改进的PSO算法及其在软测量建模中的应用

Improved Particle Swarm Optimization and Application in Soft Sensor Modeling

通过大量仿真实验,考察了粒子群算法(PSO)中粒子平均速度和算法收敛性之间的关系,提出了一种基于粒子速度反馈信息的自适应调整权重策略,同时在搜索过程中引入混沌序列。给出的收敛性分析证明,该算法可以以概率1收敛到全局最优解。对经典函数的测试计算表明,改进后的PSO算法较好地解决了基本粒子群算法中易陷入局部最优的缺点,在稳定性和收敛精度上均优于普通的PSO算法。改进的粒子群算法被用于优化神经网络的结构和参数,并将基于改进算法的神经网络用于4_CBA软测量建模中。实际应用表明,与基于其它智能算法的神经网络相比,该网络不仅有较高的泛化性能,而且有更快的学习速度和较好的实时性。

The relationship between swarm average velocity and convergence was studied in particle swarm optimization （PSO） through numerical simulations, and then a novel adaptive inertia weight particle swarm algorithm based on the particle velocity feedback （VPSO） was proposed, simultaneously, chaotic series was introduced in search. Then, the convergence analysis was given, which represents that the algorithm can converge to overall optimum. The experimental results of classic functions show that the advanced PSO algorithm solves the defect of PSO algorithm which is apt to trap in local minimums, and has great advantage of convergence property and robustness compared to PSO algorithm. In the last, the improved PSO was applied to optimize the structure and parameters in neural network （NN）. Then, the NN was applied in soft sensor modeling of 4-CBA. The experiments demonstrate that the neural network based on improved PSO has faster learning speed and better real time ability compared to other NN.