Enhanced imitation learning of robust nonlinear model predictive control via temporal convolutional neural network...

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Control Systems

Background:

• Direct current (DC) shipboard microgrids face control challenges due to hybrid energy sources and increased complexity.
• Instability can arise without grid support, necessitating advanced control strategies.

Purpose of the Study:

• To develop an enhanced temporal convolutional neural network (TCNN) to approximate the control policy for a nonlinear model predictive control (MPC) framework.
• To ensure robust voltage stabilization and stability in DC shipboard microgrids.

Main Methods:

• An enhanced TCNN was developed to approximate the control policy within a nonlinear MPC framework.
• Lyapunov-based stability analysis confirmed input-to-state stability under bounded disturbances.
• Mean-field analysis optimized neural network layer initialization for bounded and stable parameters.

Main Results:

• The TCNN controller achieved high accuracy (R² near unity, MSE of 0.07) and fast inference (10 ms).
• Controller-hardware-in-the-loop tests showed rapid voltage restoration (22.4 ms) with reduced overshoot (1.67%).
• Compared to state-of-the-art, the TCNN reduced overshoot by 62% and improved settling time by 53.25%.

Conclusions:

• The proposed TCNN controller offers a stable, accurate, and efficient solution for DC shipboard microgrid voltage control.
• Its real-time performance and superior disturbance rejection make it suitable for advanced naval power systems.