Application Research Of Fuzzy Inference System Based On Granular Computing In Time Series Data Forecasting

Time series refers to the collection of data collected at different time points to reflect the changes of a certain phenomenon or thing over time,which generally appears in many fields such as social science,economy and finance.An important goal of time series analysis is to predict time series,that is,to find the internal evolution rules of time series from observed data by using statistical methods and techniques,and to build a model to estimate the variation trend of predicted variables.Time series forecasting is to predict the future trend through the observed data,so that the decision-maker has the ability to look far into the future and make extremely favorable decisions,so it is of vital significance.Based on the fuzzy neural network and fuzzy information granule,the single-step and long-term prediction of time series are studied in this paper.In the first set of experiments,a novel self-evolving interval long short term memory(LSTM)fuzzy neural network(eIT2FNN-LSTM)is proposed.In this model,the recurrent neural network with LSTM structure is introduced into the Type-2 fuzzy neural inference system to realize the single step prediction of time series,and the validity of the proposed model is verified on multi-class data sets.In the second group of experiments,a time series model combining fuzzy information granule and recurrent fuzzy neural network is proposed.The model first processes the original data set into a time series of fuzzy information granules,and then makes a granularity level prediction on eIT2FNN-LSTM to realize the long-term prediction(multi-step prediction)of time series.The contributions and innovations of this paper mainly include the following aspects:1)A novel recurrent fuzzy neural network with LSTM mechanism is proposed as the prediction model,that is,self-evolving interval type-2 LSTM fuzzy neural network(eIT2FNN-LSTM).It can not only realize single step prediction,but also can be combined with information granule to perform prediction at the level of granularity,so as to realize long-term forecasting.Compared with the existing work,the LSTM is applied to the neural fuzzy system for the first time,which can effectively solve the problem of long-term dependence.2)When eIT2FNN-LSTM is used for single-point prediction,dynamic density clustering algorithm is adopted to determine the fuzzy rules for the structural learning of the model.The main advantage of this method is that the existing rules can be adjusted continuously by changing the characteristics of the sequential data,that is,3)the clustering center.On the basis of the combination of eIT2FNN-LSTM and sliding mode control scheme,the synchronization of chaotic system with external disturbance noise is realized,which proves the system stability based on Lyapunov stability theory.4)Based on the variable-length segmentation method of stepwise linear partitioning(SLD),a generalized zonal time-varying fuzzy information granule(GZT-FIG)is proposed,which can contain the information of variable trend and fluctuation range.According to its characteristics,based on the Hausdorff distance theory,a new distance algorithm is proposed to measure and characterize the correlation between two GZT-FIGs and apply it to the structural learning of the model.Train and predict graphs with different trends in two inference systems respectively.By using this structure and dynamic clustering method,the proposed model can avoid prediction with wrong direction and improve long-term prediction accuracy.