Performance prediction of mold flux based on PSO-BP neural network
Mold flux is an important functional material in continuous casting.In order to accurately,quickly,and low-costly obtain the physical and chemical properties of mold flux,a model was established for predicting the physi-cal and chemical properties of the mold flux(composition,melting point,melting rate,and viscosity data)using BP neural network combined with particle swarm optimization(PSO)algorithm based on the testing data from laborato-ry.Thirteen untrained test samples were selected to test the prediction accuracy of the PSO-BP model.The results showed that compared with the BP neural network prediction model,the average absolute errors of melting point,melting rate,and viscosity were reduced from 8.9 ℃,4.7 s,and 0.012 Pa·s to 8.1 ℃,2.8 s,and 0.010 Pa·s,respectively.Moreover,the error fluctuations of individual samples were reduced,and the overall prediction accura-cy was improved.Based on this model,the influence of single or multiple changes in the composition of mold flux on the physical and chemical properties was studied.By controlling other components to remain unchanged,when the basicity increased from 0.8 to 1.2,the viscosity value decreased from 0.23 Pa·s to 0.18 Pa·s.In addition,the effects of single variable adjustment and simultaneous variation of Al2O3 and MgO on the viscosity performance of mold flux were demonstrated.The model calculation results were consistent with actual theoretical laws,indicating that the predictive model of mold flux based on PSO-BP neural network can be applied to the development and re-search of mold flux,shorten the research cycle,and reduce costs.
mold fluxphysical and chemical propertiescompositionPSO-BP neural networkprediction
NETL
NSTL
万方数据
