基于多维问题的交叉算子量子粒子群优化算法

Quantum-behaved particle swarm optimization algorithm with crossover operator to multi-dimension problems

针对量子行为粒子群优化(QPSO)算法在求解多维问题时优秀维信息丢失的问题,引入交叉算子的策略,改善解的质量,提升算法性能。首先,分析了量子粒子群算法进化过程中的粒子整体更新评价策略,发现各维信息之间相互干扰,会丢失已经搜索到的优秀维信息;然后,指出如果采用逐维进化方法,会指数级增加算法的复杂度;最后,提出对进化过程中的问题解采用多点交叉的策略增加优秀维信息的保留概率,并将改进后的量子粒子群算法与线性下降参数控制策略、非线性下降参数控制策略方法通过12个CEC2005 benchmark测试函数进行了比较,并对结果进行了分析。仿真结果显示,所提算法比改进前在10个测试函数中取得了明显的改进效果,而比其他2种改进算法也在7个测试函数中取得了优势。因此该算法能够有效提升量子粒子群优化算法的性能。

According to the problem that better dimensions information of particles will loss in Quantum-behaved Particle Swarm Optimization （QPSO） algorithm when solving multi-dimensions problems, a strategy with crossover operator was introduced and the quality of solutions and the performance of algorithm would be improved. Firstly, the whole update and evaluation strategy on solutions in algorithm was analyzed and the better dimensions information of particles would loss because of the mutual interference between dimensions. Secondly, when the evolution was executed dimension by dimension, the algorithm complexity would increase exponentially. Finally, multi-crossover method was employed to increase the retaining probability of excellent dimension information. The comparison and analysis results of the proposed method, with linearly decreased coefficient control method and non-linearly decreased coefficient control method on 12 CEC2005 benchmark functions were given. The simulation results show the modified algorithm can greatly improve the QPSO performance compared with the basic QPSO in 10 functions and also get better performance in 7 functions compared with the other two QPSO variants. Therefore, the proposed method can improve the performance of QPSO effectively.