改进的粒子群神经网络在飞控中的应用  

Application of Improved Particle Swarm Optimization and Neural Network to Flight Control System

作　　者：梁坤[1,2] 宋书中[1] 马建伟[1] 

机构地区：[1]河南科技大学电子信息工程学院 [2]解放军92492部队

出　　处：《火力与指挥控制》2013年第6期121-124,共4页Fire Control & Command Control

基　　金：航空科学基金资助项目(20110142002)

摘　　要：针对BP神经网络对初始值敏感、容易陷入局部寻优且收敛速度较慢,提出用粒子群对神经网络的参数进行优化,同时设计了衰减的指数函数对惯性权重进行动态调整以提高算法性能。并应用该算法对导弹飞控系统的逆误差进行补偿,仿真结果表明,该方法对逆误差进行了有效的补偿,避免了局部寻优并提高了学习效率。Because of the disadvantages of Back-Propagation （BP） Neural Network （NN）in its easily being trapped into local maxima,sensitive to the initial value and slow convergence speed, Particle Swarm Optimization （PSO） was introduced to optimize the parameters of the BP NN. Meanwhile,the inertial weight was regulated dynamically by the decayed exponential function to enhance the performance of the PSO. The adaptive NN controller with modified PSO above is applied to the missile control system to account for the dynamic inverse error. The simulation results prove that the compensation of the inverse error is effective and the new method avoids the neural network of local optimization and improved the learning efficiency.

关 键 词：神经网络 粒子群算法 指数函数 惯性权重 导弹 

分 类 号：TJ765.2[兵器科学与技术—武器系统与运用工程]