Abstract: The particle swarm algorithm of stochastic optimization（PSO） only uses fitness function to judge metrics of the found solution, but does not evaluate the overall swarm status. This causes local convergence of the algorithm after a certain times of iterations. The optimized algorithm BPPSO uses BP neural network to classify swarm status, and performs different disturbance operations to swarm according to the divide result. It is the algorithm optimization of swarm perspective. Simulation experiment results show that the BPPSO algorithm can increase swarm variety, and improve optimization accuracy. It solves the QoS routing problem of Ad Hoc network better, and proves feasibility and validity of the proposed algorithm.

Key words: Particle Swarm Optimization（PSO） algorithm, premature convergence, Back Propagation（BP） neural network, QoS routing

摘要： 随机优化的粒子群算法（PSO）在解决待优化问题时，仅利用适应度函数对单个粒子所找到解的优劣进行判断，缺乏对种群总体状态的评估，导致算法经过一定次数的迭代后陷入局部收敛。改进算法BPPSO利用BP神经网络对种群进行状态划分，并根据划分结果对种群实施相应的扰动操作，从种群的角度对算法进行改进。仿真实验表明，改进算法能够增加种群多样性，提高优化精度，较好地解决了Ad Hoc网络的QoS路由问题，从而验证了所提算法的可行性和有效性。

关键词: 粒子群优化（PSO）算法, 早熟收敛, 向后传播（BP）神经网络, QoS路由