The Terminal Temperature Prediction Model Of VD Based On Fixed-Size Support Vector Machine

VD is an important equipment in steel external refining, especially in reducing hydrogen, nitrogen, oxygen and nonmetal impurities. Terminal temperature of VD is a key parameter relate to the quality of steel. Because of the vaccum characteristic of VD smelt process, it is hard for us to get actual time parameters of steel temperature. So it is uneasy to give an accuracy prediction of terminal temperature of steel. Usually, for most VD smelt process, people get VD terminal temperature according to experience from usual daily work, which leads to the loop hole of low prediction accuracy, further more influence the control of smelt process. Based on reasons above, this article carries out the research in VD terminal temperature prediction.Fixed-size support vector machine is a data learning method, which is fast in training, high accuracy in prediction and good at generalization. Based on analyzing support vector machine and least square support vector machine, this article illustrates the principle of fixed-size support vector machine precisely. Using Nystrom method makes it possible which in solving optimal problem in primal space. Apply quadratic Renyi entropy to select sub sample which delegate whole train sample, which shows sparsity.In order to establish a high accuracy predicting model of VD steel’s terminal temperature, it is necessary to have a deep study on VD workcraft and smelt process. Through research and analysis of VD craftwork and smelt process and based on the law of conservation of energy, article estabalishes mechanism model of VD terminal temperature model. In this article, I analyse different factors that influence steel temperature of steel and get math expression of these factors. At the sametime, through qualitative analysis and quantitive calculation, we get some factors that can be ignored. Finally I interface the math model of VD terminal temperature.Due to the lack of the data of part of parameter in mechanism model, we could on calculate the terminal temperature from mechanism model. To fix this problem, this article apply data model in predicting VD terminal temperature. Due to the complexity of VD temperature model, small data is hard to show all characters and diversification of VD terminal temperature prediction model, so it is hard for small data to get good predicion accuracy; at the sametime, big data learning theory existing long time training problem. In order to solve this controdictroy, this arcticle applys fixed-size support vector machine in VD terminal temperature prediction. Getting data model’s input parameter from mechanism model introduced in third chaptor. This article applies CSA algorithm to select parameter of kernel function and punishment parameter. From the result of simulation, compared with result of other data model, CSA-FSSVM model has shorter training time, and good prediction accuracy.Considered shortage of pure machenism model, which the model is often too complex and parameters of model may not able to be got, and shortage of pure data model, which is lack of instruction from craftwork, rely on data too much and model is hard to be explained. To solve this controdictory, this article tries to establish a parralle hybrid model of VD terminal temperature which combines machenism model and CSA-FSSVM data model. The simulation result shows that hybrid model has better predition accuracy in prediction than pure CSA-FSSVM model which is introduced in chaptor 4.