摘要
提出了基于量子理论的连续粒子群优化(Continuous Particle Swarm Optimization based on Quantum Methodology,CPSO-QM)算法,主要是采用了量子理论中的叠加态特性和概率表达特性.其中,叠加态特性可以使单个粒子表达更多的状态,潜在地增加了种群的多样性;概率表达特性是将粒子的状态以一定的概率表达出来.在基准函数的实验测试中,对比其它常用算法,结果显示本文提出的算法性能较好.在实际应用中,以丙烯腈反应器作为建模研究对象,提出了三种进化策略,实验结果显示,这三种策略训练的神经网络软测量模型都可以较好地预测丙烯腈的收率.
Continuous particle swarm optimization algorithm based on quantum methodology is propesed in the paper. The algorithm mainly uses superposition characteristic and probability representation. Superposition characteristic can make a single particle present several states. In other words, the characteristic potentially increases population diversity. Probability representation is to make particle＇s state be presented according to a certain probability. Compared with other methods for test function in the experiment, the results demonstrate the proposed algorithm is better and more effective. Additionally, acrylonitrile reactor is used as modeling object in the real application. Three evolutional schemes are used. The experimental results show that the networks trained can the better predict the acrylonitrile yield through using three evolutional schemes.
出处
《系统工程理论与实践》
EI
CSCD
北大核心
2008年第5期122-130,共9页
Systems Engineering-Theory & Practice
关键词
进化算法
粒子群
量子计算
软测量模型
evolutionary algorithm
particle swarm
quantum computing
soft sensing model
