Browsing by Author "Teklay, Ashenafi"

Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    Systematic Evolutionary Algorithm for a general Multilevel Stackelberg Problems with bounded decision variables (SEAMSP
    (Addis Ababa University, 2013-07-01) Teklay, Ashenafi; Mitiku, Semu
    Multilevel Stackelberg Problems (MSPs) are nested optimization problems which reply hierarchical decisions of subproblems. Each decision maker (DM) in the hierarchy admits the decision of those above its level (if exist), observes the response of those be- low (if exist) for each possible value of its decision variable and returns the best variable value/s of its interest. These kind of problems are known to be common in distinct areas of study. Linear MSPs are shown to be NP-hard problems by di_erent authors. The inclusion of non-linear, non-convex, non-di_erentiable and other undisciplined property of functions add further complexity to the problem. Unfortunately, real life situations are crowded with such kind of functions. Most existing algorithms in MSPs are proposed for bilevel stackelberg problems (BSPs), specially the linear version of BSPs. Systematic evolutionary algorithm for a general multilevel stackelberg problems (SEAMSP) having bounded decision spaces, has been proposed in this work. A unique feature of the algo- rithm is that it is not a_ected by the behavior of the objective and constraint functions involved in a problem. The proposed algorithm apply evolutionary algorithm concepts to MSPs, with systematic way of selecting initial populations at each iteration and a newly constructed mutation operator, which is suitable to the selection of populations. In SEAMSP, each decision space is controlled by \intelligent spies" having a nice cooperation with the spies on the other decision spaces, for representing the whole con- straint region in a random, unique, diverse and systematic way. The numerical results on various problems demonstrated that the proposed algorithm is very much promis- ing to MSPs without any limitation of the included functions, and it can be used as a benchmark for a comparison of approximate results by other algorithms