Overview
A comprehensive research implementation that combines Long Short-Term Memory (LSTM) neural networks with an Adaptive Neuro-Fuzzy Inference System (ANFIS) to control nonlinear magnetic levitation systems. This hybrid architecture leverages the strengths of both deep learning (temporal pattern recognition) and fuzzy logic (interpretable control rules).
Architecture
The system utilizes a dual-stage control strategy:
- LSTM Network: Captures temporal dependencies and sequence dynamics from the system’s state history.
- ANFIS Controller: Takes the LSTM predictions along with current error states to generate the final control signal.
Why Hybrid?
- LSTM excels at modeling complex, non-linear time-series data but lacks interpretability.
- ANFIS provides transparent, rule-based control but can struggle with long-term temporal dependencies.
- Integration: By feeding LSTM insights into the ANFIS structure, the controller achieves superior tracking performance and disturbance rejection compared to either method alone.
Implementation Details
-
Data Pipeline: End-to-end MATLAB pipeline including data chunking, outlier removal, and normalization.
-
Visualization: Extensive plotting of membership functions, error surfaces, and step responses to validate stability.