Genetic algorithm with Particle Filter for dynamic optimization problems

The optimization problem that the optimum is time-changing by following a motion law in the search space is a dynamic optimization problem. This paper introduces the optimum´s motion information to the proposed algorithms. Particle Filter is used to predict and track the changing optima. In real solution space, GA´s chromosome is the same as Particle Filter´s particle, both of which can be regarded as candidate solution. It is convenient to exchange information from the both. Two algorithms are designed to introduce the predicted particles of Particle Filter to genetic algorithm. The predicted particles serve as good genetic materials for GA in dealing with dynamic optimization problem and the optima which GA obtains in the stationary phase are viewed as observations to system state for Particle Filter. Both Particle Filter and genetic algorithm form the feedback loop and enhance the proposed algorithms´ ability of tracking the optimum. Experimental study over DF1 benchmark dynamic problem shows that the algorithms have good performance.