第四章:能控性与能观测性
一、引言
在分析和设计控制系统时,有两个基本问题需要回答:
- 能控性 (Controllability):在有限时间内,控制输入 \(\textbf{u}(t)\) 能否将系统的状态 \(\textbf{x}(t)\) 从任意初始状态 \(\textbf{x}(t_0)\) 转移到任意期望的终端状态 \(\textbf{x}(t_f)\)?这反映了输入对状态的控制能力。
- 能观测性 (Observability):在有限时间内,能否通过观测系统的输入 \(\textbf{u}(t)\) 和输出 \(\textbf{y}(t)\) 来唯一地确定系统的初始状态 \(\textbf{x}(t_0)\)(进而确定任意时刻的状态 \(\textbf{x}(t)\))?这反映了输出对状态的观测能力。
这两个概念由卡尔曼 (R.E. Kálmán) 提出,是现代控制理论的基石。
- 状态方程:\(\dot{\textbf{x}}(t) = A\textbf{x}(t) + B\textbf{u}(t)\) (反映输入对状态的影响)
- 输出方程:\(\textbf{y}(t) = C\textbf{x}(t) + D\textbf{u}(t)\) (反映输出对状态和输入的依赖)
- 系统响应:\(\textbf{x}(t) = e^{A(t-t_0)}\textbf{x}(t_0) + \int_{t_0}^t e^{A(t-\tau)}B\textbf{u}(\tau)d\tau\)
二、状态能控性 (State Controllability)
1. 定义
对于线性定常系统 \(\dot{\textbf{x}} = A\textbf{x} + B\textbf{u}\),如果在有限时间 \([t_0, t_f]\) 内,存在一个容许控制输入 \(\textbf{u}(t)\),能够将系统从任意初始状态 \(\textbf{x}(t_0) = \textbf{x}_0\) 转移到任意终端状态 \(\textbf{x}(t_f) = \textbf{x}_f\),则称系统是状态完全能控的 (completely state controllable),简称能控的。
如果只能将任意初始状态 \(\textbf{x}_0\) 转移到原点 \(\textbf{x}_f = \textbf{0}\),则称为能达原点的 (reachable to the origin)。对于线性系统,状态完全能控等价于能从原点到达任意状态,也等价于能将任意状态驱动到原点。
2. 能控性判据
a. 格拉姆矩阵判据 (Gramian Matrix Criterion)
系统 \((\textit{A, B})\) 状态完全能控的充分必要条件是:对于任意 \(t_f > t_0\) (通常取 \(t_0=0, t_f=T>0\)),能控性格拉姆矩阵 (Controllability Gramian Matrix) \[ W_c(t_0, t_f) = \int_{t_0}^{t_f} e^{A(t_0-\tau)}BB^T e^{A^T(t_0-\tau)} d\tau \] 或 (更常用的形式,令 \(\sigma = \tau-t_0\)) \[ W_c(T) = \int_{0}^{T} e^{A\tau}BB^T e^{A^T\tau} d\tau \] 是非奇异的 (即满秩的,\(\text{rank}(W_c(T)) = n\))。
- 若系统能控,则对所有 \(T>0\),\(W_c(T)\) 都是非奇异的。
- 若 \(W_c(T)\) 非奇异,则可以将任意初始状态 \(x_0\) 在时间 \(T\) 内转移到原点的控制律为: \(\textbf{u}(t) = -B^T e^{-A^T t} W_c^{-1}(0,T) e^{At_0} \textbf{x}_0\) (这里假设 \(t_0\) 为初始时刻,目标为 \(t_f=T\) 时刻到达原点,且积分变量为 \(t\)。PPT P18 的公式是 \(u(t) = -B^T e^{-A^T t} W_c^{-1}(0,T) x_0\) 驱动 \(x(0)=x_0\) 到 \(x(T)=0\))。
b. 卡尔曼能控性判据 (Kalman's Rank Condition)
系统 \((\textit{A, B})\) 状态完全能控的充分必要条件是:\(n \times nr\) 维能控性矩阵 (Controllability Matrix) \[ Q_c = \begin{bmatrix} B & AB & A^2B & \dots & A^{n-1}B \end{bmatrix} \] 的秩为 \(n\) (即 \(\text{rank}(Q_c) = n\))。
推导思路 (参考 chapter4_Controllability_and_Observability.pdf P22-P23):
