| Abstract: |  In the industrial world, critical success factors such as efficiency play an important role in reaching optimality. Technically, most of the time a company will be faced with the issue of problem dimension with respect to the constraints e.g. time and capacity. Therefore, reaching the efficiency aspect leads to the best achievement. However, the efficiency is not a trivial task since there are many real-world optimization problems which are difficult to be solved due to their high complexity. For example is combinatorial optimization where the search space possibly leads to exponential growth. We can employ exact mathematic to solve this, but then the main issue here is that it tends to unrealistic computation time. We need a method that can adjust to such this problem in the reasonable time. Thus, heuristics come as a new way dealing in the problem with considerable tradeoff regarding the achieved solution and the spent computation time. Further, the transition from heuristic to metaheuristic has come the important one. Generally, metaheuristic provide the generic strategies (abstraction) for solving arbitrary problems instead of some specific problems as heuristic deal with. Then this leads us to the question: How to choose the best parameter settings of metaheuritics strategy? The solution is to have the knowledge base who can suggest the company to decide the parameter settings of metaheuristic strategy. Technically, there are three main steps in this thesis. The first is to do the experiments of solving three benchmark problems with total 11 problem instances. There are 6 algorithms that are being employed. From the first step, the best algorithm including its parameter setting for the given problem can be analyzed. After that, the second step is to store the results of the experiments to the optimization knowledge base. The third step, which is the main objective of this thesis, is to perform fitness landscape analysis (FLA), and then characteristically cluster the benchmark problems and algorithms based on the features from FLA. The clusters will be confirmed and analyzed by the results of the experiments. So that one of the expectations is, in the future, the newly-found problems can be assigned to certain cluster which has a highly recommended parameter setting.
|
|---|