一种基于ELMAN神经网络的纸张横幅定量在线解耦方法

An Online Decoupling Method for Cross-directional Basis Weight of Paper Machine Based on ELMAN Neural Network

中文关键词:  纸张横幅定量  非方高维矩阵  强耦合  ELMAN神经网络  在线解耦

Key Words:cross-directional basis weight of paper machine  non-square high-dimensional matrix  strong coupling  ELMAN neural network  online decoupling

中文摘要:

      针对纸张横幅定量控制所存在的非方、高维强耦合特性，本研究提出了一种基于ELMAN神经网络的多变量解耦策略。具体思路如下：首先引入前置变换因子矩阵对系统进行降维化方，然后针对化方后的系统，将ELMAN神经网络和前馈解耦相结合设计解耦结构，并在此基础上引入参考模型，根据参考模型输出与系统输出之间的差值在线调整网络参数，对系统耦合进行实时补偿以实现系统的在线解耦。通过上述工作将非方高维控制问题转化为单回路群控制问题，大大降低了多输入多输出高维系统控制难度。仿真实验及现场实际应用证明了本研究所提方案的可行性和有效性，定量控制精度提升了42%，并将横幅定量偏差减小到1.88%。

Abstract:

      Aiming at the non-square and high-dimensional strong coupling characteristics of Cross-directional Basis Weight of Paper Machine， the paper proposes a feed-forward compensation online decoupling strategy based on ELMAN neural network. The idea is as follows： first， the system is reduced by introducing a matrix of prior transformation factors， and then the decoupling structure is designed by combining ELMAN neural network and feed-forward decoupling for the decoupled system. On the basis of that， a reference model is introduced and the network parameters are adjusted online according to the differences between the output of the reference model and the system， and the system coupling is compensated in real time to achieve online decoupling of the system. This transforms the non-square high-dimensional control problem into a single-loop group control problem， which greatly reduces the control difficulty of multi-input， multi-output， and high-dimensional systems. Simulation experiments and practical applications in the field have demonstrated the feasibility and effectiveness of the solution proposed in this paper， with a 42% improvement in quantitative control accuracy and a reduction in the error of Cross-directional Basis Weight of Paper Machine to 1.88%.