Distributed multi-timescale algorithm for nonconvex optimization problem: A control perspective.

Area of Science:

• Control Theory
• Optimization
• Distributed Systems

Background:

• Distributed optimization problems often involve complex, nonconvex objective and constraint functions.
• Existing methods may struggle with ensuring both solution optimality and system stability in such scenarios.

Purpose of the Study:

• To develop a novel distributed control approach for nonconvex constrained optimization.
• To ensure both optimality and stability in distributed systems with complex functions.

Main Methods:

• A virtual reference-based convex penalty function was integrated into the augmented Lagrangian.
• A two-timescale distributed primal-dual approach was designed using a consensus scheme.
• The method employs a slower subsystem for optimality and a faster subsystem for stability.

Main Results:

• The proposed method effectively addresses distributed nonconvex constrained optimization problems.
• Demonstrated ability to achieve both optimality and stability through distinct subsystems.
• Effectiveness validated across three distinct illustrative cases.

Conclusions:

• The developed two-timescale distributed approach offers a robust solution for challenging optimization tasks.
• This control-theoretic perspective provides a new framework for distributed nonconvex optimization.
• The method shows significant promise for applications requiring stable and optimal distributed control.