Study On Prediction Method Of Blast Furnace Lining Erosion Based On Neural Network

The corrosion degree of the blast furnace lining is an important factor affecting the production life of the blast furnace.Aiming at the shortcomings of traditional blast furnace erosion analysis methods of large errors and cumbersome calculations,this paper combines particle swarm algorithm and BP neural network to apply the prediction of blast furnace lining erosion appearance,and establishes the boundary line of lining erosion by using the nonlinear mapping ability of BP neural network The relationship with the temperature of the furnace wall temperature measurement point.First of all,this article is based on the two-dimensional heat transfer equation,using finite element analysis to solve the temperature distribution of the blast furnace lining.By constructing 34 groups of erosion boundaries reasonably,a two-dimensional geometric model of 34 groups of hearth sidewall cross-sections and hearth bottom longitudinal cross-sections was established,and the temperature field of the model was solved by the finite element analysis software COMSOL,and then the temperature value of the temperature measurement point was obtained.The coordinates of multiple erosion points on the 34 groups of erosion boundaries and the obtained corresponding temperature measurement points are used as the subsequent neural network samples.Then,based on the BP neural network theory,this paper designs a prediction model for the lining transverse and longitudinal cross-section erosion.30 sets of data are selected as training samples,and the remaining four sets of data are used as test samples,with temperature data and erosion point coordinates as input and output,respectively,to train and simulate the horizontal and vertical cross-section network models.Finally,this paper uses the particle swarm optimization optimized by time-varying inertia weight to improve the BP neural network model of the transverse and longitudinal cross-section erosion prediction.The same 30 sets of training samples and four sets of test samples are used to train and simulate the cross-section and longitudinal section erosion prediction network models respectively.Through comparative analysis,the maximum error of the prediction results of the BP neural network model for the four groups of samples for the horizontal and vertical cross-section erosion predictions are all between 5 and 10 cm,and the maximum error of the prediction results of the improved model for the four groups of samples are all less than 5cm.It is verified that the erosion boundary line predicted by the improved network model has a higher degree of coincidence with the actual erosion boundary line.