Researches Of Interval Prediction Model Method Based On Double Reservoir Echo State Network And Bootstrap

Nowadays,the industrial production tends to be more complex,the control nodes are increasing,so that the prediction for the key nodes need higher demand.Point prediction is always used for estimation.However,with more complex industrial process,the point prediction method cannot make uncertainty analyze for the variables by point value alone,so the point prediction method cannot meet the demand of monitoring the key variables to a certain extent.As a trend analysis method,interval prediction has a high accuracy in predicting key variables in complex industrial processes,and it can visually observe the trend of key variables,so it has become an important tool for industrial process monitoring.Therefore,this topic focuses on complex industrial processes and carries out research on estimation methods for key variable intervals,with the following three main tasks:First,an improved double-reservoir echo state network prediction model is proposed for timing data in complex industrial processes.At first,an improved principal component analysis(IPCA)is proposed for data dimensionality reduction.Secondly,an improved Immune Genetic Algorithm(IIGA)algorithm is proposed to optimize the reserve pool parameters in order to expand the parameter search range,and then solve the problem of parameter randomness and improve the model generalization accuracy.Second,based on the above point prediction model,a prediction interval model based on Bootstrap residual resampling is proposed.The residuals are extracted from the point prediction results by Bootstrap residual resampling,and then the prediction intervals are constructed by combining the model variance and noise variance.In addition,four indicators such as prediction interval coverage(PICP)and normalized mean prediction interval width(NMPIW)are introduced to judge the quality of prediction intervals.Third,simulation experiments using standard function dataset and Pure Terephthalic Acid(PTA)solvent system dataset are conducted,and the comparative analysis shows that the proposed method illustrates better on prediction accuracy and interval quality,which provides support for uncertainty analysis of key variables.