过热汽温的模拟柔性神经网络直接自适应控制系统

提出一种模拟复合正交柔性神经网络，并应用于电厂过热汽温的直接自适应控制方法。模拟柔性神经网络被用作为过热汽温控制系统中的主、副调节器，以改善控制系统的动态性能。神经网络采用5层网络结构，网络隐层节点采用带参数的Sigmoid函数构成的Laguerre（拉盖尔）复合正交多项式。网络在学习过程中，输入层与隐层之间不用调整权值，仅连续调整输出层与隐层之间的权值和隐层中Sigmoid函数的参数，以提高网络的学习适应性。网络隐层节点（处理元）是复合正交多项式的展开项，展开项的多少决定着网络的学习速度和精度。通过对具有严重参数不确定性、扰动以及大迟延的电厂过热汽温被控对象进行仿真研究，结果表明控制系统的干扰和超调明显减小，在控制系统的动态性能上，所提控制方法优于常规控制方法。电厂过热汽温控制取得了令人满意的效果。

Direct Adaptive Control System of Analog Flexible Neural Networks for Super-heated Steam Temperature

A kind of analog flexible neural network of compound orthogonal type and its application method for a direct adaptive control of super-heated steam temperature in a power plant were presented.Herein,the analog flexible neural networks were developed to employ main and minor adjustors in the control system of super-heated steam temperature to improve dynamic performances of the control system. The neural networks use the three-layer network structure.Nodes in the hidden layer use the Laguerre compound orthogonal polynomials composed of sigmoid functions with a parameter.In the network learning process,there are no weight adjustment between the input and hidden layers,then the weight values between only the hidden layer and the output layer and the parameter of Sigmoid functions in the hidden layer are adjusted continuously to improve the network learning adaptability.A node (a processing element) in the hidden layer of the network is an expansion term of the Laguerre polynomials. More or less terms determine the network learning speed and accuracy.Simulation for the super -heated steam temperature in a power plant was carried out under such a control condition that had a severe uncertainty of parameters and disturbance,as well as a large time-delay.The results show that control system can limit the disturbance and overshoot dramatically and the proposed control method is superior to conventional control ones in dynamic performances of the control system.The control of super-heated steam temperature in a power plant obtains satisfactory results.