Application Research Of Interval Prediction Based On Improved CEEMD And Mixed Kernel RVM

With the more complex of the industrial process,the diversification of the variables and the improvement of the production quality,the prediction and monitoring of the key variables in the process have a higher requirement.In the past,the accuracy of point prediction alone has not been able to meet the actual application requirements.As a probability prediction method,interval prediction can not only predict the accuracy,but also predict the trend,and provide reliability analysis for the uncertainty of the system.At the same time,with the characteristics of high noise,volatility,complexity and irregularity of industrial data,the prediction and monitoring of variables become more difficult.Therefore,the research on interval prediction method is carried out for industrial processes with high noise,volatility,complexity and irregularity.The specific work is as follows:(1)In this paper,an improved complementary integrated empirical mode decomposition de-noising method is proposed to solve the problem of high noise and non-stationary data.This method mainly integrated complementary empirical mode decomposition and sample entropy(CEEMD-SE)to decompose the data,and then analyzes the complexity of the components by the sample entropy value.The value reconstruction component obtains three parts:noise,periodicity and trend,excluding the noise components,so as to achieve the purpose of denoising.(2)A method is proposed to improve the accuracy of point prediction models based on the improved mixed kernel relevance vector machine(MRVM).MRVM mainly uses linear kernel function and the gauss kernel to synthetically measure the importance of the kernel function in the RVM structure.And the kernel density estimation method is used to optimize the kernel parameters.(3)Aiming at the trend prediction of key variables and the reliability evaluation of the system,this paper presents an interval prediction method integrating CEEMD,SE and MRVM.The simulation results are carried out through two actual industrial processes:HDPE and PTA solvent system.The simulation results show that the proposed method has a certain improvement in the precision of the point prediction;a better interval coverage rate and a narrow range of interval width are obtained in the interval prediction,and it still maintains a good prediction effect in the multi-step interval prediction.The trend of variables brings new ways to provide reliability analysis for system uncertainties.