基于新息的最小方差对偶控制

针对参数未知且随机变化的系统的最小方差控制问题，提出了基于新息的最小方差对偶控制策略，将系统参数随机变化的最小方差控制问题转化为多个基于新息的单步控制问题，利用卡尔曼滤波对参数进行估计，每步仅需求解一个非常简单的优控制问题，然后把控制加入系统，根据系统的输出获得新息，再重复卡尔曼滤波，最终可实现最小方差控制。仿真结果表明，该控制器具有对偶作用，一方面通过学习可消除系统参数的不确定性，另一方面可使系统输出稳定在期望输出值上，从而能够获得较好的参数辩识结果与良好的控制目标。

Dual Control Based on Innovations for the Minimum Variance Problem

Aiming at the system with unknown parameters that are stochastically time variant, this paper presents the dual control strategy based on innovations for the minimum variance problem, in which the problem is converted into multiple single step control problems based on innovations. In this paper, parameters are estimated using Kalman filter and in every step only a very simple optimizing control problem will be solved. Control signal is applied to the system, then innovations are obtained according to system output and then Kalman filter is repeated. Thus the minimum variance control can be realized. Simulation results show that the controller has dual property: on one hand eliminating the uncertainty of system parameters through learning and on the other hand, stabilizing system output at the expected level, thereby achieving preferable parameter estimation results and control objective.