Sliding Integral Neural Network Driven Robust Solution for Time-Varying Quadratic Programming

Time-varying quadratic programming (TVQP) requires efficient, accurate, and robust online solvers. Existing discrete-time (DT) recurrent neural networks (RNNs), however, often face a tradeoff between solution precision and noise immunity. To address this issue, a sliding integral neural network (SINN) is proposed. In particular, the interior-point (IP) method is extended to a dynamic IP (DIP) formulation with a time-varying barrier coefficient, and TVQP is reformulated as a DT error-feedback system for closed-loop design. A sliding integral control law with forward harmonic vectors is then developed to compensate for iterative differential residuals. Theoretical analyses show that the SINN achieves O(τ ⁴) steady-state accuracy, where τ denotes the sampling interval, and suppresses structured disturbances with sublinear/linear growth. Numerical and robotic arm motion-planning simulations demonstrate its high precision and robustness.