Numerical methods for approximating functional gains in LQR boundary control problems (Q5931751): Difference between revisions

This paper deals with numerical methods for solving the optimal control problem with partial differential equation systems. In section 2, the optimal control problem for the one-dimensional heat equation with boundary control is formulated. The optimal control is represented as \(u(t)= Kz(t)= -B^*\Pi z(t)\), where \(K\) is a feedback gain operator while \(\Pi\) is the solution to an ARE. \(K\) is represented as \(K\phi= \int^1_0 h(\xi)\phi(\xi) d\xi\), where \(h(\xi)\) is called the functional gain. The Riccati method is described in section 3. \(\Pi\) is represented as \([\Pi\phi](x)= \int^1_0 p(x,\xi) \phi(\xi) d\xi\). It is shown that the kernel \(p(x,\xi)\) satisfies an R-PDE. The functional gain \(h(\xi)\) is determined as \(h(\xi)= -[B^*_xp](x, \xi)= [{\partial p(0,\xi)\over \partial x},- {\partial p(1,\xi)\over \partial x}]^T\). Section 4 describes Chandrasekhar's method. Chandrasekhar equations are defined by \[ \dot K(t)= B^*L^*(t) L(t),\quad \dot L(t)= -L(t)[\widetilde A+\alpha I- BK(t)]. \] Chandrasekhar partial differential equations (C-PDE) are derived similarly in spirit to the R-PDE. In section 5, numerical methods for solving \(h(\xi)\) are considered. Riccati and Chandrasekhar equations are converted to ones with finite dimension by using finite element methods. Three finite element methods are considered: the standard method, Burns-Kang's method, and Nitsche's method. Numerical results are shown in section 6. Example 1 compares the approximate functional gains obtained by solving approximate ARE using the three finite element schemes. Nitsche's scheme provides more accurate responses than the other two schemes. In example 2, the result from the standard finite element scheme to solve a PDE is compared with Nitsche's solution to the ARE.