Research On End-point Prediction Model Of Electric Arc Furnace Based On Svm

In the steel-making process of electric arc furnace (EAF), the end-point parameters, i.e. temperature, carbon content and phosphor content of molten steel, are very important to the product quality. An exact prediction of end-point is propitious to improvement of the production efficiency. Because the end-point parameters of an electric arc furnace (EAF) are affected by both quantitative factors and non-quantitative factors, we combine GM (1, 1) model, Nonlinear Gray Bernoulli model, Gray-Markov model, Nonlinear Gray Bernoulli-Markov model, Markov model with Support Vector Machine (SVM) to produce the five combination prediction model for estimating the end-point parameter values of an EAF. The effects from the non-quantitative factors on the prediction values of end-point parameters are reflected by GM (1, 1) model, Nonlinear Gray Bernoulli model, Gray-Markov model, Nonlinear Gray Bernoulli-Markov model, Markov model; while the effects from the quantitative inputs are reflected by the SVM. The GM (1, 1) model, Nonlinear Gray Bernoulli model that reflect non-quantitative factors are established firstly, and then, their prediction value are revised by the Markov chain. Because the effect from the quantitative inputs can not be reflected by GM (1, 1)-Markov model and Nonlinear Gray Bernoulli-Markov model, the GM (1, 1)-Markov model and Nonlinear Gray Bernoulli-Markov model are certainly not free from prediction errors from the quantitative inputs. These prediction errors are compensated by the SVM model with parameters optimized by particle swarm optimization (PSO) algorithm. The final five prediction values of the end-point parameters in EAF are thus obtained, respectively, GM (1, 1)-SVM model, Nonlinear Gray Bernoulli-SVM model, Gray-Markov-SVM model, Nonlinear Gray Bernoulli-Markov-SVM model, Markov-SVM model. Meanwhile, a rolling forecasting is realized. Experiments shows that the five combination prediction model have the best prediction precision and hit probability, which are suitable for the end-point parameters prediction of EAF. Comparing to GM (1, 1)-SVM model and Nonlinear Gray Bernoulli-SVM model, the prediction precision of Gray-Markov-SVM model, Nonlinear Gray Bernoulli-Markov-SVM model and Markov-SVM model have been further improved.The innovation of the paper is as follows:(1)GM (1, 1) model, Nonlinear Gray Bernoulli model can be revised by Markov chain, two new models that Gray-Markov-SVM model and Nonlinear Gray Bernoulli-Markov-SVM model are proposed by author. Experiments shows that the prediction precision of Gray-Markov-SVM model and Nonlinear Gray Bernoulli-Markov-SVM model have been further improved, comparing to GM (1, 1)-SVM model and Nonlinear Gray Bernoulli-SVM model.(2)The popular existing LS-SVM toolbox at home and abroad is not used. According to the principle of LS-SVM, LS-SVM program based on particle swarm optimization algorithm is programmed. The prediction precision is further improved by the program.(3)The rolling forecasting is used and the real-time online prediction function is realized.