无辨识自适应算法的大滞后对象的控制方法

在实际工业生产中 ,由于对象纯滞后的存在 ,降低了控制系统的稳定性 ,使控制品质下降 ,对控制系统极为不利。对大滞后对象 ,Smith预估控制是一种重要方法 ,但常规Smith预估控制对模型的误差 (包括时间延迟的估计误差 )十分敏感 ,不适用于具有时变时延参数的系统。因此 ,常规Smith预估控制策略难以广泛用于工业控制 ,但该方法仍然得到了控制界的广泛认可。无辨识自适应控制是Marsik和Strejc提出的一种无需辨识系统参数的自适应控制算法 ,该算法简单、鲁棒性强 ,只需在线检测过程的实际输出及期望输出便可形成具有较好动态性能指标的自适应控制系统 ,但是该方法不能解决大滞后问题。借鉴无辨识自适应控制的思想和神经网络强的函数逼近能力 ,首先用一个神经网络来构成被控对象的Smith预估模型 ,然后利用无辨识自适应控制算法设计了一种适用于大滞后对象的控制器 ,两者结合 ,提出了一种简单、实用、鲁棒性强的大滞后对象控制的新方法

A Control Scheme for Plant With Large Delay Using Identification-Free Algorithm for Adaptive Control

In industry production, because of the pure delay, the stability of control system becomes lower, and the control quality will be worse, this is a great unfavorable factor to control system. For plant with large delay, Smith predictive control is an important method, but it is very sensitive to error of model (include estimation error of time delay), thus is not suitable for the system with varying time delay. Therefore the Smith predictive control is difficult to use in industry control widely, but it is still accepted by control field. Identification-free algorithm for adaptive control is an adaptive control algorithm which does not need to identify model parameter, and is put forward by Marsik and Strejc. This algorithm is simple with strong robustness. It only needs to inspect practical output and expectation output in order to form an adaptive control system with fine dynamic performance. In this paper, a new method was present that refers to the idea of identification-free adaptive control and strong approaching ability of neural network. The method first approximates Smith predictive model by using neural network, and then used the identification-free algorithm for adaptive control to design a controller which is suitable for plant with large delay. This method is a simple, applicable, robust strong new method, and can be applied to plant with large delay.