دانلود کتاب طراحی الگوریتم های اکتشافی برای بهینه سازی سخت

This book is intended for students, teachers, engineers and scientists wishing to become familiar with metaheuristics. Frequently, metaheuristics are seen as an iterative master process guiding and modifying the operations of subordinate heuristics. As a result, the works in the field are organized into chapters, each presenting a metaheuristic, such as simulated annealing, tabu search, artificial ant colonies or genetic algorithms, to name only the best known. This book addresses metaheuristics from a new angle. It presents them as a set of basic principles that are combined with each other to design a heuristic algorithm. When addressing a new problem, we try to solve it by exploiting the knowledge acquired by experience. If the problem seems peculiarly difficult, a solution that is not necessarily the best possible is accepted. The matter is to discover the solution with a reasonable computational effort. Such a resolution method is then called a heuristic. By analysing a whole menagerie of metaheuristics proposed in the literature, we identified five major basic principles leading to the design of a new algorithm: 1. Problem modeling The most delicate phase when confronted with a new problem is its modeling. Indeed, if a problem is taken by the “wrong end,” its resolution can be largely compromised. Naturally, this phase is not the prerogative of metaheuristics. 2. Decomposition into sub-problem When one has to solve a complex problem or an instance of large size, it is necessary to decompose it into simpler or smaller sub-problems. These may themselves be difficult. Hence, they must be approached by an appropriate technique, for example a metaheuristic. 3. Building a solution When a suitable model is found, it becomes easy to build a solution to the problem, even if it is not good or even inapplicable in practice. One of the most common constructionmethods is a greedy algorithm, which may even provide exact solutions for simple problems such as the minimum spanning tree or the shortest path. 4. Modifying a solution The next step tries to improve a solution by applying slight modifications. This approach can be seen as a translation to the discrete world of gradient methods for differentiable optimization. 5. Randomization and learning Finally, the repetition of constructions or modifications makes it possible to improve the quality of the solutions produced, provided that a random component and/or a learning process are involved.