基于群体信息挖掘的协同差分进化算法及其应用

Cooperative coevolutionary differential evolution algorithm based on colony information mining and its application

为了提高差分进化算法的寻优速度和寻优效能,提出了一种基于群体信息挖掘的协同差分进化算法。该算法首先利用协同进化的思想,将种群分解成不同的子种群,每个子种群采用不同的差分策略进行独立的差分和交叉操作,再将各个子种群合并成一个种群,并根据每个个体的适应度值挑选出最优个体;为了提高差分进化算法的局部搜索能力,引入了多元回归分析和模式搜索算法,对于整个种群,利用最小二乘法求得种群的寻优方向信息,并以此来指导种群中的每一个个体进行模式搜索。仿真试验和在精对苯二甲酸生产过程对羧基苯甲醛含量软测量模型参数估计中的实际应用表明:该算法的性能比传统的差分进化算法有较大的提高,取得了较好的效果。

To solve the defect of poor search ability and bad precision of optimal result of differential evolution algorithm, a cooperative co-evolutionary differential evolution algorithm based on colony information mining （CCDE） was proposed. It first split the population into several sub-populations based on cooperative co-evolution. Each sub-population did the differential operations and crossover operations individually, using different differential strategies. Then it combined all the sub-populations into a whole population, and picked out one best individual depending on the fitness of each individual. The multiple regression analysis and pattern search algorithm were joined into the algorithm to improve the local search ability. It used the information of search direction gained by the least square to guide each individual of the entire colony to do the pattern search. The simulation experiment and the application in parameter estimation of 4-carboxybenzaldehyde content soft sensor in the production of pure terephthalic acid showed that the performance of the algorithm was much better than differential evolution algorithm and the result was good.