Fuzzy neural network technique for system state forecasting.

Area of Science:

• Computational intelligence
• System dynamics and forecasting
• Machine learning for time series analysis

Background:

• Traditional forecasting methods struggle with multiple datasets, exhibiting limited modeling capacity and opaque reasoning.
• System state forecasting often relies on diverse datasets from distinct operational conditions.
• Accurate prediction requires effective information extraction from heterogeneous data sources.

Purpose of the Study:

• To propose a novel fuzzy neural network (FNN) for enhanced system state forecasting.
• To improve forecasting accuracy by effectively extracting information from multiple datasets.
• To introduce a new parameter estimation technique for improved modeling.

Main Methods:

• Development of a fuzzy neural network (FNN) integrating autoregressive (AR) nodes for linear correlations and nonlinear nodes for nonlinear correlations.
• Introduction of the Laplace particle swarm (LPS) method for efficient parameter estimation and enhanced modeling accuracy.
• Validation through diverse applications including Mackey-Glass data, exchange rate prediction, and gear system prognosis.

Main Results:

• The developed FNN effectively captures linear and nonlinear dynamics within multiple datasets.
• The Laplace particle swarm (LPS) method significantly improves parameter estimation and overall modeling accuracy.
• The FNN-LPS approach demonstrates superior performance in system state forecasting across various test cases.

Conclusions:

• The proposed fuzzy neural network (FNN) combined with the Laplace particle swarm (LPS) method offers a robust solution for system state forecasting with multiple datasets.
• This integrated approach effectively models complex system dynamics and improves predictive accuracy.
• The methodology shows promise for real-world applications in financial forecasting and predictive maintenance.