交叉前置式粒子群优化算法及其在催化裂化C3含量软测量中的应用

摘要

An improved particle swaim optimization with pre-crossover(PSOPC) was proposed to avoid the premature convergence of particle swarm optimization algorithm. An auxiliary population was introduced in which the particles with low fitness but high diversity after each generation were stored.The pre-crossover between the particle in the swarm and the individual in the extra population was implemented, which helps increase the diversity of the particle swarm so as to improve the global convergence. PSOPC was used to train BP neural network to construct a soft-sensing model of C3 concentration of the fluid catalytic cracker unit (FCCU). The experimental results show that the model based on PSOPC and neural network has good precision and strong generalization.%针对粒子群优化算法易陷入局部极小点,出现早熟收敛的问题,本文提出了一种交叉前置式粒子群优化算法(PSOPC).该算法引入了一个辅助种群,将其中的个体与粒子群中的粒子在粒子更新之前进行预先的交叉操作.辅助种群所包含的是每次迭代后所生成的适应值较差但多样性较高的粒子.前置式交叉操作能够增加粒子群的多样性,有效改善算法的全局收敛能力.标准函数测试结果表明,PSOPC比基本PSO具有更好的优化性能.最后,将PSOPC应用于催化裂化装置干气中C3含量软测量建模,通过与实际的工业数据的对比,结果表明基于PSOPC的神经网络C3含量软测量模型具有较高的精度和较强的泛化能力.