Metaheuristic Techniques

Most real-world search and optimization problems involve complexities such as non-convexity, nonlinearities, discontinuities, mixed nature of variables, multiple disciplines and large dimensionality, a combination of which renders classical provable algorithms to be either ineffective, impractical or inapplicable. There do not exist any known mathematically motivated algorithms for finding the optimal solution for all such problems in a limited computational time. Thus, in order to solve such problems to practicality, search and optimization algorithms are usually developed using certain heuristics that though lacking in strong mathematical foundations, are nevertheless good at reaching an approximate solution in a reasonable amount of time. These so-called metaheuristic methods do not guarantee finding the exact optimal solution, but can lead to a near-optimal solution in a computationally efficient manner. Due to this practical appeal combined with their ease of implementation, metaheuristic methodologies are gaining popularity in several application domains. Most metaheuristic methods are stochastic in nature and mimic a natural, physical or biological principle resembling a search or an optimization process. In this paper, we discuss a number of such methodologies, specifically evolutionary algorithms, such as genetic algorithms and evolution strategy, particle swarm optimization, ant colony optimization, bee colony optimization, simulated annealing and a host of other methods. Many metaheuristic methodologies are being proposed by researchers all over the world on a regular basis. It therefore becomes important to unify them to understand common features of different metaheuristic methods and simultaneously to study fundamental differences between them. Hopefully, such endeavors will eventually allow a user to choose the most appropriate metaheuristic method for the problem at hand.