【作者】 曹海云；

【导师】 李守巨；

【作者基本信息】 大连理工大学 ， 计算力学， 2008， 硕士

【摘要】 倒立摆控制系统通常用来检验控制策略的效果，是自动控制、机械电子等领域中非常典型的较为理想的实验装置。倒立摆系统是一个非线性、强耦合、多变量和自然不稳定的系统，这就需要给倒立摆系统施加适当的力使倒立摆系统在一个很小的范围内移动，从而使摆的摆角在很小的范围内变化。根据神经网络的自组织、自适应和泛化学习能力，应用神经网络解决倒立摆控制问题。BP神经网络在计算的过程中，需要调节其连接权系数，一般情况下采用梯度下降法来调节其权系。人工神经元网络为非线性系统建模提供了快速、简便的学习能力和应用性能，以及精确表达过程的非线性-、复杂性的能力。其中多层前馈神经元网络以其结构直观、算法简便和理论上能逼近任意非线性连续映射的能力而得到了最广泛的应用。但是经典的BP算法存在收敛速度慢、数值稳定性差等缺点，为此许多学者将非线性优化理论中的许多算法引入到多层前馈网络的训练学习中，起了一定的改善作用。拟牛顿算法中的BFGS优化算法在求解无约束优化问题中被认为是最好的方法。论文采用BFGS方法来训练BP神经网络的权值，从而达到控制倒立摆系统的目的。通过对倒立摆系统和离散时间系统的模拟应用，表明所提出方法的控制效率和有效性。模拟结果表明所提出的非线性系统控制方法给出了更精确的全局最优解和更快速的收敛速率。更多还原

【Abstract】 Inverted pendulum control system which is a kind of perfect equipment in autocontrol, mechanism and electron is used to test the result of control. Inverted pendulum system is a nonlinear, coupling, variable and erratic system. In order to ensure the swing angle and the position of dolly to chang in a small range, a proper force must be applied to the inverted pendulum system. Neural network can be used to solve the problem of inverted pendulum control because of its ability of sel-organization, sel-adaption and generalization. BP neural network can work out how much force applied to the dolly through gradient degressive algorithm to adjust the joint weights.Artificial neural networks offer speediness, simple and convenient learing ability and application performance for nonlinear system model. Multilayer feedforward neural networks are used widely because of its intuitionistic structure, simple and convenient algorithm and the ability to approach random nonlinear continuum mapping. However, BP neural network lies the problem of slow convergence speed and the bad numerical value stability and so on. A lot of algorithms in nonlinear optimization theory are used into the learning of multilayer feedforward neural networks, and get some improvement. This article uses the BFGS optimization method which is considered to be the best optimization algorithm in solving nonrestrain question, to train the weight of BP neural network.The control of nonlinear system is applied to some simulation examples of a discrete-time system and the inverted pendulum model system to demonstrate the performance and control efficiency of the proposed method. The simulation results show that the suggested methods give better global convergent characteristics and faster convergence.更多还原