摘要
采用结合小波包分析的特征提取及神经网络的非线性映射等特性的小波神经网络系统,实现高炉铁水中Si含量的预报和控制。原始操作信息采用灰关联分析预选,网络结构设计采用剪除法确定隐含层节点,采取自适应和加动量项调整学习速率等措施。结果表明,系统具有更高的学习精度和更快的收敛速度,当允许误差为±0.02时,命中率达到87.5%,并且减少了系统参数特征量,优化了系统辨识和模型建立过程。
The prediction and control of the silicon content in molten iron from blast furnace is realized by use of a wavelet neural network with combination of the feature extracting of wavelet package analysis and nonlinear mapping of neural network. The production operating information is pre-selected by Grey Interconnect Degree Analysis. The automatic adopt the number of imply layers nodes, self-adaptation study and momentum items rate adjusting measures etc. are applied in network structural design. The results show that the system is accurate in learning and faster in convergence, the prediction accuracy is up to 87.5% while the average error is 0.02, and the number of systemic parameters is decreased and the processes of systemic identifying and modeling are optimized.
出处
《有色金属》
CSCD
北大核心
2005年第2期106-110,共5页
Nonferrous Metals
基金
国家杰出青年科学基金项目(60425310)
