Intelligent Data Prediction Model of Catalyst and Temperature Combination Strategy Based on Stochastic Forest Model

The technology of ethanol conversion to olefins, alkanes and aromatics is a strategic choice for China’s petrochemical industry and has great development potential. Firstly, the data prediction model was established with the parameters and temperature in the catalyst combination as the input and the C4 olefin yield as the output. Therefore, we use the experimental data to train the three models of random forest, BP neural network and gradient lifting regression (GBR), and compare the predicted data with the experimental real data. Through the error analysis of RMSE, MAE and R-squared indexes, it is found that the predicted value obtained by the random forest model is closest to the real value. Then, in order to obtain the specific parameter value when the C4 olefin yield is the maximum, we use the particle swarm optimization (PSO) to build the model, and take the relationship between output and input in the random forest model as the fitness function to obtain the optimal composition and temperature of the catalyst without temperature limit and when the temperature is lower than 350 °C.