Research On Non-Linear Time Series Prediction Based On Intuitionistic Fuzzy Cognitive Maps

Time series are a collection of data with temporal relationships,which are widely existed in nature and social production activities.Subject to the internal/external influences,time series are usually characterized by non-linearity,uncertainty,and incompleteness.How to learn the potential patterns from historical data and make prediction for the future trends have been a hot topic for researchers.Fuzzy cognitive maps,as a soft computing method with knowledge representation and inference capability,are widely used in time series prediction.This work makes further improvements based on fuzzy cognitive map to analyze and study time series,mainly including the following contents.(1)In order to effectively model time series under uncertain environment,a novel higher-order intuitionistic fuzzy cognitive map is proposed.Intuitionistic fuzzy set is introduced into higher-order fuzzy cognitive maps to perform feature learning on time series with long temporal dependencies.The proposed method not only effectively improves the knowledge representation and learning ability for uncertain data containing strong noise and high fluctuations,but also enhances the ability for long temporal dependencies modeling.In addition,to capture the fluctuation characteristics of the data,variational mode decomposition is utilized to decompose time series into sequences of various frequencies,and fine features on different scales can be obtained.(2)The combination of reservoir computing and granular mechanism is introduced into fuzzy cognitive map,based on which a novel prediction model is proposed for multi-dimensional time series.While preserving the advantages of fuzzy cognitive maps,such as interpretability and causal inference,reservoir computing can be utilized to enhance the learning capability of streaming data and realize the learning of large-scale data features with an efficient way.Meanwhile,in order to tackle the high fluctuations existing in time series and enhance the robustness of model training,the type of weights of fuzzy cognitive graphs is improved by granular technique.The above intelligent model has achieved promising performance in various real-world datasets,such as traffic time series and EEG signals.