IEEE Trans Neural Netw Learn Syst. 2013 May;24(5):837-43. doi: 10.1109/TNNLS.2013.2238638.
Considering the mismatching of model-process order, in this brief, a self-tuning proportional-integral-derivative (PID)-like controller is proposed by combining a pole assignment self-tuning PID controller with a filter and a neural compensator. To design the PID controller, a reduced order model is introduced, whose linear parameters are identified by a normalized projection algorithm with a deadzone. The higher order nonlinearity is estimated by a high order neural network. The gains of the PID controller are obtained by pole assignment, which together with other parameters are tuned on-line. The bounded-input bounded-output stability condition and convergence condition of the closed-loop system are presented. Simulations are conducted on the continuous stirred tank reactors system. The results show the effectiveness of the proposed method.
考虑到模型-过程顺序不匹配，在这篇简要介绍中，通过将极点配置自整定 PID 控制器与滤波器和神经网络补偿器相结合，提出了一种自整定比例-积分-微分（PID）类似的控制器。为了设计 PID 控制器，引入了一个降阶模型，其线性参数通过具有死区的归一化投影算法进行识别。高阶非线性通过高阶神经网络进行估计。PID 控制器的增益通过极点配置获得，与其他参数一起在线调整。给出了闭环系统的有界输入有界输出稳定性条件和收敛条件。在连续搅拌釜式反应器系统上进行了仿真。结果表明了所提出方法的有效性。
