Online Nash-optimization tracking control of multi-motor driven load system with simplified RL scheme

Area of Science:

• Control Engineering
• Artificial Intelligence
• Systems Science

Background:

• Existing optimal tracking control solutions primarily focus on single-input systems.
• Multi-motor driven load systems (MMDLS) present unique challenges due to differing motor dynamics.

Purpose of the Study:

• To address the optimal tracking control problem (OTCP) for unknown MMDLS.
• To develop a control strategy that achieves a Nash equilibrium for all motor inputs.
• To optimize individual performance indices as a result of the Nash equilibrium.

Main Methods:

• Utilized an identifier to reconstruct MMDLS dynamics, eliminating the need for an accurate system model.
• Employed a simplified reinforcement learning (RL) structure with critic neural networks (NNs) to approximate cost functions.
• Designed steady-state and RL-based controls using identified dynamics and Nash optimization.
• Implemented a novel adaptation algorithm for adaptive optimization of NN weights and learning gains.

Main Results:

• Demonstrated weight convergence for both the identification algorithm and the RL structure.
• Proved the stability of the Nash-optimized MMDLS.
• Numerical simulations confirmed the correctness and superior performance of the proposed methods.

Conclusions:

• The proposed RL-based framework effectively solves the OTCP for unknown MMDLS.
• Achieving a Nash equilibrium enables simultaneous optimization of individual motor performance.
• The adaptive learning gains enhance the robustness and efficiency of the control system.