Memorised quasi-time-fuel-optimal feedback control of perturbed double integrator (Q1614405)

Given a dynamic controlled object \[ \ddot{x}=f(\dot{x},x,t)+g(\dot{x},x,t)u \;\;\;\;\;\;\;(1) \] where \(f: \mathbb{R}^{2} \times \mathbb{R}_{+} \rightarrow \mathbb{R}\), \(g: \mathbb{R}^{2} \times \mathbb{R}_{+} \rightarrow \mathbb{R}\), \ \ \(u \in U=\{-1, 0, +1\}\), the functions \(f\), \(g\) are constrained as follows: \[ | f(\dot{x},x,t)| \leq F, \;\;\;0<G_{1} \leq g(\dot{x}, x, t) \leq G_{2}, \;\;\;G_{1} \leq G \leq G_{2}, \;\;\;{F+G_{2}+G_{1} \over {G}} < {1 \over 2}\;\;\;(2) \] The authors look for a control function that minimizes the integral criterion \[ J=\int_{t_{0}}^{t_{f}}(\rho+| u| )dt,\;\rho > 0 \] where \(t_{0}\) is the starting time, \(t_{f}\) is the target time when reaching \((0,0)\). Because of very weak regularity of the right-hand side of (1), the strict optimal control design is practically impossible. Thus, the presented concept, called memorised feedback control, is based on the classical solution of the deterministic problem of the object \(\ddot{x}=u,\;\;| u| \leq 1\). Here, robust convergence of the feedback system has been proved using sequence theory. The comparison of the obtained results with the control of the closed-loop structure of the system \(\dot{x}_{1}=x_{2},\;\dot{x}_{2}=u\) shows that the new controller overcomes overshoot and sliding in the controlled process (1). Moreover, applying the proposed quasi-optimal control requires simple technical means, only.