基于量化输出和信道通讯的系统辨识与状态估计

With the fast development of Internet of Things and other new-high technologies, it has attracted a lot of attention to study the system identification and state estimation with quantized outputs and communication channels. Most of the existing work focuses on either the quantization or the communication unreliability, but the research subject to both them is relatively rare. The main goal of this project is to establish identification algorithms and state estimators with quantized outputs and communication channels, and complete the performance analysis. For several kinds of typical quantized output systems, identification algorithms will be designed to estimate the unknown parameters in the cases that the channel transfer matrix is known and unknown, respectively. Under the persistently exciting inputs, we will prove the convergence, give the convergence rate, and discuss the optimality of the algorithms. Considering both the quantization and communication uncertainties, the minimum mean-squared error estimators will be established for multiple-input and single-output systems, we will also analyze the stability, convergence and robustness and so on, and extend the results to the cases of multiple-input and multiple-output systems and distributed estimation. The results and methods obtained will be applied to the multi-agent systems and networked control systems, and the consensus algorithms will be given based on quantized measurements and communication channels.

伴随着物联网等高新科学技术的兴起，基于量化输出和信道通讯的控制理论受到了国内外学者的广泛关注，现有的相关研究大多数单一地侧重于量化或者信道通讯，把两者结合在一起的理论还相对匮乏，亟待发展和完善。本项目旨在建立基于量化输出和信道通讯的辨识算法和状态估计器，并完成对它们的性能分析。具体内容包括：针对几类典型的量化输出系统，分别在信道转移概率矩阵已知和未知的情况下构造辨识算法，在持续激励输入条件下证明算法的收敛性，给出估计误差的收敛速度，并讨论算法的最优性；在同时考虑量化误差和信道不确定性的前提下，建立多输入单输出离散线性随机系统的最小均方误差估计器，给出稳定性、收敛性、鲁棒性等性能分析，进而把取得的结果拓展到多输入多输出系统和分布式估计的情形；相关结果和方法在多自主体系统和网络控制系统中的应用，给出基于量化测量和信道通讯的趋同算法等。

实际当中，由于测量手段、生产成本或元器件的能力等限制，一定条件下的系统输出不得不量化成低精度的数据，然后通过信道传输给接收者。一般来讲，接收到的可用数据与测量到的输出存在较大差距，这就向传统的控制理论提出了新的要求和挑战。已有研究大多数单一地侧重于量化或者信道通讯，把两者结合在一起的理论还相对匮乏，亟待发展和完善。本项目旨在建立基于量化输出和信道通讯的辨识算法和状态估计器，并完成对它们的性能分析。..主要内容和结果包括：1）基于量化输出和信道通讯的系统辨识。设计了相应系统的辨识算法，证明了收敛性，给出了收敛速度，讨论了量化阈值、系统噪声、信道不确定性等对辨识精度的影响；2）基于量化输出和信道通讯的状态估计。针对线性离散时不变系统，推导出了递推形式的最小均方误差估计器，分析了预测误差协方差矩阵的性质，讨论了量化策略和信道不确定性对估计质量的综合影响等。..此外，还研究了有限脉冲系统在量化输入和量化输出下的辨识问题，给出了渐近有效的参数估计算法，并把取得的结果拓展到了Wiener系统和Hammerstein系统等非线性系统；提出了面向决策的系统辨识研究框架，设计了资源分配算法，建立了其收敛性能；在不规则采样和随机采样下研究了线性连续时间系统的辨识问题。..本项目资助发表SCI期刊论文10篇，其中IEEE Transactions on Automatic Control长文1篇，Automatica 长文和短文各2篇；以及会议论文5篇。