论文部分内容阅读
Hammerstein模型是化工过程中最常用的模型之一,它由非线性静态环节和线性动态环节串连组成,适合描述pH过程和具有幂函数、死区、开关等非线性特性的过程.这类模型的控制问题可以分解为:线性模型的控制问题和非线性模型的求根问题.针对Hammerstein模型提出了一种基于神经网络的模型预测控制策略,采用一组神经网络拟合非线性部分的逆映射.这种方法不需要假设Hammerstein模型的非线性部分由多项式构成,并且避免已有研究在无根和重根情况下存在的问题.最后通过仿真试验证明了以上结论.
The Hammerstein model is one of the most commonly used models in chemical processes. It consists of nonlinear static links and linear dynamic links in series and is suitable for describing pH processes and nonlinear processes such as power functions, dead zones, switches, etc. Such models The control problem can be decomposed into: the control problem of the linear model and the root of the nonlinear model. Aiming at the Hammerstein model, a neural network-based model predictive control strategy is proposed. A neural network is used to fit the inverse of the nonlinear part This method does not need to assume that the nonlinear part of the Hammerstein model is composed of polynomials, and avoids the existing problems in the absence of roots and heavy roots. Finally, the above conclusion is proved through simulation experiments.