Artificial Bee Colony Algorithm

Dynamic Link Library (DLL), v1

Artificial Bee Colony (ABC) algorithm is an optimization algorithm based on the intelligent foraging behaviour of honey bee swarm, proposed by Dervis Karaboga in 2005.

A brief guide on implementing the library:

The library implementation consists of two primary functions:

initABC() – the initialization function

runABC() – performs a single run of the algorithm and returns the solution obtained

Acoording to the design of the ABC algorithm, possible solutions are denoted as food sources that the honeybees try to investigate and forage. The model of a food source is implemented using a C/C++ structure which looks like this:

struct foodSource {
	double dim[MAX_DIM_COUNT];
	double objFxValue; 
	double fitness;
	double trials;
	double prob;
};

Here, MAX_DIM_COUNT is a constant that indicates the maximum number of dimensions of a solution that the program can handle.

To use the algorithm, it must be first initialized by invoking the **initABC() **function to which a number of parameters need to be passed, the prototype of which looks like this:

void initABC (
		  int dimCount, 
		  int colonySize, 
		  int maxIterations, 
		  int trialsLimit, 
		  double lB[], 
		  double uB[], 
		  objectiveFunction objFxPtr, 
		  constraintFunction consFxPtr, 
		  int maxMin, 
		  char dumpFileName[]
); 

The parameters have been described in detail:

• dimCount (integer) – number of dimensions or variables in the optimization problem. It is limited to a maximum value of 50 as set in the library source by default (can be modified).

• colonySize (integer) – size of the bee colony, i.e. the hive size. It is limited to a maximum value of 100 as set in the library source by default (can be modified).

• maxIterations (integer) – maximum number of iterations (also known as MCN or the maximum cycle number) that the algorithm will attempt to go through before it terminates deciding that it has obtained its solution

• trailsLimit (integer) – number of consecutively failed attempts the algorithm will tolerate to improve a bad solution before it is discarded

• lb (double precision floating point array) – a set of values denoting the lower limits of the various dimensions of the problem. Array length is hence equal to dimCount

• ub (double precision floating point array) – a set of values denoting the upper limits of the various dimensions of the problem. Array length is hence equal to dimCount

• objFxPtr – a function pointer that will hold the address of the objective function of the problem. Its return type is double and it accepts one input parameter of the foodSource type.

• consFxPtr – another function pointer that will hold the address of any additional function that will check whether a randomly generated possible solution to the problem does satisfy any other constraints imposed by the problem at hand if required. Its return type is integer and it accepts one input parameter of the foodSource type. It should return 1 if all constraints are passed, and 0 if one or more constraint check fails for the generated possible solution.

• maxMin (integer) – an argument that denotes whether the problem to be solved is a maximization problem (pass the value 0) or minimization problem (pass the value 1)

• dumpFileName (*string of maximum length 100) – a string that holds the name of a data-sheet text file that will be outputed by the program containing the best solution generated per iteration in the algorithm, thus giving an indication of the convergence of the solution.

The next step in the process is much simpler.

Invoke the runABC() function to perform a single run of the ABC algorithm and obtain a generated solution for that run. The generated solution returned by the function is of the foodSource type.

The runABC() function takes one input parameter of the integer type which dictates whether to generate a data-sheet text file for this run (pass non-zero value is yes, or zero if no). Hence the runABC() function looks simply like this:

foodSource runABC(int dumpData); 

The custom function pointer type definitions and the primary function prototypes mentioned above are provided as follows:

typedef double (*objectiveFunction)(foodSource);

typedef int (*constraintFunction)(foodSource);

typedef void (*initABCFunc)(int, int, int, int, double[], double[], objectiveFunction, constraintFunction, int, char[]);

typedef foodSource (*runABCFunc)(int); 

Thus, you can see that by simply invoking these two functions from a DLL, the ABC algorithm can be functionally run and implemented on any multi-modal optimization problem.

The project contains 3 folders:

• abcalgo32 - the source for a 32-bit implementation of the library in DEV-C++ (v5.4.1 used)
• abcalgo64 - the source for a 64-bit implementation of the library in DEV-C++ (v5.4.1 used)
• rastriginTest - example code for calling the library (both 32-bit and 64-bit) and using it to find solution to the Rastrigin's Function. It contains the following files:
  • a 64-bit compiled DLL of the library (abcalgo64.dll) [compiled using TDM-GCC 4.7.1 64-bit]
  • a 64-bit version of a sample program using the library (testRast64.cpp)
  • a 32-bit compiled DLL of the library (abcalgo32.dll) [compiled using TDM-GCC 4.7.1 32-bit]
  • a 32-bit version of a sample program using the library (testRast32.cpp)