Research On Complex Industrial Processes Modeling Based On Kernel Method

Industrial processes are always with those characteristics: the high degree of structure complexity, strongly nonlinear behavior, time varying characteristics, close coupled and high degree of uncertainty, few useful information, etc. With the development of technology and productivity, the systems of the industrial process control become more complex due to the lack of precise, formal knowledge about system. Owing to the lack of precise mathematical model, it is difficult, even impossible, to control such complex systems. Since Prof. Vapnik proposed the Support Vector Machine (SVM) based on Statistical Learning Theory in 1995, kernel method based machine learning algorithm has been developed rapidly. It becomes one of the hot points in academic research now and has been widely used in pattern recognition, system identification successfully, etc. Therefore, based on the background of nonlinear and complex industrial processing, this paper applies the kernel method to the feather selection, modeling and prediction of the data of industrial process.The paper proposes a principal component fuzzy neural network model based on kernel method and greedy algorithm, which map the input variables into the high dimensional space in order to fully mine the hidden information. Through greedy algorithm to select the principal components, which are import into the fuzzy neural network model, the network reflects the hidden relationship between the data using rules, avoid the explosion of the rules and provide an efficient method to improve the prediction accuracy of the fuzzy neural network. Additionally, according to the statistics regression method, the paper constructs a Bayesians model based on the kernel method. Comparing with the traditional artificial experiences, the model gets rid of the influence of random factors, and improves the accuracy of dynamic process control. Using increment regression model and Bayesians regression model to build the model of the industrial process respectively, the paper compare the results of both the methods. And the Bayesians model based on the kernel method and automatically estimate the parameters, avoiding influence of manpower interference.In the end, take BOF steelmaking for example, the paper computes the second blowing oxygen volume using the result of the Bayesians model based on the kernel method, which combines the theory of three steps of decarburization to construct the exponential model, in order to direct the real time of the content of carbon and endpoint condition. And the paper modifies the integral model of carbon, combining the data detected by exhaust gas analysis, to compute the total amount of decarburization, so that a new approach to predict the carbon content of the endpoint is provided.