A Serial Hardware Structure Of Particle Swarm Optimization With Multi-Swarm Strategy

Particle Swarm Optimization(PSO) is a powerful optimization algorithm. Several improved PSO algorithms and Multi-Swarm PSO are proposed to further increase the search ability. Therefore, the system can deal with different kinds of problems efficiently.However, software implementation of Multi-Swarm PSO is not fast enough to be applied into real-time application. To solve the above problem, several hardware structures of PSO can be used to improve the calculation speed, which including the hardware/software co-design structure, the parallel hardware structure and the serial hardware structure. The serial strucuture can simultaneously achieve a higher speed and a lower hardware overheard.Based on the above analysis, a serial hardware structure of Multi-Swarm PSO is proposed in this research. The serial hardware structure can drastically decrease the calculation time compared with software implementation. Moreover, its hardware cost is within an acceptable range since all particles share the same calculation unit.The proposed serial hardware structure consists of three modules, which are the Calculation Unit, the Storage Unit and the Control Unit, respectively. The Calculation Unit is employed to calculate the position value and fitness value of particles. The Storage Unit is utilized to store the system parameters and calculation results. The Control Unit is designed to update the best value, control the algorithm flow and coordinate all modules in system. By employing pipeline structure, bit operation,ROM storage and other strategies, the calulaiton speed can be significantly improved. Moreover, through the cooperation of all modules, the proposed hardware structure can execute its function stably and figure out the optimal solution smoothly.According to the experimental results, the proposed structure can achieve 14-15 times improve ratio though the performance of software platform is better than hadware, and the hardware cost is acceptable. Thus, the proposed system can balance the stability and hardware overhead, and drastically improve the calculation speed.