改进人工神经网络法进行二(2,2,6,6-四甲基-4-哌啶基)马来酸酯合成反应工艺参数优化

针对正交设计的局限性，采用人工神经网络对正交实验数据建模。经过近千种不同因素不同水平之间的组合模拟，预测出以分子筛负载钛酸四异丙酯为催化剂，由马来酸二甲酯和2,2，6，6-四甲基-4-哌啶醇催化合成新型受阻胺光稳定剂的关键中间体二(2，2，6，6-四甲基-4-哌啶基)马来酸酯的最佳反应工艺参数，在最佳工艺条件下的神经网络模拟收率为96，8％，10次实验室小试和10L反应器中放大实验的平均收率为96．6％。

The Process Optimization for Synthesizing Bis(2,2,6,6-tetramethyl-4-piperidinyl) Maleate by Neural Network Technique

Bis(2,2,6,6-teramethyl-4-piperidinyl) maleate(BPM) is a key intermediate for preparing new types of hindered amine light stabilizers. The authors had developed new technology for synthesizing BPM from dimethyl maleate and 2,2,6,6-tetramethyl-4-piperidinol with zeolite supported tetraisopropyl titanate as catalyst. The optimal reaction conditions were determined through orthogonal design. Using the experimental data of the orthogonal design as training sample set, a new model was constructed by improved back-propagation artificial neural network(BP-ANN). When the BP-ANN model could approximate to the variational regularities of orthogonal design, widened the range of the most factors level, then using the 988 kinds combinations of all widened factors at different levels as simulation input sample, the optimal reaction technological conditions for synthesizing BPM can be obtained by trained BP-ANN model. Under this condition, 5 times laboratory-scale test and 5 times pilot-plant test in 10L reactors were performed, the average yield reach 96.6%. The experimental results are in good agreement with the predicted yield which is 96.8%.