Research On Modeling Of Hydrodynamic Parameters Of Trickle Bed Reactor

The trickle bed reactor is a gas-liquid-solid three-phase fixed-bed reactor with a simple structure.The scattered catalysts in the reactor bed form intricate flow channels,and the gas-liquid two-phase flow is affected by a variety of the impact of interaction.Liquid holdup and pressure drop are two important hydrodynamic parameters of trickle bed reactors,and are also important indicators to measure reactor performance.They play a key role in the design and optimization of the reactor,and are affected by the characteristics of the bed and the gas.The influence of parameters such as liquid flow rate and fluid properties.At present,the prediction methods of fluid dynamics parameters of trickle bed reactors include empirical models,phe-nomenological models,computational fluid dynamics models,and machine learning models.The first three methods are more traditional and most used,but the reactor has two-phase gas-liquid fluid flow.There are complicated rules and the influence mechanism of flow is not com-pletely clear,which leads to the problems of low accuracy and poor generalization of traditional forecasting models.The machine learning algorithm is a novel modeling method,which can complete the modeling of the reactor hydrodynamic parameters without fully clarifying the mechanism,and its modeling efficiency is higher than that of the traditional modeling method.Therefore,this manuscript uses the machine learning algorithm of Random Forest and Deep Belief Network to model the fluid force parameters of the trickle bed reactor.First,according to the prior knowledge of the trickle bed reactor,the experimental data published in the literature were integrated as the original data,and 80%of the original data was randomly divided into the training set,and the remaining 20%of the original data was used as the test set.Then the random forest method is used to evaluate the feature importance of the training set,and the feature combination with cumulative importance close to 95%is selected as the input of the prediction model.After standardized preprocessing,the data is input into the deep belief network model.The number of hidden layers and the number of hidden layer nodes of the DBN model were optimized with the mean square error as the performance index.The optimal combination of structural parameters of the DBN model was determined through com-parative analysis,and finally the reactor liquid holdup and pressure drop models were trained separately.The model test shows that the MSE of the RF-DBN prediction model(including the fluid holdup and pressure drop model)are 1.2e-4 and 1.23,MRE are 4.81%and 15.03%,and R~2 are 0.967 and 0.975,respectively.Compared with other models,the RF-DBN model has the highest model accuracy and the strongest generalization ability,and the model is sensitive to changes in input characteristics,which conforms to the gas-liquid two-phase flow mechanism.Finally,using the GUI function of MATLAB to design a trickle bed reactor hydrodynamic parameter prediction system based on the RF-DBN model,the system realizes the prediction function of the trickle bed reactor liquid holding capacity and pressure drop and custom training RF-DBN model functions,which provide convenience for the application of trickle bed reactor prediction models.