atom. The quarry atoms used a modified speed update equation in which in add-on to being influenced by the personal and planetary best, the quarry was besides repelled from the marauder atom. The influence of the marauder atom on the quarry particles grew exponentially as the quarry atoms came closer to the marauder atom.
2.2.2 PSO with Niching
Niching algorithms are those algorithms which are capable of turn uping multiple solutions to a job. Niches can be defined as dividers of an environment which represent one solution to a job. Speciation is the procedure of happening multiple niches or solutions. Speciess are the dividers of a population viing within an environment. They are the group of atoms which converge on a individual niche [ Engelbrecht, 2006 ] .
Niching algorithms can work either in consecutive manner, parallel manner or a combination of both. In consecutive manner, the algorithm tries to happen the optima consecutive. Once optima have been located, the hunt infinite is adjusted so that the found optima are non included in future hunts. In parallel manner, the algorithm will seek to happen the niches in analogue and will maintain a record of these niches for the whole continuance of the hunt procedure. Algorithms uniting both consecutive and parallel attacks will by and large seek to happen the niches in a consecutive mode. Once a niche has been found, it is non
instantly refined but is maintained in analogue and refined together with all other found niches [ Engelbrecht, 2006 ] . Consecutive niching can be implemented by put to deathing the PSO multiple times, get downing with a different drove each clip. The nonsubjective map is changed after a solution is found. The atoms which move towards already found solutions are therefore penalized. This is a simple method of implementing consecutive niching and was employed by Kassabalidis et Al. [ Kassabalidis, 2002a ] [ Kassabalidis, 2002b ] to invert a trained nervous web utilizing PSO. Another consecutive niching technique is to utilize nonsubjective map stretching. In this attack, one time an optimum is located, it is removed from the hunt infinite by utilizing nonsubjective map stretching ; the drove is so reinitialized and allowed to seek for other lower limit. This procedure continues in an iterative mode until a certain threshold has been achieved. This technique was used by Parsopoulous et Al. in a niching based attack for turn uping multiple optima of a uninterrupted map [ Parsopoulos, 2001a ] , [ Parsopoulos, 2001b ] [ Parsopoulos, 2004 ] . Brits et Al. proposed the nbest PSO [ Brits, 2002a ] [ Brits, 2002c ] which is a parallel niching PSO algorithm. Britishs suggested the usage of a redefined aim map which rewards a atom when it is closer to any of the possible solutions. In add-on to this, spacial vicinities are used to make species. The atoms move towards the best solution of their nearest topological vicinity, which is defined as the centre of mass of the places of the n closest atoms at any given clip. Although nbest PSO can expeditiously turn up multiple optima, it has trouble in keeping niches.
2.2.3 Sub-swarm based PSO with Niching
There are some PSO discrepancies which are capable of happening multiple solutions to multimodal jobs by using a sub-swarm based niching attack. These include the NichePSO and Species based PSO.
In the NichePSO presented by Brits et Al. [ Brits, 2002a ] [ Brits, 2002b ] [ Brits, 2002c ] [ Brits, 2005 ] , the algorithm ab initio has a individual chief drove with all the atoms. The chief drove so continues to engender smaller sub-swarms as it converges towards possible solutions. Each sub-swarm therefore maintains a separate niche and the atoms within the sub-swarm continue to germinate and better the solution. When farther betterments to the
solutions found by the sub-swarms are non possible, so the algorithm is considered to hold converged. The sub-swarms that are created are wholly independent and no information is exchanged between sub-swarms. During the class of executing if a atom from the chief swarms moves into an country being explored by a sub-swarm so that atom is absorbed into the sub-swarm. Whenever two sub-swarms start to research the same optimum so both these sub-swarms are besides merged together organizing a bigger sub-warm which takes advantage of the experience gained by both of the sub-swarms. Britishs et Al. [ Brits, 2003 ] have compared the scalability of NichePSO with two familial algorithm niching techniques, consecutive niching and deterministic crowding. The writers concluded that NichePSO has an upper manus on more complex jobs. The species-based atom drove optimisation ( SPSO ) proposed by Parrott and Li [ Parrot, 2006 ] relies on the construct of speciation to keep niches and happen multiple optima for multimodal optimisation jobs. During each of the loop, SPSO identifies atoms which are to be used as species seeds. The species seed has the best fittingness amongst all atoms in that species and is considered the vicinity best for others atoms in that species.
