求解时变非线性优化的平滑预设时间零化神经网络及其应用OA

In existing prescribed-time zeroing neural networks,the derivative of the computational error with respect to time was found to be discontinuous at the prescribed time when solving time-varying nonlinear optimization problems,which could easily lead to actuator chattering in practical applications.Therefore,a smooth prescribed-time zeroing neu-ral network was proposed,which ensured that the derivative of the error trajectory remained continuous at the prescribed time.Furthermore,a whole-process smooth prescribed-time zeroing neural network was proposed,which ensured the smoothness of the error trajectory at the initial time and thereby achieved a smooth transition over the entire time do-main.Theoretical analysis was conducted on the prescribed-time convergence and smoothness of the computational error for the two proposed networks,and the impact of the smoothing gain on the networks'convergence performance was discussed.For the repetitive motion planning problem of a redundant manipulator with initial error,a smooth repetitive motion planning scheme was proposed and solved using the proposed smooth prescribed-time zeroing neural network.The smooth operation of the manipulator was ensured by this computational scheme,and convergence of the end-effector position error was achieved.Simulation results of time-varying nonlinear optimization and the redundant manipulator validate the effectiveness of the proposed networks and repetitive motion planning computational scheme.