Figure 4 - uploaded by André Restivo
Content may be subject to copyright.
Source publication
The optimization of parameter driven simulations has been the focus of many research papers. Algorithms like Hill Climbing, Tabu-Search and Simulated Annealing have been thoroughly discussed and analyzed. However, these algorithms do not take into account the fact that simulations can have dynamic scenarios. In this dissertation, the possibility of...
Context in source publication
Context 1
... all solutions from a certain simulation will produce nc sets of solutions: ( C → 1 , C → 2 , . . . , − C → nc ). A second step in the aggregation process would be taking each one of the sets and analyze if their scenarios also obey any kind of pattern. It is expected, from simulation scenarios, to be scattered without any scenario classes emerging. However, if we look at a solution class at a time there is a clear possibility that such classes appear. This happens because similar solutions are expected to solve similar scenarios. In Figure 4.10 we can see an example of a simulation with two input parameters ( p 1 and p 2 ) and two environmental parameters ( s 1 and s 2 ). As can be seen, the scenarios in this example are rather scattered but when we select only the scenarios solved by the solution class C we can see a scenario cluster emerging ( D ). It is also possible, that one class of solutions solves more than one class of scenarios, so the same clustering algorithms that can be applied in the Solution Aggregation problem (as has just been seen in Section 4.3.4.1) should, and can, also be applied in the Scenario Aggregation case. To implement the aggregation of scenarios, the K-Means clustering algorithm (explained in Section 3.6.1) was used. In order to implement this particular algorithm some choices had to be ...
Citations
... However, the system allows different simulation configurations to be used by means of system extensibility. The system is composed by several modules that will be introduced in the following paragraphs and are explained in full detail in [5]. To begin, a module that will interact with the different type of simulations was needed. ...
The application of optimization algorithms to parameter driven simulations and agents has been thoroughly explored in literature. However, classical optimization algorithms do not take into account the fact that simulations normally have dynamic scenarios. This paper analyzes the possibility of using the classical optimization methods, combined with clustering techniques, in order to optimize parameter driven agents, in simulations having dynamic scenarios. This will be accomplished by optimizing the agents in several random static scenarios and clustering the optimum results of each of these optimizations in order to find a set of typical solutions for the agent parametrization problem. These typical solutions can then be used in dynamic scenario simulations as references that will help the agents adapt to scenario changes. The results of this approach show that, in some cases, it is possible to improve the outcome of simulations in dynamic environments while still using the classical methods developed for static scenarios.
