摘要
催化裂化装置是一个高度非线性、时变和长时延、强耦合、分布参数和不确定性的复杂系统。为解决催化裂化过程的优化控制问题 ,采用多层前馈神经网络进行辨识、建模 ,用周期图检验法对模型检验 ,用改进的 Frank- Wolfe算法进行稳态优化计算 ,并以大港炼油厂实际生产过程的稳态数据进行试验和验证 ,说明神经网络适合于解决非线性复杂生产过程的辨识。
A FCCU (Fluid Catalysis and Cracking Unit) is a highly non linear, transient system with long time delays, intensive coupling, and distributed parameters. The optimal control problem for the FCC process is solved using the MFNN (multilayer feedforward neural network) for system recognition and modeling, periodogram analysis for model testing, and the advanced Frank Wolfe algorithm for steady state optimization computation. Steady stable state data was used from the production process at Dagang Oil Refinery Works to train and test the neural network. The results prove that the neural network is effective for the system recognition, modeling and stable state optimal control of complex non linear production processes.
出处
《清华大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2000年第7期70-73,共4页
Journal of Tsinghua University(Science and Technology)
关键词
催化裂化
稳态优化控制
工业过程
炼油厂
DCS
fluid catalysis and cracking (FCC)
multilayer feedforward neural network (MFNN)
system recognition
modeling
stable state optimal control (SSOC)
