基于粒子群优化算法改进的XGBoost模型制备C4烯烃工艺条件优化

C4烯烃是生产清洁友好燃料等化工产品的重要原料，提升C4烯烃收率，增大C4烯烃产量是生产过程重要的目标之一。针对乙醇偶合制备C4烯烃这一化学反应样本数量多，特征数量少等特点，本文提出一种组合模型：粒子群算法改进的XGBoost模型。首先将XGBoost模型与数据的拟合效果作为粒子群算法的目标函数，通过粒子迭代确定XGBoost模型的最优超参数; 然后通过对变量设置一定的步长构造仿真数据。最后，将优化后的XGBoost模型与仿真数据进行拟合，拟合优度由76%提升至93%。根据预测结果确定了C4烯烃的最大收率和最佳反应条件，得到C4烯烃收率的最大值为43.52%。实验结果表明，改进后的XGBoost模型在误差和精度方面都优于原始模型。

Optimization of processing conditions for C4 olefin production based on particle swarm improved XGBoost model

C4 olefin is an important raw material for the production of clean and friendly fuels and other chemical products. Increasing the yield of C4 olefin is regarded as one of the important objectives in production. For the chemical reaction of C4 olefin produced by ethanol coupling, which has a large number of samples and a small number of features, a combination model: particle swarm optimization (PSO) improved XGBoost model is proposed in this paper. Firstly, the fitting effect of XGBoost model and data is taken as the objective function of PSO, and the optimal hyperparameters of XGBoost model are determined by particle iteration. Then the simulation data is constructed by setting a certain step size for the variables. Finally, the optimized XGBoost model is fitted to the simulation data, and the goodness of fit is improved from 76% to 93%. According to the prediction results, the maximum yield and optimal reaction conditions of C4 olefin were determined, and the maximum yield of C4 olefin was 43.52%. The experimental results show that the improved XGBoost model is superior to the original model in terms of error and accuracy.