基于符号跟随系统的无源控制设计

研究了基于符号跟随系统(SFS)的无源控制设计问题,讨论了该控制方案的适用范围.通过构造出的一个反例,说明了并联一般的无源系统后SFS无法保证实现系统的无源化.进一步分析了引入SFS后的单位负反馈可能导致闭环系统病态的问题,由此给出了在反馈通道中引入严格真无源系统的控制方案,得到了基于无源系统理论的输出反馈控制,并作了相应的特性分析.仿真以非最小相位系统为例,所得结果表明了该控制方案的有效性.SFS的引入拓宽了经典控制方案中参数的可调范围,可以较好地改善闭环系统的动态性能.     

Adaptive Numerical Approach for Optimal Control of a Single Train

This paper considers the optimal control problem of a single train, which is formulated as an optimal control problem of nonlinear systems with switching controller. The switching sequence and the switching time are decision variables to be chosen optimally. Generally speaking, it is very difficult to solve this problem analytically due to its nonlinear nature, the complexity of the controller,and the existence of system state and control input constraints. To obtain the numerical solution, by introducing binary functions for every value of the control input, relaxing the binary functions, and imposing a penalty function on the relaxation, the problem is transformed into a parameter optimization problem, which can be efficiently solved by using any gradient-based numerical approach. Then, the authors propose an adaptive numerical approach to solve this problem. Convergence results indicate that any optimal solution of the parameter optimization problem is also an optimal solution of the original problem. Finally, an optimal control problem of a single train illustrates that the adaptive numerical approach proposed by us is less time-consuming and obtains a better cost function value than the existing approaches.     

一类不确定非线性系统的鲁棒自适应跟踪

讨论了一类含线性未知参数和未建模动态的不确定非线性系统的自适应跟踪问题,通过backstepping过程给出了自适应控制规律。在该控制作用下,所有状态信号和控制信号都是有界的,且可保证未建模动态对输出的影响任意地小。     

基于AP和BP神经网络算法的手写数字识别

针对现有的手写数字识别技术不适合大规模应用的问题,提出了一种基于AP和BP神经网络的快速手写数字识别算法。首先对预处理后的样本通过AP算法(affinity propagation)聚类消除冗余,重新构造样本空间;然后构造BP(误差反向传播)神经网络模型,学习测试集合样本。采用UCI机器学习数据库中的数据进行实验,结果表明,算法的识别正确率可达96.10%,高于BP神经网络算法的识别正确率94.88%,且执行时间约为后者的10%,具有较高的实用价值。     

基于误差函数的自然梯度法及其在MLP学习中的应用

在激活函数选为误差函数的条件下,给出了求Fisher信息矩阵逆的算法,并进行仿真验证.首先将自然梯度法应用于多层感知器学习中,并与BP(back-propagation)算法进行了比较,说明了自然梯度法能够加快学习速度,并且可能避免学习中出现平坦区现象,即使出现平坦区,也不会很严重.仿真验证了自然梯度法上述良好的学习性...