Application Research Of Combination Model Of SVM And Neural Network In Prediction Of Thermal Efficiency Of Tubular Furnace

At present,energy resources are very scarce,and the problem of environmental pollution is becoming more and more serious.Enterprises urgently need to increase energy conservation and emission reduction efforts.As the largest energy-consuming device in petrochemical enterprises,the importance of tubular heating furnace to the oil production process is self-evident.At present,the research on energy efficiency optimization of tubular heating furnaces by enterprises mainly focuses on improving the structure of heating furnaces and related processes.However,for enterprises,the most concerned issue should be the energy-saving optimization of heating furnace equipment that has been put into use.Enterprises are required to improve the thermal efficiency of the heating furnace.Most companies still use traditional methods to measure thermal efficiency,which has a certain hysteresis,making it difficult to truly improve thermal efficiency.For this reason,this paper studies the thermal efficiency prediction of tubular heating furnaces,and designs a combined model for thermal efficiency prediction of tubular heating furnaces.Make the combined prediction accuracy higher than the prediction accuracy of each single model before the combination.The main work of this paper includes:Explain the fundamentals of support vector machines,neural networks,and combined forecasting,and select a single forecasting model to study.Mechanism analysis is carried out on the original data,and the feature selection and extraction of the data are carried out through correlation analysis,nuclear principal component analysis(KPCA)and visual analysis to determine the experimental data required for the final construction of the thermal efficiency prediction model.Then a single prediction model is studied,a thermal efficiency prediction model based on LSSVR is constructed,the regularization coefficient and kernel function parameters are optimized by the improved grid search method(Grid Search),and the prediction results show that the optimized model accuracy has been improved.Next,a thermal efficiency prediction model based on error back-propagation neural network is constructed,the network is optimized by the additional momentum(Momentum)stochastic gradient descent method,and an adaptive learning rate is introduced.The experimental results show that the prediction accuracy of the model is greatly improved compared with that before optimization.On this basis,the LSSVR prediction model and the BP neural network prediction model are combined,and a nonlinear combination prediction model based on neural network is established.Using the powerful nonlinear mapping ability of neural network to fully mine the information of each single prediction model,the mean square error of the combined prediction model is reduced by 67.7% and 63.3% respectively compared with the single prediction model.Therefore,the nonlinear combined model of LSSVR and BP neural network is used as the final model for the thermal efficiency prediction of the tubular heating furnace.