Simulation Test Of Interval Prediction Method Based On Fuzzy Information Granularity And Improved Recurrent Extreme Learning Machine

At present,with the increasing complexity of large-scale industrial production processes,the number of control points is increasing,and the prediction of key variables is increasingly demanding.Generally,the prediction of key variables is mostly based on point prediction methods.However,for the more complex process and uncertain monitoring variables,it is no longer sufficient to rely on the point prediction method to establish the system model.Meanwhile,in the actual production process,the prediction of data range for a certain period of time in the future is more practical than the actual data prediction.Therefore,this paper focuses on the research and analysis of the uncertainty of key variables in industrial processes.The contents are as follows:First,the industrial production process is mostly a dynamic sequential time series process,which the data is massive and the variables are interdependent showing high nonlinearity and uncertainty.Fuzzy information granulation method is introduced to preprocess the data that can effectively reduce the amount of calculation,avoid excessive learning,long time and other issues,so as to achieve the extraction of data features.In the process of fuzzy granulation,it is noted that the feature extraction effect is restricted by the membership function of fuzzy information granulation.Therefore,the membership function is selected to extract the features of time series data better.In order to verify the validity of its feature extraction,the data of the Shanghai Composite Index are used for comparison experiments and the membership function is selected.Second,as the hidden layer nodes are set randomly in traditional ELM,the model structure is often not stable and the model does not have dynamic characteristics,which cannot describe the industrial production process in a sequential manner.Therefore,this paper proposes an improved recurrent ELM.First,a feedback layer is added between the output layer and the hidden layer of ELM to memorize the output data and rate of change of the hidden layer,thereby dynamically updating the output of the feedback layer.Second,the model introduces regularization penalty to reduce the model structure error.Thus,the proposed model can reflect the dynamic characteristics of the system,which can adapt to the time series data of industrial production processes well.In order to verify the prediction accuracy of the model,the comparison experiments are carried out using the data of the Shanghai Composite Index.The experiments verify that the proposed IRELM model is superior comparing with other models.Third,combined with the above methods,this paper proposes an interval prediction model based on FIG and IRELM,and the interval evaluation indicators are applied to evaluate the effect of the proposed interval prediction model.At the same time,it is applied to pure terephthalic acid(PTA)solvent system for the interval prediction of key process variables.The experimental results show that the predicted interval has higher interval coverage and narrower interval width.In summary,the proposed interval prediction model based on FIG and IRELM in this paper can solve the problem of system uncertainty very well.