令 \(t_0=0\),目标是将初始状态 \(\textbf{x}(0)=\textbf{x}_0\) 转移到终端状态 \(\textbf{x}(t_f)=\textbf{0}\)。
\(\textbf{0} = e^{At_f}\textbf{x}_0 + \int_0^{t_f} e^{A(t_f-\tau)}B\textbf{u}(\tau)d\tau\)
\(\textbf{x}_0 = -\int_0^{t_f} e^{-A\tau}B\textbf{u}(\tau)d\tau\)
根据凯莱-哈密顿定理,\(e^{-A\tau} = \sum_{k=0}^{n-1} \alpha_k(\tau)A^k\)。
\(\textbf{x}_0 = -\sum_{k=0}^{n-1} A^k B \int_0^{t_f} \alpha_k(\tau)\textbf{u}(\tau)d\tau\)
令 \(\boldsymbol{\beta}_k = -\int_0^{t_f} \alpha_k(\tau)\textbf{u}(\tau)d\tau\),则
\(\textbf{x}_0 = \sum_{k=0}^{n-1} A^k B \boldsymbol{\beta}_k = \begin{bmatrix} B & AB & \dots & A^{n-1}B \end{bmatrix} \begin{bmatrix} \boldsymbol{\beta}_0 \\ \boldsymbol{\beta}_1 \\ \vdots \\ \boldsymbol{\beta}_{n-1} \end{bmatrix} = Q_c \boldsymbol{\beta}\)
要使对于任意 \(\textbf{x}_0\) 该方程组都有解 \(\boldsymbol{\beta}\) (进而能找到对应的 \(\textbf{u}(t)\)),必须 \(\text{rank}(Q_c)=n\)。
例题 (参考P24) \(\dot{\textbf{x}} = \begin{bmatrix} 1 & 2 & 1 \\ 0 & 1 & 0 \\ 1 & 0 & 3 \end{bmatrix}\textbf{x} + \begin{bmatrix} 1 & 0 \\ 0 & 1 \\ 0 & 0 \end{bmatrix}\textbf{u}\) \(A = \begin{bmatrix} 1 & 2 & 1 \\ 0 & 1 & 0 \\ 1 & 0 & 3 \end{bmatrix}, B = \begin{bmatrix} 1 & 0 \\ 0 & 1 \\ 0 & 0 \end{bmatrix}\) \(AB = \begin{bmatrix} 1 & 2 \\ 0 & 1 \\ 1 & 0 \end{bmatrix}\) \(A^2B = A(AB) = \begin{bmatrix} 1 & 2 & 1 \\ 0 & 1 & 0 \\ 1 & 0 & 3 \end{bmatrix}\begin{bmatrix} 1 & 2 \\ 0 & 1 \\ 1 & 0 \end{bmatrix} = \begin{bmatrix} 2 & 4 \\ 0 & 1 \\ 4 & 2 \end{bmatrix}\) \(Q_c = \begin{bmatrix} B & AB & A^2B \end{bmatrix} = \begin{bmatrix} 1 & 0 & 1 & 2 & 2 & 4 \\ 0 & 1 & 0 & 1 & 0 & 1 \\ 0 & 0 & 1 & 0 & 4 & 2 \end{bmatrix}\) 由于 \(Q_c\) 是 \(3 \times 6\) 矩阵,其秩最大为3。通过观察前三列 \(\begin{bmatrix} 1 & 0 & 1 \\ 0 & 1 & 0 \\ 0 & 0 & 1 \end{bmatrix}\) 已经满秩,所以 \(\text{rank}(Q_c)=3=n\)。系统能控。
例题:倒立摆系统 (参考P27) \(A = \begin{bmatrix} 0 & 1 & 0 & 0 \\ 0 & 0 & -1 & 0 \\ 0 & 0 & 0 & 1 \\ 0 & 0 & 11 & 0 \end{bmatrix}, B = \begin{bmatrix} 0 \\ 1 \\ 0 \\ -1 \end{bmatrix}\) \(AB = \begin{bmatrix} 1 \\ 0 \\ -1 \\ 0 \end{bmatrix}, A^2B = \begin{bmatrix} 0 \\ 1 \\ 0 \\ -11 \end{bmatrix}, A^3B = \begin{bmatrix} 1 \\ 0 \\ -11 \\ 0 \end{bmatrix}\) \(Q_c = \begin{bmatrix} 0 & 1 & 0 & 1 \\ 1 & 0 & 1 & 0 \\ 0 & -1 & 0 & -11 \\ -1 & 0 & -11 & 0 \end{bmatrix}\) 计算其行列式或进行初等行变换可知 \(\text{rank}(Q_c)=4=n\)。系统能控。
