A Penalty Strategy Combined Varying-Parameter Recurrent Neural Network for Solving Time-Varying Multi-Type...

Area of Science:

• Optimization
• Machine Learning
• Control Theory

Background:

• Time-varying quadratic programming (TVQP) problems present significant computational challenges.
• Existing methods struggle with complex constraints, including inequalities and bounded conditions.

Purpose of the Study:

• To propose and analyze a novel penalty strategy combined varying-parameter recurrent neural network (PS-VP-RNN) for solving TVQP problems.
• To address TVQP problems with equality and multitype inequality constraints.

Main Methods:

• A novel penalty function transforms inequality constraints into a penalty term within the objective function.
• A varying-parameter recurrent neural network (VP-RNN) is designed to incorporate this penalty term for TVQP.
• The global convergence of the proposed PS-VP-RNN is theoretically proven.

Main Results:

• The PS-VP-RNN successfully solves TVQP problems with equality constraints.
• The method demonstrates capability in handling inequality and bounded constraints.
• Numerical simulations confirm the effectiveness and accuracy of the PS-VP-RNN.

Conclusions:

• The PS-VP-RNN offers a robust and accurate approach for solving complex TVQP problems.
• This method provides a unified framework for TVQP with various constraint types.
• The validated effectiveness opens avenues for applications in dynamic optimization and control systems.