Study Of Least Squares Support Vector Machine Prediction Model Based On Temperature Of Molten Iron In The Blast Furnace

Blast furnace temperature prediction process is a not touch, can’t see through the nonlinear system, Process parameters are uncertain, nonlinear, high dimension features. The temperature prediction model is based on the silicon content of molten iron as the object of study,Although there is a positive correlation between silicon content and temperature, but also thepresence of nonlinear, So how to describe the change of temperature with the silicon contentof molten iron is the lack of scientific, Therefore, based on the physical temperature of furacetemperature is the temperature of molten iron as the research object, and to establish the forcasting model. Support vector machine is a kind of algorithm recently developed, which cansolve the classification problem but also can solve the regression problem, is considered tobe the replacement algorithm and neural network algorithm. At the same time can well dealwith small samples, nonlinear, high dimension, fast convergence speed, good generalizationability, So the author adopts the prediction model of the temperature of molten iron to buildsupport vector machine technology. The main work of this paper has:1. The blast furnace process data in the collection, part of the data has missing values,noise, outliers, different classes and lag issues such as order parameter, If you do not doprocessing will directly influence the performance of the model, in order to improve themodel performance, this paper will take corresponding measures for treatment of theseproblems before the modeling.2. Based on the principal component analysis (PCA) and support vector machine (SVM)method of metal temperature time series model is established, using the principalcomponents analysis method to reduce the dimensionality of the processed data, time seriesdata principal component, as input variables of the model, and then using support vectormachine algorithm to establish the temperature of molten iron prediction model, parametersof particle swarm optimization algorithm to optimize the model and uses the linear decreasing inertia weight, finally gets the prediction results, and the optimized support vectormachine model of time series model, RBF neural network time series model to docomparison of particle swarm, the forecast precision of least squares support vector machinemodel of time series is higher, shorter computing time.3. Put forward a kind of clustering and least square support vector machine (LS-SVM)based on the method of establishing the temperature of molten iron multiple regressionmodel, this method can solve the prediction accuracy of single model of low, poorgeneralization ability problem, firstly, subtractive clustering is used to determine the numberof clustering the samples to be classified, the use of K-means for classification of trainingsamples, prediction model then using least squares support vector machine algorithmestablishes different categories, at the same time using particle swarm algorithm to optimizethe parameters of the model, and the test sample into a different model to obtain the forecastresult, the final weighted and get the final prediction results, in contrast to the least squaressupport vector time series model and the model of support vector machine is established, theleast squares support vector machine model of the hit rate is higher than other models.