关键词: 浆态床渣油加氢, 关联模型, BP神经网络, 卷积神经网络, 遗传算法, 预测精度, 产物分布

Abstract: Pearson correlation coefficient method was used to select the variables in the residue slurry bed hydrogenation test data sample, 17 intput variables with low correlation were obtained by optimization, and a BP neural network model with the structure of 17-14-7 was established to simulate the distribution of hydrogenation products (cracking gas, gasoline, diesel, vacuum gasoline, residue, insoluble matter) and hydrogen consumption. Aiming at the problem of data imbalance and small amount of data in the original data, data expansion was relational model. The results showed that the improved BP neural network model had a structure of 17-15-7 after data expansion, and its prediction accuracy was greatly improved, with the maximum relative error reduced from 66% to 6.57%. In order to further improve the prediction accuracy and calculation efficiency of the model, the convolutional neural network model was introduced, and the maximum relative error of the model was reduced to 5.38%. Combining genetic algorithm with neural network model to optimize the model, the accuracy of model prediction of reaction product distribution could be further improved, and the maximum relative error of Convolutional neural network optimized by genetic algorithm was less than 2%.

Key words: slurry residue hydrocracking, association model, BP neural network, convolutional neural network, genetic algorithm, prediction accuracy, product distribution