2.2.4 PSO for Dynamic Optimization
Now we move on towards dynamic optimisation. Carlisle and Dozier [ Carlisle, 2000 ] have presented a technique which allows the PSO to be adapted for dynamic environments. This technique involves resetting the atom memories sporadically to the current places leting the drove to track the optima with really small operating expense. At the disbursal of an extra map rating the `reset ‘ can besides be triggered by observing when the optima has moved excessively far. Eberhart and Shi [ Eberhart, 2001 ] have employed the PSO in tracking optima in a dynamic environment. Hu and Eberhart [ Hu, 2002 ] have suggested that a fixed figure of atoms be re-randomized to happen new optima every bit shortly as a alteration is detected in the environment. Jin and Branke [ Jin, 2005 ] have discussed some of the bing evolutionary optimisation techniques for dynamic environments.
Blackwell and Branke [ Blackwell, 2006 ] have proposed that the population of atoms be split into a set of droves interacting locally by exclusion parametric quantity and globally through anti-convergence operator. In add-on charged or quantum atoms are used to keep diverseness. The species-based atom drove optimisation ( SPSO ) of Parrott and Li [ Parrot, 2006 ] , mentioned above in context of multimodal jobs, can besides be used for dynamic optimisation jobs.
2.3 Honey Bee Foraging Behavior Inspired Particle Swarm Optimization Algorithm ( HBF-PSO )
We now describe a fresh algorithm, which we have called HBF-PSO, [ Baig, 2007 ] and show that it can be applied to unimodal, multimodal and dynamic optimisation jobs. The chief feature of the HBF-PSO is the incorporation of three new characteristics, in the PSO algorithm. We propose coincident geographic expedition and development of the fittingness landscape, along with a chalkboard type of information sharing mechanism. These characteristics allow parallel seeking for fittingness extremums and resettlement of hunt towards assuring countries. These characteristics are derived from a theoretical account of the nutrient scrounging behaviour of honey bees. It has been observed that lookout bees gather information and the bees at the hive get asynchronous updates by agencies of waggle dances of the lookout bees. If a promising flower spot is discovered, bees are sent from the hive for scrounging. The measure of allocated bees is relative to the quality of spot. In HBF-PSO, several droves of particle-bees eatage ( hunt for peak fittingness ) in a corporate and co-ordinated mode. A swarm assembles in a promising part and so moves towards the top of the extremum. Furthermore, there are scout particle-bees that are placed indiscriminately and execute a local hunt. The forage and exploratory survey are allowed to go on for limited continuance, sufficient plenty for a drove to scrounge its part of allotment ( converge on extremum fittingness ) . Matching to the shake dance of the bees at the hive, a chalkboard is maintained which may be updated asynchronously whenever some utile information is available either from the exploratory survey or scrounging particle-bees. Particle-bees are reallocated after analysing the information on the chalkboard.
