摘要
针对高维非线性系统,分析了基于多分辨分析的正交小波网络的建模能力,并用于循环流化床锅炉燃烧过程的动态建模.根据现场采集的实时数据进行网络训练和泛化实验。
Circulating Fluidized Bed Boiler (CFBB) has a splendid future among all kinds of coal-burning furnaces.At the same time,how to control the CFBB is regarded as a challenging problem because of its strong nonlinear?coupling multivariable,time delay and time-varying characters.Neural-network-based predictive control,which takes neural networks as predictive model,has strong robustness in nonlinear MIMO process control and is recommended to be adopted in the CFBB control;The wavelet network,a type of feed-forward basis function network,is chosen to build the nonlinear predictive model due to its fast convergence,small size and especially the linear relationship between its node outputs and weight coefficients which could be expediently adjusted online and this is very important in the CFBB control.In this paper,the orthogonal wavelet network is proposed in the dynamic modeling of the combustion process of CFBB, not only for it can effectively avoid the problem of 'curse of dimensionality',but for it has more significant identifying accuracy and smaller network size when adopted in the low-frequency chemical processes.The industrial data collected from two kinds of industrial processes are used as trained samples and predicted samples.Both theory analysis and application results show that the learning accuracy and the generalization capability of this wavelet network are satisfying.
出处
《厦门大学学报(自然科学版)》
CAS
CSCD
北大核心
2005年第4期525-529,共5页
Journal of Xiamen University:Natural Science
基金
厦门市科技计划项目(3502Z20021090)