基于QPSO的小波神经网络控制器

Wavelet Neural Network Controller Based on Improved QPSO

通过理论分析基本PSO-WNN算法,在收敛速度慢和泛化性能低等方面自身有很明显的缺陷。根据单个粒子稳定收敛这一条件,提出一种改进的粒子群算法,并利用这种改进的算法来重新训练,重新得到一组小波神经网络权值,这样就能建立一种高效的粒子群小波神经网络控制器。最后进行测试,可以比较改进后的算法与基础POS-WNN,得出网络更容易全局收敛,函数逼近误差变小,迭代次数减少,分类精度升高。最后根据二级倒立摆的特点,设计小波神经网络控制器模块,对实验结果进行仿真研究,从而验证了小波神经网络控制器的有效性。

Through theoretical analysis of the basic PSO-WNN algorithm, it has obvious defects of the slow rate of convergence and low generalization performance. According to stable convergence condition of the individual particles, an improved particle swarm optimization, and use of this improved algorithm to re-train and re-get a group of wavelet neural network weights are put forward, so that an efficient particle swarm wavelet neural network controller can be built. By last test, we can compare the improved algorithm and the basic POS-WNN to come to the network global convergence easily, the function approximation error becomes smaller, the number of iterations is reduced, the classification accuracy is increased. Finally, according to the characteristics of the double inverted pendulum, wavelet neural network controller module is designed, the experimental results of simulation is studied and the validity of the wavelet neural network controller verified.