2.3.1 Related Work
Some research workers have tried in the yesteryear to mime the behaviour of honey bees in their algorithms. Abbass [ Abbass, 2001a ] [ Abbass, 2001b ] has presented the Marriage in honey Bees Optimization ( MBO ) algorithm which is modeled after the coupling flight of queen bees. Pham [ Pham, 2006a ] [ Pham, 2006b ] has proposed the Bees algorithm which is inspired by the nutrient scrounging behaviour of the honey bees. The chief feature of Bees algorithm is the inactive sampling of fittingness landscape. The Artificial Bee Colony algorithm has been proposed by Karaboga [ Karaboga, 2005 ] , [ Basturk, 2006 ] , [ Karaboga, 2007 ] . It is a population based algorithm in which a possible solution to the optimisation job is represented by the place of a nutrient beginning and the nectar sum of a nutrient beginning corresponds to the quality ( fittingness ) of the associated solution. Examples of other applications based on honey bee behaviour include a routing algorithm [ Wedde, 2004 ] [ Farooq, 2009 ] . Several bugologists and sociologist have besides tried to pattern the behaviour of honey bees [ Loengarov, 2006 ] . We elaborate a little more on the Bees algorithm and Artificial Bee Colony algorithms in the undermentioned paragraphs because they are comparatively closer to our proposed algorithm. Bees Algorithm Pham et Al. [ Pham, 2005 ] , [ Pham, 2006 ] have proposed the Bees Algorithm which is a population-based algorithm that mimics the nutrient scrounging behaviour of droves of honey bees. In its basic version, the algorithm performs a sort of inactive vicinity trying combined with inactive random trying. The Bees algorithm starts with the lookout bees being placed indiscriminately in the hunt infinite. The fittingness of the lookout bees is evaluated, after which, the bees that have the highest fittingness are chosen as selected bees and sites visited by them are chosen for inactive vicinity sampling. The algorithm takes samples in the vicinity of the selected sites by puting bees in them, taking more samples near the best sites. For each site merely the bee with the highest fittingness is selected to organize the following bee population. The staying bees are used for random sampling. These stairss are repeated until a stopping standard is met.
Artificial Bee Colony algorithm The Artificial Bee Colony algorithm has been proposed by Karaboga [ Karaboga, 2005 ] , [ Basturk, 2006 ] , [ Karaboga, 2007 ] . It is a population based algorithm. The settlement consists of three groups of bees: employed bees, looker-ons and lookouts. A possible solution to the optimisation job is represented as the place of a nutrient beginning and the nectar sum of a nutrient beginning corresponds to the quality ( fittingness ) of the associated solution. The figure of the employed bees is equal to the figure of solutions in the population. At the first measure, a indiscriminately distributed initial population ( nutrient beginning places ) is generated. An employed bee produces a alteration on the beginning place in its memory and evaluates the fittingness at that place ( calculates the nectar sum ) . If the fittingness ( nectar sum ) of the new place is higher than that of the old place, the bee memorizes the new beginning place and forgets the old one ; otherwise it keeps the place of the old one in memory. After all the employed bees have evaluated the new places, the looker-ons go to these places with more looker-ons traveling towards better places and less looker-ons traveling towards less fit places. The looker-ons besides produce a alteration on that place and measure the fittingness at that place. The lookouts indiscriminately select places to measure. This rhythm continues until the expiration standard is meet. Besides if the fittingness of certain employed bee does non better for some clip so that employed bee is converted to a lookout bee.
2.3.2 Food Foraging Behavior of Honey Bees
In this subdivision, we present our theoretical account of the nutrient scrounging behaviour of honey bees. For the development of the HBF-PSO algorithms the existent forage procedure is simplified as follows. At the start of the nutrient scrounging procedure lookout bees are sent out from the hive to look for flower spots. These bees return with information about the measure, quality and way of flower spots and pass on this information to the other bees by agencies of a shake dance. Extra bees are so recruited based on available information to scrounge the ascertained flower spots, with more bees being attracted towards richer flower spots. The exploratory survey and scrounging procedure continues at the same time during the whole harvest home season.
