离散时间MDP矩最优模型——折扣依赖于历史的情形

本文在S、A(i)(i∈S)均匀可列集情形下,建立了折扣依赖于历史的矩最优模型。给出了折扣总报酬k阶矩在各类策略下的统一表达式;讨论了矩最优策略的结构与性质;证明了矩最优方程在给定条件下,存在唯一的有界解。

连续时间折扣矩最优模型及其与离散时间拟折扣矩最优模型的关系——Q 矩阵族未必保守的情形

本文是首次在转移率矩阵族为一般 Q 矩阵族(未必保守亦未必一致有界)的条件下,研究状态空间与决策集均为可数集的连续时间折扣矩最优模型(M_k-CTMDP);提出离散时间折扣依赖于状态与决策的拟折扣矩最优模型(β_k-GTMDP);并揭示二者之间的关系;给出在 f~∞下折扣总报酬 k 阶矩向量 μ_k(f)满足:kαμ_k(f)=kr(f)(?)μ_(k-1)(f)+Q(f)μ_k(f)及μ_k(f)=kP^(min)(kα,f)(r(f)(?)μ_(k-1)(f))的简洁表达式;给出报酬矩最优是矩最优方程组唯一有界解的一个很弱的充分条件与解法;给出矩最优策略存在的充要条件与若干性质.本文结果对 MDP 理论的发展与应用有重要意义,而且对跳跃型马氏过程的一类积分型泛函的研究与应用也颇有意义.

Stochastic Optimal Control of First-Passage Failure for Rectangular Thin Plate Vibration Model under Gaussian White-Noise Excitations

A rectangular thin plate vibration model subjected to inplane stochastic excitation is simplified to a quasinonintegrable Hamiltonian system with two degrees of freedom. Subsequently a one-dimensional Ito stochastic differential equation for the system is obtained by applying the stochastic averaging method for quasi-nonintegrable Hamiltonian systems. The conditional reliability function and conditional probability density are both gained by solving the backward Kolmogorov equation numerically. Finally, a stochastic optimal control model is proposed and solved. The numerical results show the effectiveness of this method.