Abstract for: Genetic Algorithms for Multi-Objective Optimization in Dynamic Systems
This study uses parametric search to meet multiple goals in the behavior of dynamic systems. Parameters are searched using genetic algorithm. Main aim of this study is to discuss how multi-objective parameter search gives essential information about the system. A nonlinear electric circuit is one of the two dynamic models in this paper used for parameter optimization. The electric circuit model shows oscillatory behavior. A fitness function which evaluates period and amplitude and compares it with the desired oscillatory pattern is proposed. It is shown that time horizon for simulation based optimization can be crucial. The second model is a generic System Dynamics model, the stock management problem with second order supply line. The policy parameters are weight of stock adjustment and supply line adjustment. A fitness function that evaluates the settling time, overshoot, and steady state error is proposed. The search results provide some insight on both the fitness function and the system. The obtained results are satisfactory and they show that the response time of the system can be decreased by small overshoot. The paper is a step towards simulation based parameter search becoming an essential support toolbox for model building and policy design in System Dynamics.