一种自适应CMAC 软测量与控制模型.docVIP

一种自适应CMAC软测量与控制模型王华秋 (计算机学院软件工程系重庆400050) 摘　要：通过研究了小脑模型神经网络(CMAC)存在的不足，设计了基于自适应CMAC算法，对CMAC的重要组成部分概念映射、学习率和惯性系数进行了自适应设计，提高CMAC的计算速度和精度以满足复杂动态环境下的非线性实时控制的需要。结合沉降分离控制系统工艺优化的需求，提出了基于自适应CMAC的沉降系统软测量和控制模型，用于准确实时地预测稀释矿浆固含，在此基础上进行絮凝剂投放措施优化。工业试验说明了该模型在对化学量软测量的预测精度和快速性上具有明显的优越性，本模型已应用于某氧化铝厂工艺优化系统中动态调节絮凝剂投放量，节省了生产成本，取得了明显的经济效益。自适应小脑模型神经网络沉降分离软测量絮凝剂稀释固含TP273 TF821 TF355　　文献标识码：A　　国家标准学科分类代码：520.2060 Soft sensor and control model with self-adaptive CMAC Wang Huaqiu (Software Engineering Department of Computer College, Chongqing University of Technology, Chongqing 400050, China) Abstract：The paper studies the existing deficiency of the cerebella model articulation controller (CMAC) and designs an adaptive algorithm to self-adjust the important components of CMAC，such as the concept mapping， learning and inertia factor. The self-adjustive CMAC improves the speed and accuracy of calculation to meet the complex and dynamic demand in non-linear environment for real-time control. Based on the process optimization demand of the capacity thickener separation control system，the paper presents a solid-liquid content soft sensor and flocculent control model self-adaptive CMAC to forecast the solid-liquid content of diluted pulp accurately and quickly. On this basis the flocculant flow quantity is optimized. Industry test shows that in terms of the accuracy and rapidity of the chemical soft sensor，the proposed model has obvious advantages. And the model has been applied to certain alumina plant to optimize dynamically the flocculant flow quantity to save the production cost and achieve obvious economic benefits. Key words：self-adaptive CMAC; capacity thickener and separation; soft sensor，flocculantsolid-liquid content of diluted pulp 1　引　　言 本文研究了小脑模型神经网络(CMAC)， 这是一种表达复杂非线性函数的表格查询型自适应神经网络该网络可以通过学习算法改变表格的内容，具有信息分类存储的能力[1]。由于CMAC的非线性逼近能力较好被广泛应用于故障诊断[2]、传感器测量[3]、冶金过程控制[4]、化工过程控制[5]等领域。为了提高CMAC的计算速度和精度来满足复杂动态环境下的非线性实时控制的需要许多学者对CMAC进行了改进，文献[6]对CMAC的存储单元个数C进行了设置该算法对一个样本进行学习后可以对其相邻的样本产生一定的效应输入向