一种基于非线性滤波的倒立摆控制方法研究

对于倒立摆这样的强非线性系统,采用传统的BP算法存在着收敛速度慢、易陷入局部极小值的缺陷,而采用卡尔曼滤波方法则会带来很大的模型误差。为了解决上述问题,提出了基于粒子滤波优化神经网络的方法。首先建立了倒立摆神经网络控制器的物理模型并将模型粒子化,而后用粒子滤波算法对粒子进行优化估计,将估计结果作为网络的权值应用到倒立摆控制中,采用离线训练方式,仿真比较了卡尔曼滤波和粒子滤波两种方法控制效果,结果表明,新算法较卡尔曼滤波方法在控制性能上有明显提高。

Study of Inverted Pendulum Control Method Based on Nonlinear Filter

The inverted pendulum model is a strong nonlinear system.Using BP algorithm to controll inverted pendulum has some drawbacks such as falling into local minima and slow convergence.And using Kalman filter will lead to huge linearized error.In order to solve those problems,a method called optimize neural network based on particle filter is introduced.The physical model of the inverted pendulum controller is founded,and particle filter is to used estimate the neural network parameters.Then those parameters are applied as the weights of the neural network to controll the inverted pendulum.The control effect comparison between Kalman filter and particle filter are obtained in the mode of off-line.The simulation result shows that the performance of particle filter improves markedly than Kalman filter both on speed and precision.