The concept that forms the basis for a simulated annealing solution generation process centers on the cooling of materials in a heat bath, a process known as annealing.
Simulated annealing, as a solution generation (search) process, is not considered a traditional mathematical solution generation process, but rather a heuristic. It relies on a set of logic to iteratively adjust a solution, allocating and reallocating resources to various uses, until a very good solution to a problem has been located. The process of arriving at the optimal design for a system by annealing has been suggested as an example of an evolutionary solution process that is modeled by statistical means. The concepts involved in using simulated annealing for solving management problems suggest that the search will not become trapped in local optima; however, locating the global optimum solution is problematic, and the most we can say about the solutions generated for complex problems is that what is reported as the 'best solution' is the best local optima that was found during the search.
While the heuristic process does not guarantee that the global optimal solution to a problem has been located, the advantage of using the heuristic is that a good solution to a difficult problem can be located rather quickly. Traditional mathematical programming techniques that involve linear equations and binary decision variables, such as mixed integer or integer programming, use special heuristics, such as the cutting plane or branch and bound methods, to find the optimal solution of problems that are not too large. Within these special heuristics, linear programming is used as a subroutine. Two issues related to traditional mathematical programming techniques are important: mixed integer and integer programming may require a significant amount of time to generate an optimal solution to a complex planning problem, and there are no known mathematical solution processes that can guarantee optimal solutions to general nonlinear optimization problems, other than complete enumeration. The commonly cited strengths of simulated annealing compared to other heuristics such as tabu search and genetic algorithms are the ease of implementation and the relative insensitivity of the solution time to problem size. Closely related heuristics to simulated annealing include threshold accepting, record-to-record travel, and the great deluge - the main difference being the criteria for accepting a nonimproving solution in order to avoid becoming trapped in local optima.
Was this article helpful?