Our theoretical account of the above biological procedure is as follows. The settlement is supposed to dwell of a fixed figure of bees. Initially all the bees go out in random waies to happen
flower spots. After a certain clip they all return to the bee hive and analyse the collected information about the profusion ( quality and measure ) of the spots that they have discovered. Several bees go out to scrounge the ascertained flower spots and other bees go out to reconnoiter the staying part indiscriminately. After passing a certain clip in foraging/scouting, all of the bees once more return to the bee hive and analyse the gathered information. These alternate turns of analysis and foraging/scouting are repeated once more and once more until the whole country has been sufficiently searched. These two stages are explained in more item below. Book-Keeping & A ; Analysis Phase Some facets of the history of reconnoitering and scrounging are recorded and analyzed at the hive ( matching to the shake dance ) . During the analysis stage, the profusion of each flower spot is compared with the profusion of other spots. The bees decide which of the spots need to be foraged exhaustively in the following unit of ammunition. Each selected spot is assigned a separate drove composed of a few bees. The staying bees are assigned to execute random hunt. Richness record of each spot being foraged is kept and monitored during subsequent analysis. Based on this record a spot may be declared as wholly foraged and bees are no longer assigned to that spot. Furthermore, the richer parts are foraged on a precedence footing. A current spot may be abandoned on the find of a new richer spot. After the forage of the richer spots is complete the bees may return to this abandoned spot once more. Scouting & A ; Foraging Phase Each selected flower spot is foraged by a drove of several bees in a concerted and corporate mode. On the other manus, each lookout bee ( get downing from an arbitrary point ) explores its encompassing part locally and without any aid from other bees. This forage and reconnoitering procedure continues for some pre-specified clip continuance. At the terminal of that clip period the bees gather once more at their hive, pool the information and get down the analysis stage.
2.3.3 HBF-PSO Algorithm
In HBF-PSO one bee is one atom, i.e. one complete solution of the job. A fixed figure of bees ( H ) are created and placed indiscriminately in the fittingness landscape. The fittingness at each of these points is calculated. The points are sorted on the footing of their fittingness and bee droves are created around each of the m best points. If any two or more of the best points are near one another than they are considered as belonging to the same part and merely one drove is created for them. In other words, overlapping of droves in non allowed during their creative activity. This status encourages that there may be merely one development per promising part. The creative activity of droves agencies that n bees have been assigned to each of these droves. The staying bees ( f = h – m * N ) are used as lookouts and placed indiscriminately. These two types of bees are called foraging bees and reconnoitering bees, severally. The forage and reconnoitering bees hunt for a little, fixed figure of loops in the undermentioned mode. The n forage bees search for a extremum as a drove in a co-ordinated and corporate mode harmonizing to classical PSO algorithm. Mean while each of the degree Fahrenheit lookout bees searches around its stead of arrangement entirely. This is done harmonizing to PSO algorithm without societal constituent. After a few loops of coincident forage and reconnoitering a planetary analysis takes topographic point. The best fittingness reported from each of the drove and the best fittingness of each of the lookout bees are all sorted and the droves for the following loop are determined on the footing of the information. An old drove is allowed to last merely if its best fittingness comes within the first m best points. If one of these best points has been freshly discovered by a lookout bee, so a new drove is created around it. As mentioned before, no imbrication is allowed during the creative activity of droves. This eliminates one of the two droves which have come near to one another ( the drove with the lower fittingness is eliminated ) . This mode of making droves gives the algorithm a penchant for researching the higher fittingness parts on a precedence footing which normally leads to the find of higher extremums foremost in multi extremum environments.
If a drove has converged on a extremum its best fittingness value becomes dead. This phenomenon is kept under observation for each drove. If a drove is non describing any
betterment in fittingness during a few back-to-back analysis stages, so it is assumed to hold reached a extremum. The drove is disbanded and the part around its best fittingness is considered as wholly foraged. A wholly foraged part is so considered tabu and is non included in any farther hunt. In instance of dynamic optimisation job, upon declaring a part as tabu, a bee is placed on the best place to supervise for alteration. The fittingness of this bee is revaluated during each analysis stage. Once the fittingness is found to hold changed, it is removed from the foraged part list and becomes available for hunt in subsequent loops. Please note that this monitoring is merely employed in dynamic environment. The scouting/foraging followed by analysis continues until a stopping status is fulfilled. This status can be a prescribed figure of rhythms or the find of a pre-specified figure of extremums or sufficient forage of the hunt infinite ( for illustration, if we are unable to happen undiscovered parts with a fittingness greater than 10 % of the best fittingness ) or a combination of these three conditions.
Figure 2-1 Overview of HBF-PSO algorithm
2.3.4 Summary of HBF-PSO Algorithm
An overview of the HBF-PSO algorithm is presented in Figure 2-1 and the flow chart is given in Figure. 2-3. The algorithm can be summarized in the stairss outlined in Figure 2-2.
