基于差分进化粒子群混合优化算法的软测量建模

Soft sensor modeling based on differential evolution-particle swarm optimization based hybrid optimization algorithm

针对乙烯生产过程中,用传统方法难以直接完成对乙烯收率的在线测量的问题,提出了一种新型差分进化粒子群混合优化算法,建立了乙烯收率软测量建模。改进算法将优化过程分成两阶段,两分群分别采用粒子群算法和差分进化算法同时进行。迭代过程中引入进化速度因子进行算法局部收敛性判断,通过两个群体间的信息交流阻止算法陷入局部最优。对高维复杂函数寻优测试表明,算法的整体优化性能均强于基本粒子群算法和差分进化算法。应用结果表明,基于改进算法的软测量模型具有测量精度较高、泛化性能较好等优点。

In the process of ethylene production, ethylene yield cannot be measured on line via traditional approaches. To resolve this problem, a novel differential evolution （DE） particle swarm optimization （PSO） based hybrid optimization algorithm （DEPSO） was proposed. Then a soft sensor model for real time measuring ethylene yield was constructed. The procedure of optimization was divided into two phases and the particles were divided into two sub-swarms, one sub-swarm searched via PSO and the other searched via DE at the same time. Evolution speed factor was introduced in judging local convergence of algorithm during the process of iteration, with two sub-swarms exchanging information in each iteration to avoid local optimum. Optimization test on several complex functions with high-dimension indicated that the improved algorithm performed better than standard PSO and DE in whole optimization capability. Application results showed that the soft sensor model based on the improved algorithm had high measurement precision as well as good generalization ability.