基于离散过程神经元的乙烯生产装置软测量

针对传统M—P神经网络模型的时间依赖性问题，提出将离散过程神经元应用到乙烯裂解炉软测量中，并将Fletcher—Reeves修正的改进变梯度学习算法应用到离散过程神经元网络，达到提高过程神经元网络的训练速度的目的。最后用乙烯装置的生产数据进行仿真研究，仿真结果表明该改进算法具有明显的快速收敛性，实现了乙烯产率的预测。

Soft Sensing of Ethylene Plant Based on the Discrete Process Neural

For the time-dependent problem of M-P neural network＇ model, applied the discrete process neural to the ethylene plant soft sensing, and the application of the Alternating Gradient training algorithm amended by Fletcher-Reeves algorithm on the discrete process of neural network. Finally, use the ethylene production plant production data to emulate. Simulation results show that the improved algorithm has obvious rapid convergence. Achieve a prediction of ethylene production.