A Dini-Derivative-Aided Zeroing Neural Network for Time-Variant Quadratic Programming Involving Multi-Type...

Area of Science:

• Optimization and Control Theory
• Computational Neuroscience
• Robotics

Background:

• Time-variant quadratic programming (QP) with multi-type constraints is common in various applications.
• Existing zeroing neural network (ZNN) solvers for QP have limitations, including potential failure, approximate solutions, and complex parameter tuning.

Purpose of the Study:

• To introduce a novel ZNN solver, Dini-ZNN, for time-variant QPs with multi-type constraints.
• To overcome the limitations of existing ZNNs by utilizing a continuous, non-differentiable projection operator.

Main Methods:

• The proposed Dini-ZNN solver employs the upper right-hand Dini derivative of a projection operator as a mode switcher.
• Theoretical analysis rigorously proves the convergence and optimality of the Dini-ZNN solver.
• Comparative validations assess the performance against existing methods.

Main Results:

• The Dini-ZNN solver demonstrates guaranteed problem-solving capability and high solution accuracy.
• The novel solver eliminates the need for hyperparameter tuning, simplifying its application.
• Effectiveness is verified through comparative studies.

Conclusions:

• The Dini-ZNN solver offers a robust and efficient solution for time-variant QPs with complex constraints.
• Its successful application in robot kinematic control highlights its practical utility.
• This work advances ZNN-based optimization techniques.