一种多变量广义预测自校正控制器及其应用

A Multivariable Self-Tuning Generalized Predictive Controller and its Application

本文将Clarke等人提出的广义预测控制算法推广到P正则CARIMA模型描述的多变量系统,提出了一种新的多变量自校正控制器。这种多变量广义预测自校正控制器采用了显式算法,系统模型参数由带变遗忘因子的UD分解递推最小二乘法估计。该算法能够用于控制输入输出个数不同的系统,它具有鲁棒性好、需要较少的被控过程的先验知识、输出稳态无偏而且使用比较方便等优点。在座舱温湿度系统中实时控制的结果证明了该算法的有效性和实用性。

In attempting to overcome some of the problems associated with the application of current selftuning controllers in real processes, a novel multivariable self-tuning controller was presented in this paper. The proposed explicit multivariabie self-tuner, based upon the certainty equivalence principle, involved a parameter estimation algorithm and a control law calculation which uses the estimated process parameters. The parameters of each loop of the system under control were estimated using the UD factored least-squares method combined with variable forgetting factor technique,instead of the standard RLS,to avoid numerical instability and covariance windup of the estimator. The control law algorithm was derived from extending the generalized predictive controlor GPC of Clarke et al to admit a class of multivariable Systems described by the P-canonical CARIMA model. To evaluate the properties of this new multivariable self-tuning controller, several experimental applications were performed on the temperature and humidity control system of aground-based environmental cabin. The experimental results indicated that the proposed algorithm is applicable to the cases where the number of inputs is not necessarily equal to the number of outputs and where the plant structure is not necessarily the form of the assumed P-canonical CARIMA model. These experiments also showed that this algorithm has strong robustness properties, it seems to be insensitive to process deadtime, nonlinearities, dynamic variations, and even its own tuning parameters. So, it requires less prior knowledge regarding the system to be controlled, and it is easy to use. Besides, this controller ensures good rejection of load disturbances with offset—free regulation, it is simple enough to implement in a personal computer.