基于AEPSO优化的神经网络多步预测控制

为提高神经网络预测控制的性能,提出了基于自适应扩展粒子群优化的神经网络预测控制方案。基本PSO算法中,每个粒子的更新受粒子个体极值和局部极值的影响,为了提高其全局收敛性,采用多粒子策略,使每个粒子的更新受更多其他粒子的影响;为提高收敛速度,采用自适应策略,对参数c0进行自适应调整,使c0随着迭代次数的增加而逐渐减小,这样,在PSO算法的搜索过程中,随着迭代次数的增加,搜索区域会越来越小,从而加快PSO算法收敛速度。运用该算法实现神经网络预测控制中的滚动优化,在有限时域内对控制序列进行寻列,给出基于粒子群优化的神经网络预测控制系统的稳定性证明。仿真结果表明,基于粒子群优化的神经网络预测控制系统具有良好的跟踪性能。

Multi-steps predictive control of neural networks based on AEPSO optimization

In order to enhance the index of neural network predictive control,a neural network predictive control method was provided by using a self-adaptive expand PSO algorithm.Each particle was refreshed by using individual and local extremum in the basic PSO algorithm.To improve the global convergence,particle was refreshed by multi-particle strategy;at the same time,the parameter c0 was self-adaptively adjusted and in such a way c0 gradually decreased with the increase of iterative time,as the searching area of the PSO algorithm was reduced and the convergence velocity of PSO algorithm was effectively improved.The improved PSO was used as the optimizer of the neural network predictive control and optimized the control serial in the finite time field.The stability of predictive control of neural network using improved PSO as optimizer is proved.Simulation results show that the method has better track performance.