Research On SVM Based Temperature Model And Its Parameters’Optimization Of Reheating Furnace

Steel rolling heating-furnace, as of steel billet heating equipment, is an important part in the metallurgical industry, its main function is heat the steel billet in production line, in accordance with the requirements of the production process. Because heating-furnace control system possesses a multivariable and nonlinear system with large net lag and a high degree of coupling and other characteristics, it is difficult to establish a more accurate and practical model, and it is unlikely to control the temperature precisely, thus the research on steel temperature model for heating furnace produce is of great significance.The regenerative furnace is used as the background for the thesis, steel temperature modeling is established by the method based on multivariate statistical technique.Three steel temperature model have been got respectively:the model regressed by the method Least Squares Support Vector Machine (LSSVM); the model regressed by LSSVM after Principal Component Analysis (PCA) method for data processing; steel temperature model of reheating furnac regressed by LSSVM after Kernel Principal Component Analysis (KPCA) method deal with our large amount of data, in order to get high quality input data.Learning and generalization ability of LSSVM is determined almost by the ultra-parameter value, so it is necessary to optimize ultra-parameter of the slab temperature modeling. The thesis determines and optimizes the ultra-parameter used10fold cross-validation method, but the10fold cross-validation consums time especially, and the results are not optimal.The thesis uses the basic Particle Swarm Optimization (PSO) to get the global optimal ultra-parameters, and then establish a better model, by using the optimal parameters. Simulation results demonstrate the model with the ultra-parameter optimized by the PSO obtain better results.Parameter settings of basic PSO needs to repeatedly try, and PSO exist the earliness problem and the stability problem. To overcome these problems, an improved PSO algorithm is proposed, based on the standard PSO algorithm. It integrates the evolution theory and the multi-particle group thinking, which can avoid local optimization while maintaining good convergence speed.Finally, the summary and the prospect on this research are given.