乙炔羰基化合成丙烯酸甲酯催化性能的人工神经网络模拟

GARBF Artificial Neural Network Simulation of the Catalytic Performance of the Synthesis of Methyl Acrylate via Acetylene Carbonylation

通过神经网络技术可找出催化工艺与催化性能之间的关联性,从而对催化性能进行预测,达到提高研究效率的目的。本文针对训练样本中奇异样本对神经网络模型预测能力和泛化能力的影响,将遗传算法思想引入神经网络,构建神经网络模型动态训练集,建立了遗传算法-神经网络模型(GARBF);利用GARBF模型对乙炔羰基化合成丙烯酸甲酯催化性能进行预测模拟。结果表明:与RBF相比,GARBF的预测精度明显提高,对于六组测试集,平均相对误差从2.94%降低到1.18%,体现了更强的泛化能力。

Through the neural network technology, the correlation between catalytic process and catalytic performance can be found,and the catalytic performance can be forecasted, so as to improve the research efficiency. In view of the influence of the abnormal training samples on the prediction ability and the genealization capacity of the neural network model. This paper intro-duces genetic algorithm（GA）into neural network to construct neural network dynamic training set, and to establish GARBF neural network model, which is used in the prediction simulation of the catalytic performance of the synthesis of methyl acrylate via acetylene carbonylation. Compared with RBF,the prediction ability of GARBF improved obviously. The average relative er-ror of the six groups of test sets has been reduced from 2. 94% to 1.18% ,demonstrating stronger generalization capacity.