Begin N = population size ; M = figure of droves ; Initialize parametric quantities ; Generate initial population of bees by measuring fittingness at a random topographic point ; While ( expiration standards is non met ) Compile a list of all best fittingness ; Sort this list in falling order ; While ( thousand droves have non been created ) Remove the current top fittingness from the list. If ( neighborhood convergences any other vicinity of a drove ) Continue ; End if If ( neighborhood convergences any country declared as wholly foraged ) Continue ; End if If ( retrieved fittingness is due to a drove ) Continue with the old drove ; Else Place one bee on the place where that fittingness has been obtained ; Place other n – 1 bees indiscriminately in the adjacent part around it ; End if End while Allocate staying bees for random exploratory survey ; Run PSO for each drove ; Run PSO with out societal constituent for lookout bees ; If ( any drove has converged and become stagnant ) Mark the corresponding part as wholly foraged ; If ( optimisation job is dynamic ) Place an excess bee at the planetary best place of foraged part to supervise it ; End if End if If ( optimisation job is dynamic ) Evaluate the fittingness of all excess bees allocated for the monitoring of foraged parts ; If ( fittingness has changed ) Remove part from foraged parts list ; End if End if End while End
Figure 2-2 Summary of HBF-PSO algorithm
2.3.5 Parameters of HBF-PSO
The list of parametric quantities which need to be set by the user is as follows: ( I ) entire figure of bees to be used, ( two ) the figure of bees in each drove, ( three ) the figure of reconnoitering bees, ( four ) the figure of droves, ( V ) the size of the vicinity where droves are ab initio formed, ( six ) the figure of loops in a foraging/scouting stage, ( seven ) the parametric quantities of the PSO algorithm used by the forage bees for corporate hunt, and ( eight ) the parametric quantities of the PSO algorithm used by the lookout bees for stray searching. For the clip being we leave aside the parametric quantities of points ( I ) , ( seven ) and ( eight ) since they can be borrowed from literature of classical PSO. We concentrate on the four parametric quantities listed from ( two ) to ( six ) . The ratio of scrounging bees to reconnoitering bees: This ratio is dependent on the type of fittingness landscape being explored. It can be fixed for the full tally of the algorithm or maintain variable as a map of the fittingness values of the current spots being explored. If the spots are of high fittingness we can delegate more bees to acquire to their extremums rapidly and subsequently on transfer them for geographic expedition. Number of elect parts ( droves ) for a given foraging period: We can hold a fixed figure of droves throughout the continuance of the HBF-PSO algorithm. Though new droves may be created and old droves disbanded, but the entire figure may be kept the same. We can besides take to hold a variable figure of droves by enforcing a minimal fittingness threshold on the fittingness values viing for making a new drove. In both the instances the maximal droves allowed will hold to be a sensible figure maintaining in position the figure of bees available for scrounging ( at least two or three bees have to be assigned to each drove and we besides have to reserve some bees for reconnoitering intents ) . The size of a spot ( i.e. the part specifying the spot ) : Regions may necessitate to be defined for swarm formation at the beginning of the foraging stage. The size of these parts can be kept changeless throughout the executing of the algorithm ( e.g. one ten percent of the scope of a dimension, centered on the best fittingness point known for that spot at the clip of analysis ) . The part can besides be defined in a variable manner after a elaborate analysis of the recent scrounging history of a spot.
Figure 2-3 Flowchart of HBF-PSO algorithm.
The figure of loops specifying the foraging period: This can be fixed at a sensible arbitrary figure ( e.g. 15 ) . This can besides be made self-adaptive and can be shortened if we are making fittingness tableland earlier or if we observe oscillations in swarm motion ( bespeaking multiple extremums in the same country ) . Similarly, it can be lengthened if we observe that droves are being created in the same part once more and once more in back-to-back analysis. In this chapter after given an overview of sub-swarm based PSO algorithms, PSO algorithms with niching and sub-swarm based PSO algorithms with niching, we presented our proposed honey bee scrounging behaviour inspired PSO algorithm. In the following chapter we present the experimentation that we have carried out to prove its public presentation.