Research On The Application Of Metacognition Based Fuzzy Inference System In Time Series Prediction

Time series is an ordered set of data arranged in chronological order.The purpose of studying time series is to find the time series relation between data,and then to predict the future data changes reasonably according to the known historical data.In recent years,time series modeling and prediction based on machine learning has attracted the attention of researchers.For example,neural network,support vector machine and other learning models are widely used in the research of time series data.Based on the existing research,this paper uses metacognition and intuitionistic fuzzy neural network to discuss the problem of time series.Based on the characteristics of time series,it explores the structure of learning model and learning algorithm.The main work is as follows:(1)Data preprocessing is added to reduce the complexity of data and improve the efficiency of system operation.This paper studies the temporal relations contained in the original time series by using empirical mode decomposition method to map one-dimensional data to a multidimensional model,then it explores the changing trend of data expressed in different dimensions.By analyzing the trend of data in multi-dimensional,we can find the possible instability value in the original data and deal with it pertinently.By preprocessing the time series,we can get a smoother time series on the basis of preserving the characteristics of the data subject,which provides convenience for further research.(2)Based on fuzzy neural network,intuitionistic fuzzy sets are used to expand the fuzzy link.The original fuzzy neural network only pays attention to membership degree,but the real problems are often ambiguous.So it is impossible to use membership degree as a single value to summarize,which leads to the omission of important information in the model and the inaccurate prediction results.By adding the concept of hesitation degree,the fuzzy sets are extended to intuitionistic fuzzy sets.Membership degree and non-membership degree are used to express the relationship between samples and rules at the same time,which makes the prediction model more applicable in dealing with practical problems.(3)In the multi-layer fuzzy neural network model,through adding self-feedback link in the network,using the gated recurrent neural network is to memorize the characteristics expressed by the learned data samples,this paper discusses the long-range time-series dependence between the data before and after the sequential learning on the linkage effect of the prediction system itself.In the system construction,the vertical influence of time layer and the horizontal state of data flow between multi-layer networks are considered simultaneously,so as to find the proper connection and construction method for self-feedback link.(4)In order to construct the prediction model more efficiently,the model structure and parameter updating process are studied.Because the continuous variability of time series needs the online update of reasoning system,the efficiency of system construction becomes an important index to measure the system performance.In the construction process,metacognition is used to endow the system with the ability of autonomous learning.The calculation results and output results generated when the current sample passes through the system are used to establish a relationship with the model parameters,so as to judge the possible impact of the current sample on the model,to guide the prediction system to make a correct judgment and make the system construction intelligent finally.