人工智能助力臭氧催化剂SrFexZr1-xO3的开发

将人工智能应用于催化臭氧氧化催化剂SrFe_xZr_1-xO₃的开发过程，采用共沉淀法制备了50种不同配方的催化剂，考察聚乙二醇（PEG）投加量、煅烧时间、老化时间、氨水投加量和铁掺杂量对SrFe_xZr_1-xO₃催化剂催化臭氧降解间甲酚反应活性的影响.同时，利用人工神经网络（ANN）和响应面（RSM）对催化剂合成条件与TOC去除率和间甲酚转化率的关系进行拟合，训练集中ANN的R²值分别为0.91和0.97，高于RSM的R²值0.35和0.41；在4组测试集上ANN的均方误差（MSE）分别为9.87和17.67，远小于RSM的23.89和28.87.结果表明，ANN模型对催化剂制备过程的复杂体系具有更好的拟合和泛化能力.在ANN训练好的模型中通过枚举法寻找最优合成条件为：PEG投加量为19.00%，煅烧时间为1.25 h，老化时间为26.50 h，氨水投加量为6.21 mL，铁掺杂量为3.37%，所得催化剂为SrFe_0.13Zr_0.87O₃-B.最佳反应条件下，间甲酚转化率和TOC去除率分别达到98.52%和17.21%，优于空白组的73.46%和1.86%.

Artificial intelligence assisted the development of ozonation catalyst SrFexZr1-xO3

Artificial Intelligence was first applied for catalyst development of SrFe_xZr_1-xO₃ in the catalytic ozonation process. 50 different catalysts were prepared by co-precipitation method, the effect of PEG additive amount, calcination time, aging time, NH₃·H₂O additive amount and iron doping amount on catalytic ozonation of m-cresol by SrFe_xZr_1-xFeO₃ were investigated. Artificial neural network (ANN) and response surface methodology (RSM) were proposed to fit the relationship between catalyst synthesis condition and TOC removal and m-cresol conversion. In the training set, R² of ANN was 0.91 and 0.97 which was bigger than 0.35 and 0.41 in RSM. Also, the mean square error (MSE) of ANN was 9.87 and 17.67 which was much less than 23.89 and 28.87 of RSM in 4 test data. This indicated that the ANN model had a better fitting and generalization ability in the complex system of catalyst preparation than RSM. In the model trained by ANN, the optimal synthesis condition was searched by enumeration and the best synthesis condition was that PEG additive amount was 19.00%, calcination time was 1.25 h, aging time was 26.5 h, NH₃·H₂O additive amount was 6.21 mL and iron doping amount was 3.37%, the resulting catalyst was SrZr_0.97Fe_0.03O₃-B. Under reaction conditions, m-cresol conversion and TOC removal reached 98.52% and 17.21% respectively, which were superior to blank group 73.46% and 1.86%.