连续碳酸化分解过程分解率的预测模型  

Research on Resolution Ratio Prediction Model of Continuous Carbonation Decomposition

作　　者：袁湘环[1] 谢永芳[1] 

机构地区：[1]中南大学信息科学与工程学院,湖南长沙410083

出　　处：《计算机测量与控制》2009年第4期695-696,722,共3页Computer Measurement &Control

基　　金：国家自然科学基金重点项目(60634020);教育部新世纪人才计划资助项目(NCET-07-0867)

摘　　要：连续碳酸化分解过程(简称碳分)是烧结法生产氧化铝的重要环节之一,其末槽分解率直接影响着氧化铝的产量和质量。碳分工艺是一个大滞后、非线性的复杂工业过程,分解率很难在线检测。针对此问题,论文提出了一种基于粒子群优化(PSO)的神经网络预测模型,对末槽分解率进行预测,以便实现分解率在线控制。该预测模型采用BP神经网络来构建,利用粒子群算法对神经网络的参数进行优化;结合实际工艺,对所建预测模型进行仿真研究,实验结果表明,建立的预测模型具有较好的泛化能力和较高的自学习能力,并具有较好预测精度,能为分解率的在线及优化控制提供指导。The process of the continuous carbonation decomposition （PCCD） is the key procedure in the sintering alumina production. The alumina production quality and production are mainly affected by the last resolution ratio in the process of PCCD. Since PCCD is a nonlinear industrial process with long delay time, it is difficult to detect the last resolution ratio online. For this problem, based on Particle Swarm Optimization （PSO）, a back-propagation neural network （BP） prediction model was proposed to forecast the last resolution ratio. The main idea is that the BP neural network was used to construct the prediction model of the last resolution ratio, and PSO was employed to optimize parameters of BP neural network in the proposed model. The experiment was performed using practical data, and the results show that the model possesses better generalization capability, higher self-learning ability and preferable precision, which can provide guidance to the optimal Control of the resolution ratio.

关 键 词：连续碳酸化分解 预测模型 粒子群优化算法 BP神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]