Forecasting of Multidimensional Time Series Basing on Fuzzy Rule-Based Models
| Autor: | Vadim Borisov, Victor Luferov |
|---|---|
| Rok vydání: | 2019 |
| Předmět: |
0209 industrial biotechnology
Mathematical optimization Fuzzy rule Series (mathematics) Basis (linear algebra) Computer science media_common.quotation_subject 02 engineering and technology Task (project management) Variety (cybernetics) Interdependence Reduction (complexity) 020901 industrial engineering & automation 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing media_common Parametric statistics |
| Zdroj: | 2019 XXI International Conference Complex Systems: Control and Modeling Problems (CSCMP). |
| Popis: | The limitations of existing models and methods for forecasting multidimensional time series are, first of all, the complexity of integration of the indirect influence of interdependent time series under uncertainty, and also reduction of this complex task to forecasting the totality of individual one-dimensional time series. A variety of fuzzy rule-based cognitive models is used to forecast multidimensional time series under uncertain conditions. And the direct and indirect influence of interdependent components of these series is considered. An approach to the structural-parametric adjustment of fuzzy rule-based cognitive models for solving problems of forecasting multidimensional time series is presented. The approach provides: firstly, the parametric adjustment of neuro-fuzzy models of mutual influence weights for all concepts of this model; secondly, the structural adjustment of the model due to changes in the “retrospective depth” for the components of multidimensional time series. The accuracy of forecasting of the multidimensional time series is estimated on the basis of the proposed models. It is compared with the existing approaches based on the forecasting of such series, represented by sets of one-dimensional time series. The greater accuracy of forecasting the multidimensional time series by the developed models is achieved by: first, taking into account the direct and indirect influence of interdependent components of time series under uncertain conditions; secondly, the implementation of flexible structural and parametric adaptation of the model; thirdly, the possibility of setting different “retrospective depths” for components of multidimensional time series. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|