摘要
研究了人工神经网络在化工过程测量数据校正中的应用,提出了新的样本构造方法和神经网络的在线训练策略。对某乙烯装置裂解气分离系统测量数据, 应用自行设计开发的改进算法的神经网络与数据校正系统集成运行, 结果表明基于神经网络的数据校正技术能对测量数据中所含的随机误差和过失误差进行同时校正,提高过程数据的精度和校正过程的稳定性,同时满足数据校正的实时性要求。
The application of artificial neural networks (ANN) in on-line data rectification of chemical process measurements was studied. A modified forward ANN using resilient back propagation algorithm was developed and integrated in data reconciliation system. With studying the separation process of ethylene cracking gases, automatic pattern acquisition, on-line training and on-line data rectification were realized. It is shown that precision of data obtained by simultaneous reconciliation and error detection is improved greatly. The proposed method utilizes abundant historical data to make up insufficiency of space redundancy. In addition, it avoids degradation of error detection power for small magnitude measurements due to their little contribution to constraint residue. ANN integrated operation strategy allows automatic and parallel on-line training according to changes of steady states. It overcomes operational limitation of ANN to some extent, thereby enhancing quality and stability of data rectification in practice. In contrast to traditional methods, this method is especially suitable for rigorous real-time application with less computation expenses.
出处
《高校化学工程学报》
EI
CAS
CSCD
北大核心
2003年第3期319-324,共6页
Journal of Chemical Engineering of Chinese Universities
基金
国家自然科学基金(29976015
20225620)
国家重大基础研究规划(G20000263)
教委博士点基金的资助。
