通讯访问约束下的线性系统参数可辨识性

On parameter identifiability of linear multivariable systems under communication access constraints

考察了线性离散时间随机系统在访问约束下的模型参数在线辨识问题.对于参数的变化为Gauss-Markov过程描述的时变线性多变量系统,将其输入输出模型转化为状态空间模型,即可使时变参数的在线辨识问题转化为系统状态估计问题.对于系统中存在的因访问约束造成的有限通讯,采用一个二值向量函数描述系统中各个信号的通讯状态,其中"1"表示该信号被传输,"0"表示通讯中断.运用信息论的测度和方法,根据由互信息定义的可辨识性概念,讨论了线性时变系统的参数在线可辨识性问题,得出在输出检测信号传输存在访问接入约束情况下的参数可辨识性条件.理论分析和仿真结果表明:在通讯访问受限的条件下,系统存在合理的通讯接入策略,使其保持模型参数的可辨识性.

The problem of on-line parameter identification was discussed for linear multivariable discrete time stochastic systems under communication access constraints. Under the assumption that the time- varying parameter can be modeled as a Gauss-Markov process, the input-output model of the discussed systems is transformed to the state space description, and then the problem of parameter identification is transformed to the problem of state estimation. To deal with the limited communication caused by access constraints, a binary-valued vector function is adopted to describe the sensors＇ access status, where 1 means ＂accessing＂ and 0 means ＂not accessing＂. By using the information theoretic measures and approaches, the concept of on-line parameter identifiability, which was defined by mutual information, was discussed under the sensors＇ access constraints, and a condition for parameter identifiability was derived. Theoretical analysis and simulation results show that there is a proper communication strategy preserving the identifiability of the system under access constraints.