| Information from data is used by the time series model to predict and analyze the state of the system,which has been widely applied in many fields such as industry,economy and medical treatment.With the increasing scale and complexity of modeling data,people hope that time series models can not only predict future information,but also provide the trend of modeling objects,and then obtain a certain semantic interpretation of the modeling results.Information granulation method is considered to discuss the granular representation of time series data,the modeling of interval time series and the evaluation of forecast results in this thesis.The main contents are as follows.Firstly,information granulation method is applied to convert time series data into granular data.A reconstructed interval-valued time series model whose data is measured by granularity criteria is established for different time scales.Then,an interval time series prediction method based on the compensation strategy is proposed.The function coefficient autoregressive model is used to predict the interval time series,and the genetic algorithm is adopted to identify its coefficients.Then the broad neural network is used to simulate the residual series.By constructing a compensation sequence,the prediction accuracy of the model for interval data is further improved.Besides,the combined forecasting strategy is proposed to model the interval time series.Fuzzy C-means clustering is adopted to determine the sub-regions and threshold parameters of the model,which can help to establish the segmented threshold autoregressive sub-model to realize the initial prediction of the time series.The least square method is applied to solve the parameters,and the broad learning system of incremental learning structure is used to dynamically compensate the error sequence.Then the variable weight segmentation compensation model is designed to correct the residual sequence.At the same time,the prediction interval is measured by the information granularity criterion,and the semantic interpretation is provided for the prediction result.Different types of data sets are numerically simulated,and the results show that the compensation forecasting method and the combined forecasting method designed in this paper can effectively fit the interval-valued time series.Finally,the proposed method is applied to predict the cardiovascular indexes of uremic patients during dialysis treatment,and the corresponding information management system is developed.The system has functions such as data management,preprocessing analysis,data prediction,result statistics and evaluation,which provides technical support for the management and quantitative analysis of diagnosis and treatment data for uremia patients. |
