Application Research Of Interval Prediction Method Based On Bootstrap And Improved Extreme Learning Machine

At present,with the increasing complexity and scale of industrial production process,there are more and more kinds and numbers of control variables.In the early traditional industry,point prediction is the main method to predict the key variables.However,due to the difference of the actual situation and the uncertainty of the relevant variables in the specific production process,the point prediction method alone cannot meet the actual industrial needs.At the same time,in the actual production process,compared with the data point prediction,it is more meaningful to predict the change range of data in a certain stage in the future for guiding practice.Therefore,this topic studies and analyzes the change trend of key variables in the industrial process,focusing on the interval prediction method and its application in the actual industrial process,as follows:Firstly,In order to predict the change trend of the related variables in the industrial process,this paper proposes a point prediction method based on MKELM-EF.The model mainly measures the importance of the kernel function in the prediction model according to the proportion of the two kernel functions.At the same time,the adaptive particle swarm optimization(APSO)algorithm is used to optimize the relevant kernel parameters in the model The prediction accuracy and convergence speed of the model are optimized.Secondly,aiming at the trend prediction of key variables,this paper proposes an interval prediction method based on bootstrap and improved limit learning machine.Among them,bootstrap is used to resample the original data and MKELM-EF is used as the regression algorithm.The combination of the two can construct an effective prediction interval.Singer function data is used to test the model to verify its effectiveness.Thirdly,select the key variables of PTA solvent system to verify the model.Including the performance verification of the point prediction model MKELM-EF and the interval prediction method based on bootstrap and MKELM-EF.In the point prediction experiment,the existing methods are selected for comparison experiment,and the results show that MKELM-EF model has some improvement in prediction performance.Meanwhile,several different interval prediction methods based on bootstrap and different prediction models are used for comparative experiments.The experimental results show that compared with other comparison methods,the method proposed in this paper has some improvement in the performance of prediction accuracy,and can take into account other indicators such as interval average width on the premise of meeting the interval coverage.