Abstract: The paper proposes an improved adaptive Particle Swarm Optimization（PSO） algorithm which integrates gradient search method, breeding method and the all-time optimal location information of the first N particles. With chaotic inertia weight, the renewal of each particle’s speed and position is considered with not only the information of its own all-time and global optimal location information, but also the information of other all-time optimal location information, while the weights of other particles’ all-time optimal location information change adaptively with the number of iterations; meanwhile, particles location updates in their negative gradient direction and the particle locations increase linearly with iterations; when the particles stop updating, cross may be hold in local optimum position. The experimental results verify that the algorithm has better convergence speed and convergence precision than those relevant algorithms.

Key words: Particle Swarm Optimization（PSO） algorithm, gradient search, breeding method, self-adaption, inertia weight

摘要： 提出了一种融合梯度搜索法、繁殖法并结合前[N]个粒子历史最优位置的改进自适应粒子群优化算法。算法选用混沌惯性权重，每个粒子速度和位置的更新不仅考虑自身历史最优和全局最优位置，还受其他粒子历史最优位置的影响，且其影响程度的权重随迭代次数自适应变化；同时粒子位置随迭代次数以线性递增的概率进行负梯度方向更新；当粒子更新停滞时，对可能处于局部最优位置的部分粒子进行杂交。仿真实验结果表明，该算法比其他相关算法具有更好的收敛速度和收敛精度。

关键词: 粒子群优化算法, 梯度搜索, 繁殖法, 自适应, 惯性权重