single population met a heuristics

8/3/2019 Single Population Met a Heuristics

1/18

Single PopulationMetaheuristics


2/18

Concepts

Coding.

Random Generation.

Some Operators.

Fitness Function.

Stopping Criteria.

Constraints.


3/18

Some Metaheuristic Classifications

Nature inspired versus nonnature inspired: Many metaheuristics are inspired bynatural processes: evolutionary algorithms and artificial immune systems frombiology; ants, bees colonies, and particle swarm optimization from swarm intelligenceinto different species (social sciences); and simulated annealing from physics.

Memory usage versus memoryless methods: Some metaheuristic algorithms arememoryless; that is, no information extracted dynamically is used during the search.Some representatives of this class are local search, GRASP, and simulated

annealing. While other metaheuristics use a memory that contains some informationextracted online during the search. For instance, short-term and long-term memoriesin tabu search.

Deterministic versus stochastic: A deterministic metaheuristic solves an optimizationproblem by making deterministic decisions (e.g., local search, tabu search). In stochastic

metaheuristics, some random rules are applied during the search (e.g., simulatedannealing, evolutionary algorithms). In deterministic algorithms, using the same initialsolution will lead to the same final solution, whereas in stochastic metaheuristics,

different final solutions may be obtained from the same initial solution. Thischaracteristic must be taken into account in the performance evaluation ofmetaheuristic algorithms.


4/18

Cont.

Population-based search versus single-solution based search: Single-solution based algorithms (e.g., local search, simulated annealing)manipulate and transform a single solution during the search while inpopulation-based algorithms (e.g., particle swarm, evolutionary algorithms)a whole population of solutions is evolved. These two families havecomplementary characteristics: single-solution based metaheuristics areexploitation oriented; they have the power to intensify the search in localregions. Population-based metaheuristics are exploration oriented; theyallow a better diversification in the whole search space. In the next chaptersof this book, we have mainly used this classification. In fact, the algorithmsbelonging to each family of metaheuristics share many search mechanisms.

Iterative versus greedy: In iterative algorithms, we start with a completesolution (or population of solutions) and transform it at each iteration usingsome search operators. Greedy algorithms start from an empty solution,and at each step a decision variable of the problem is assigned until acomplete solution is obtained. Most of the metaheuristics are iterativealgorithms.


5/18

Local Search


6/18

Others


7/18

Simulated Annealing


8/18

Threshold Accepting


9/18

Record-to-Record & Great Deluge


10/18

Tabu Search

InitializeIdentify initial Solution, Create empty TabuList, Set BestSolution=SolutionDefine TerminationConditionsDone=FALSERepeatif value of Solution> value of BestSolutionthen BestSolutionDSolutionif no TerminationConditionshave been met thenbegin

add Solutionto TabuListif TabuListis full then delete oldest entry from TabuListfind NewSolutionby some transformation on Solutionif no NewSolutionwas found orif no improved NewSolutionwas found for a long time then

generate NewSolution at randomif NewSolutionnot on TabuListthenSolutionD NewSolution

endelseDone=TRUE

until done=TRUE


11/18

Iterated Local Search


12/18


13/18

Variable Neighborhood Search


14/18

Guided Local Search


15/18

Smoothing Algorithm


16/18


17/18

Noisy Algorithm

Some Types of Noise

Uniform Noise

Logarithmic Noise

Others..


18/18

GRASP

Greedy Randomized Adaptive Search Procedure

single population met a heuristics

Documents