Research On The Soft Sensing Of Aviation Kerosene Parameters Based On KPCA And Mixed Kernel LS-SVM

In petrochemical enterprises, under the condition of the existing detection techniques and technological conditions, dry point and flash point as quality indicators of aviation kerosene are difficult to be measured online. How to realize online real-time measurement of the aviation kerosene quality parameters is the hot and difficult issue in modern process control field.In this paper, according to off-line measurement of aviation kerosene quality index has long time lag problem, a petrochemical company rectification device is as the research object. Based on the theory of crude oil distillation, various factors which affect dry point and flash point are analyzed. As a fist step 23 inputs have been chosen by the method of Mechanism Analysis. The process variables read from the DCS system of this enterprise, the laboratory values of dry point and flash point, which constitutes the required data as modeling samples. The collected sample data needed for modeling would be true signal of tested samples, while the actual signal can potentially corrupted by various types of noises. S-G filter is used to reduce the noise. Savitzky-Golay(S-G) is one of the filters which can smoothen out the signal without much destroying its original properties. Some of the secondary variables may be interrelated, KPCA was applied to choose the nonlinear principal component of the model in input data space, the effective information was extracted to eliminate redundancy variables. Finally, a methodology based on genetic algorithm (GA) to optimize the parameters and mixed kernel least square support vector machine regression (MKLS-SVM) for online quality prediction in atmospheric distillation column is presented. Using the industrial process data as the sample, we develop the multiple input multiple output soft sensor for the dry point and flash point of aviation kerosene. The simulation results show that the KPCA-MK-LS-SVM regression has better abilities of approach and generalizing for practical problems with small sample, nonlinear and high dimension. It can meet the quality requirement of the aviation kerosene in factory.