Research On Prediction Model Of Dangerous Gas Diffusion Based On Neural Network

With the rapid development of China in the industrial field,incidents such as fires,explosions,and casualties caused by toxic gas released are also increasing.In gas diffusion simulation analysis,the requirements of computational fluid dynamics on computers limit its application.This study establishes surrogate models of computational fluid dynamics to predict the change trend of gas concentration with time and the spatial distribution characteristics of gas,the research mainly starts from the following three aspects.First of all,in order to use fewer sample points to achieve better spatial coverage,this paper uses HSS to design the value of the input variable.Based on the theory of fluid mechanics combined with COMSOL software,numerical simulation of the aerosol diffusion process in two-dimensional space is carried out,which provides reliable experimental data for the study.The grey correlation analysis reveals the influence of inlet air velocity,inlet aerosol concentration and aerosol particle density on gas concentration.Then this paper proposes an improved LSTM neural network concentration prediction model,which can use historical data to construct a corresponding trend model to predict the gas concentration value in the future.Setting the model parameters of the LSTM neural network to achieve the optimization effect and comparing it with several traditional time series prediction models,the LSTM neural network can obtain better performance when the data volume is not large.Then,a prediction model of improved LSTM network is proposed.The actual value is used as the input for the next step in each step of prediction during training,and the network parameters are revised to perform subsequent prediction.Simulation experiments show that the prediction accuracy of the improved model is improved by 59.2%,which verifies the effectiveness of the model.Finally,in order to further study the dynamic distribution of gas concentration with time and space in two-dimensional space,BP neural network optimized by genetic algorithm prediction models are proposed to study the gas diffusion process concentration.Considering the relevant influence parameters of time,space coordinates and concentration,four different models of input and output variables are constructed.The results show that the RMSE of the optimized model is reduced by 34.92%,40.65%,77.61%,and 81.06% respectively.The RMSE of Model 4 can reach 6.63e-04,which can show excellent performance for data prediction.The method proposed in this paper has been verified in the numerical simulation of aerosol diffusion,and the research has potential application value for industrial and environmental safety assessment.