|
Pattern recognition methods,capable of advanced decision making without a precise description of the pro-cess model,have been used in many cases to implement advanced control systems.Such techniques have been ap-plied to structural identification of nonlinear systems and distributed parameter systems,direct control of systemsoperating in unfamiliar environments,and dynamic modelling,state estimation,optimizing,fault diagnosis,predictive control,self-tuning control,intelligent control and so on.An account of... Pattern recognition methods,capable of advanced decision making without a precise description of the pro-cess model,have been used in many cases to implement advanced control systems.Such techniques have been ap-plied to structural identification of nonlinear systems and distributed parameter systems,direct control of systemsoperating in unfamiliar environments,and dynamic modelling,state estimation,optimizing,fault diagnosis,predictive control,self-tuning control,intelligent control and so on.An account of these techniques and their ap-plication to controls is given in this paper.Possible new applications of pattern recognition to process control sys-tems are suggested. 模式识别方法本质上是一种分类方法,它不需要过程的数学模型就可做出高级决策。模式识别方法已在过程控制中的许多领域得到了应用,如非线性和分布参数系统的结构辨识、在未知环境下运行系统的控制、自适应控制、智能控制、动态建模、状态估计、优化控制、故障诊断等。本文将概述模式识别方法在以上各方面(?)情况,并指出今后可能的研究方向。 This paper presents a summarized definition of artificial neural network and its research method. The basis of its development and the relation between the basis and the information processing are also dealt with. Attamp is made theoretically to enable the processing system designer to be aware of the relation on mathematical essenre between the network and the processing,so that he can know what are the very problems to be solved by means of this network technique for more ideal results. 本文阐述了人工神经网络的概括定义、研究方法、发展的基础以及这些基础与信息处理的关系,试图从思想或概念上使信息处理系统的设计者懂得在使用人工神经网络技术设计信息处理系统时,究竟那些特性总是实质性的和必不可少的;并了解人工神经网络与信息处理在数学本质上的联系,进而让人们知道究竟那些问题更适于用人工神经网络技术来解决。 The learning procedure of neural networks can be regarded as a problem ofestimation (or identifying) parameters with a nonlinear or linear observation equation. Allalgorithms having been put up were assumed to have constant learning rates or constantaccelerative. In this paper, making use of Kalman filtering, we derive a new back-propaga-tion algorithm whose learning rate is computed by Riccati difference equation. 神经网络的学习过程本质上可以看成是一组线性或非线性观测方程的参数估值或参数识别问题。在已经提出的许多学习算法中,学习速度或者加速度常常假定为定值。本文运用卡尔曼滤波原理,提出了一种新的神经网络学习算法。该算法的学习速度是由带时间参数的里卡蒂微分方程来确定的。
|