Research On Online Correction Method Of Continuous Casting Solidification Heat Transfer Model

In the process of continuous casting,the secondary cooling control affect the casting machine production and quality of slab directly and is the key of slab internal organization and internal defects.Therefore,the determination of surface temperature of slab in secondary cooling area,is very important to formulate reasonable cooling system and improve the slab casting machine's production and slab's quality.People usually use mechanism of solidification heat transfer model to predict the surface temperature of slab.However,the process of continuous casting is complicated,and working conditions often change,the long-term accuracy of model application not guaranteed.Therefore,the research of slab heat transfer model which has the function of self correction has practical application meaning.For this purpose,this paper studies a kind of online correction method of continuous casting solidification heat transfer model.Based on the real-time accurate prediction of slab surface temperature to realize the purpose of heat transfer model to implement the on-line calibration.The main research content is as follows:(1)Analyzed the production technology of continuous casting slab solidification and heat transfer process and characteristics,established the continuous casting solidification heat transfer model.(2)Taking a billet caster as the research object,aiming the problem of long-term accuracy of solidification heat transfer model,on the basis of the established mechanism model,Integrating the concept of soft measurement,this paper proposes a method of combining the auxiliary model and mechanism model,to correct the output of mechanism model.The auxiliary model based on the principle of BP neural network and support vector regression(SVR),it can predict the difference between correction point temperature and the slab surface temperature outside the secondary cooling zonethrough the measurement data at anytime.Then use the temperature of slab's surface add to forecast difference one by one to gain the temperature of each correction point's temperature.Implementation of slab surface temperature field of the accurate calculation.(3)For the auxiliary model based on BP neural network,combined with the study of specific issues,determine the BP neural network layers,each layer node number and related parameters Settings.Aiming at the existing defects of BP neural network,genetic algorithm combined with BP neural network,the method of using genetic algorithm to encode the network weight threshold,and the fitness function and the related parameters,to optimize the network weights and threshold value,improve the performance of the BP neural network auxiliary model.(4)For the auxiliary model based on SVR,combined with the specific issues,select the kernel function,use the method of cross validation to optimize the coefficient c and the width of kernel function.In view of the limitations of thecross-validation optimization,propose a method which combined the particle swarm algorithm and SVR.Using its global searching characteristic to search the best parameter c and of the model,to improve its performance.(5)Using Matlab to simulate the data,verify the feasibility of all solutions.The results show that the prediction accuracy of correction method based on BP neural network and SVR improves after optimization of genetic algorithm and particle swarm optimization.By contrast,the correction accuracy of method which is based on PSO-SVR is the highest.The MSE between correction value and actual value is lower than 4.Taking a group of field measurement data,it cost 0.285 second to complete the correction.Therefor this method can satisfy the requirement of continuous casting solidification heat transfer model of online correction.