利用串/并联神经网络建立化学反应器模型

Modeling of a chemical reactor with serial/parallel neural networks

研究了使用线性动态神经网络与非线性的静态网络相结合的混合建模方式解决复杂非线性系统的建模问题。使用混合神经网络建模,可以降低单个网络的训练难度,基于此,也可将非线性系统控制策略的求解分解,转换为线性系统的求解。从而改善使用单一神经网络建模存在的精度不高以及训练时间长等不足,也为非线性系统控制策略的求解提供方便。本文以一个典型多变量系统———连续搅拌釜式反应器(CSTR)作为仿真对象,详细研究和实现了两类神经网络串联和并联的混合建模方法,并对结果进行了比较。

Modeling studies of complex nonlinear systems by hybrid neural networks are described in this paper. The hybrid neural networks consist of linear dynamic neural networks and nonlinear static neural networks. Using such a hybrid neural network, the difficulties in training a single network can be decreased, and the solution for a nonlinear control strategy can be reduced to solving for a linear system. Thus, this method overcomes the shortcomings of long training time and lower accuracy observed for one single neural network. Both the serial and parallel neural networks have been used to model a CSTR. A comparison between the two hybrid neural networks on the basis of the results obtained is described.