文献类型：期刊文献

英文题名：GRU-based model-free adaptive control for industrial processes

作者：Sun, Jinggao[1];Wei, Ziqing[1];Liu, Xing[1]

机构：[1]East China Univ Sci & Technol, Key Lab Smart Mfg Energy Chem Proc, Minist Educ, Shanghai 200237, Peoples R China

年份：2023

卷号：35

期号：24

起止页码：17701

外文期刊名：NEURAL COMPUTING & APPLICATIONS

收录：;EI(收录号：20232114125451);WOS:【SCI-EXPANDED(收录号:WOS:000990949800001)】；

语种：英文

外文关键词：Model-free adaptive control; Recurrent neural network; Gated recurrent unit; Attention mechanism

摘要：Many industrial processes have the characteristics such as large time delay, strong coupling or nonlinearity, which make them difficult to control. When exact process models are available, model-based control methods can achieve good performance. However, it is difficult to satisfy in many cases. This paper proposes a novel Model-free adaptive control scheme with an encoder-decoder structure based on gated recurrent unit (GRU) network and attention mechanism. The control objective is to make the process output track a known reference input. The controller does not need to know an accurate process model and can adjust the weight of each neuron of the neural network according to the error signal to achieve process control. The gating mechanism of GRU neural network enables the controller to take full advantage of the system's history information. Lyapunov-based stability analysis is provided to guarantee the stability of the whole control system. Some process simulations and a Wood/Berry distillation column example show that only by adjusting a few parameters, the proposed controller can control a multivariable process with coupling or large time delay well.