Real-time model predictive control using a self-organizing neural network.

Area of Science:

• Control Engineering
• Artificial Intelligence
• Nonlinear System Dynamics

Background:

• Model predictive control (MPC) often faces computational challenges with complex nonlinear systems.
• Accurate predictive modeling is crucial for effective MPC performance.
• Existing methods may struggle with real-time implementation for highly dynamic systems.

Purpose of the Study:

• To propose a real-time model predictive control (RT-MPC) strategy for nonlinear systems.
• To enhance computational efficiency and modeling accuracy in MPC.
• To ensure stability and satisfactory performance in closed-loop systems.

Main Methods:

• Development of a self-organizing radial basis function neural network (SORBFNN) for concurrent structure and parameter learning, serving as the predictive model.
• Enhancement of a fast gradient method (GM) to reduce computational cost and enable online suboptimization of the RT-MPC.
• Stability analysis and steady-state performance evaluation of the proposed closed-loop system.

Main Results:

• The SORBFNN model significantly improves prediction accuracy with uniformly ultimately bounded modeling error.
• The enhanced GM effectively reduces computational burden, allowing for online RT-MPC optimization.
• Numerical simulations and experimental results confirm satisfactory tracking and disturbance rejection capabilities.

Conclusions:

• The proposed RT-MPC based on SORBFNN offers an efficient and effective solution for controlling nonlinear systems.
• The integration of SORBFNN and enhanced GM addresses computational complexity while maintaining high performance.
• The method demonstrates practical applicability and robustness through experimental validation.