Application And Research Of Particle Swarm-Support Vector Machine In Leakage Accidents Of Benzene Storage Tanks

Benzene is a common dangerous chemical in chemical production.Once a benzene storage tank leaks,if it does not take effective emergency treatment measures in time,it will easily lead to poisoning and even fire and explosion accidents.At the same time,due to the existence of a large number of dangerous substances in chemical production,it is prone to chain reaction,which seriously threatens people’s lives and property safety.Therefore,we need to pay attention to benzene storage tank leakage accidents.Predicting the risk of benzene storage tank leakage accidents before the accident and real-time prediction of the concentration of benzene storage tank leakage accidents are of great significance in reducing the risk of benzene storage tank leakage accidents and improving the emergency response capability of benzene storage tank leakage accidents.At present,the research on risk prediction of hazardous chemicals leakage mainly focus on the combination of qualitative and quantitative analysis of dangerous sources and the analysis of accident consequences,but it is difficult to predict the risk of leakage of benzene storage tanks before the accident.The research on the prediction of the leakage concentration of hazardous chemicals mainly includes theoretical calculation,experimental measurement and numerical simulation analysis,but there are shortcomings such as limited conditions and extremely time-consuming calculation,which are not applicable to the emergency scene of leakage accidents.Support vector machine model is one of the main machine learning algorithms,which can effectively deal with small sample and nonlinear problems,but the accuracy of its model will be affected by its internal parameters.Therefore,this paper proposes the use of global optimization algorithm particle swarm optimization(PSO)to optimize the support vector machine(SVM)model parameters,establish a particle swarm-support vector machine(PSO-SVM)model,and innovatively apply it to the risk prediction of benzene storage tank leakage accidents before accidents and the real-time prediction of benzene storage tank leakage concentration in accidents.The main work of this paper is as follows:First of all,we use PSO-SVM model to predict the risk of benzene storage tank leakage accident before accident.To verify the risk prediction performance of PSO-SVM model,We compared the mean square error,correlation coefficient and parameter optimization results of the predicted values and test sets of PSO-SVM,GA-SVM and GS-SVM,furthermore,the weight adjustment method in the PSO-SVM model and the influence of the population number on the risk prediction value are discussed.It is found that the PSO-SVM model has the smallest mean square error between the predicted risk value and the test set,the highest correlation coefficient and the best prediction effect.The PSO-SVM risk prediction model built by the linear weight reduction is the best and the best in the test set.The number of groups does not affect the prediction accuracy of the PSO-SVM model but affects the time consumption.Then,we use the PSO-SVM model to predict the concentration of benzene storage tank leakage accidents in real time.In the real-time prediction of concentration,we use ALOHA simulation to obtain sample data.In order to verify the concentration prediction performance of the PSO-SVM model,we compared the relative error,mean square error and correlation coefficient of the predicted and simulated values of the three models of PSO-SVM,GA-SVM and BP neural network.It is found that the correlation coefficient between the PSO-SVM model concentration prediction value and the simulation value is the highest,and the relative error and the mean square error are the smallest,indicating that the PSO-SVM concentration prediction model performs better.Then to further verify the correctness of the established PSO-SVM concentration prediction model,we separately predict the concentration distribution of the 12th and 13th leakage scenarios.By dividing the scope of the accident,it is found that the shape of the impact range of the accident is basically consistent with the shape of the ALOHA simulation,and the predicted impact range is basically similar to the simulated range.Finally,we apply the PSO-SVM model to the benzene storage tank leakage scenario of a coalification base,use FDS simulation to obtain the concentration sample data,establish a PSO-SVM based dangerous location concentration prediction model,and compare the prediction results with the FDS simulation values.The mean square error and correlation coefficient between the predicted concentration value and the simulated value are then further calculated to characterize the predicted result.It is found that the predicted concentration value of PSO-SVM is roughly the same as the overall trend of FDS simulation value,and the error is within the acceptable range.And there is a small mean square error and a high correlation coefficient between the predicted value and the simulated value,which indicates that the concentration prediction accuracy based on the PSO-SVM dangerous position is better and can be used in practical engineering applications.