Particle Swarm Optimization In Dynamic Optimization In Applied Research,

Many optimization problems are often dynamic in real world, their optimization parameters, constraints, optimization objectives etc are time-varying. These problems require optimization algorithm to detect changes accurately at all times and response timely after environment changes. These two aspects are the main foundations to evaluate algorithm's performance in dynamic environment and also the main handles to apply algorithm to dynamic optimization. Particle Swarm Optimization algorithm model is simple, good and robust. It is an effective method to solve complex, nonlinear problems. Therefore, improving Particle Swarm Optimization algorithm model and making it more adaptive to time-varying optimization problem are practically significant to engineering applications. To this end, this dissertation focuses on the characteristics of Particle Swarm Algorithm which are suitable for solving dynamic optimization problem, and makes them improved, then improves the algorithm's capability of tracking, detection and response for changes in dynamic environment.Firstly, for improving Particle Swarm Optimization algorithm's capability of detection and response for changes, the dissertation selects some detective particles at random, through the changes information of their fitness to detect changes in environment; at the same time, chaotic mutation has been introduced as responded strategy. However, because of the algorithm responds too frequently, wasting a lot of computer resources, also responding too frequently makes the convergence rate of the algorithm greatly reduced. Following, it introduces swarm diversity to control the responded frequency adaptively. The simulative results of single-mode high-dimensional dynamic environment show that the improved algorithm can detect changes more accurately, and respond more quickly.Later, for multi-mode complex dynamic optimization problems, the dissertation proposes an improved particle swarm optimization model. Through using chaotic sequences to make swarm distributing evenly, then the algorithm's performance has been improved in early evolutionary. For the characteristics of abstract dynamic environment model are time-varying, the dissertation uses different abstract environment models in different time to evaluate particles in same time, through the changes information of global best location and global best fitness to determine changes in environment. At the same time, the dissertation uses swarm diversity and the distance between particle's current position and the global best location in next moment as a responded condition. These combine with the method of improved resetting evolutional orientation, not only be able to track extreme point after the changes timely, but also to maintain a good balance between exploitation and exploration. So that particle swarm optimization algorithm has been further improved for complex dynamic optimization problems.