Adaptive Sliding Mode Control of Dynamic Systems Using Double Loop Recurrent Neural Network Structure.

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Neural Networks

Background:

• Controlling nonlinear dynamic systems often requires accurate modeling of unknown system dynamics.
• Traditional neural networks (NNs) may have limitations in capturing complex internal states and output signals simultaneously.
• Recurrent neural networks (RNNs) offer potential but can be improved for enhanced approximation capabilities.

Purpose of the Study:

• To propose a novel adaptive sliding mode control system utilizing a double loop recurrent neural network (DLRNN).
• To enhance the approximation performance for unknown nonlinear system dynamics.
• To validate the effectiveness of the DLRNN-based controller on a microelectromechanical system (MEMS) gyroscope.

Main Methods:

• A new three-layer RNN structure (DLRNN) with dual feedback loops storing firing weights and output signals was developed.
• The DLRNN was integrated into an equivalent controller to approximate unknown nonlinear dynamics.
• Online adaptive laws were used to update DLRNN parameters for optimal approximation.

Main Results:

• The DLRNN demonstrated superior approximation performance compared to standard NNs and single-loop RNNs.
• The adaptive sliding mode controller with DLRNN achieved good tracking properties when applied to a MEMS gyroscope.
• The DLRNN accurately and rapidly estimated unknown dynamics, exhibiting more stable internal states than other NNs.

Conclusions:

• The proposed DLRNN structure effectively captures internal states and output signals, leading to enhanced approximation.
• The adaptive sliding mode control system incorporating DLRNN provides a robust solution for controlling nonlinear dynamic systems.
• The DLRNN offers a promising approach for accurate and stable modeling in applications like MEMS gyroscope control.