c. PBH (Popov-Belevitch-Hautus) 秩判据
系统 \((\textit{A, B})\) 状态完全能控的充分必要条件是:对于 \(A\) 的所有特征值 \(\lambda_i\) (包括重根),矩阵
\[ \begin{bmatrix} \lambda_i I - A & B \end{bmatrix} \]
的秩均为 \(n\) (即行满秩)。
或者等价地说,不存在 \(A\) 的左特征向量 \(\textbf{v}^T \neq \textbf{0}\) (即 \(\textbf{v}^T A = \lambda \textbf{v}^T\)) 使得 \(\textbf{v}^T B = \textbf{0}\)。
例题 (参考P31)
\(\dot{\textbf{x}} = \begin{bmatrix} -7 & 0 & 0 \\ 0 & -5 & 0 \\ 0 & 0 & -1 \end{bmatrix}\textbf{x} + \begin{bmatrix} 2 \\ 0 \\ 9 \end{bmatrix}u\)
特征值为 \(\lambda_1=-7, \lambda_2=-5, \lambda_3=-1\)。
对于 \(\lambda_1=-7\): \(\begin{bmatrix} -7I-A & B \end{bmatrix} = \begin{bmatrix} 0 & 0 & 0 & 2 \\ 0 & 2 & 0 & 0 \\ 0 & 0 & 6 & 9 \end{bmatrix}\),秩为3。
对于 \(\lambda_2=-5\): \(\begin{bmatrix} -5I-A & B \end{bmatrix} = \begin{bmatrix} 2 & 0 & 0 & 2 \\ 0 & 0 & 0 & 0 \\ 0 & 0 & 4 & 9 \end{bmatrix}\),秩为2 (\(
3. 能控标准型 (Controllable Canonical Form)
如果系统已经处于可控标准型(如第二章传递函数法得到的): \[ A_c = \begin{bmatrix} 0 & 1 & 0 & \dots & 0 \\ 0 & 0 & 1 & \dots & 0 \\ \vdots & \vdots & \vdots & \ddots & \vdots \\ 0 & 0 & 0 & \dots & 1 \\ -a_0 & -a_1 & -a_2 & \dots & -a_{n-1} \end{bmatrix}, \quad B_c = \begin{bmatrix} 0 \\ 0 \\ \vdots \\ 0 \\ 1 \end{bmatrix} \] 则该系统总是状态完全能控的。可以验证其能控性矩阵 \(Q_c\) 的秩为 \(n\)。 (参考P28,PPT给出的可控标准型是 \(A\) 的最后一行是特征多项式系数的负数,\(B\) 是最后一个元素为1的列向量,这种形式的系统总是能控的。)
4. 线性变换与能控性
线性状态变换 \(\bar{\textbf{x}} = P\textbf{x}\) (P为非奇异阵) 不改变系统的能控性。 即,若 \((\textit{A, B})\) 能控,则 \((\textit{P A P}^{-1}\textit{, P B})\) 也一定能控。 证明:\(\bar{Q}_c = \begin{bmatrix} \bar{B} & \bar{A}\bar{B} & \dots & \bar{A}^{n-1}\bar{B} \end{bmatrix} = P Q_c\)。因为 \(P\) 非奇异,所以 \(\text{rank}(\bar{Q}_c) = \text{rank}(Q_c)\)。
5. 按约旦标准型的能控性判据 (Jordan Form Criterion)
将系统矩阵 \(A\) 通过相似变换化为约旦标准型 \(J = P^{-1}AP\),则 \(\bar{B} = P^{-1}B\)。 系统 \((A,B)\) 能控的充要条件是:
- 对应于 \(J\) 中每个具有相同特征值 \(\lambda_k\) 的所有约旦块,\(\bar{B}\) 中与这些约旦块的最后一行相对应的诸行向量是线性无关的。
- 特别地,如果特征值 \(\lambda_k\) 在 \(J\) 中只出现于一个约旦块中(即几何重数等于1),则条件简化为 \(\bar{B}\) 中与该约旦块最后一行对应的行向量非零。
- 如果 \(A\) 可以对角化为 \(\Lambda = \text{diag}(\lambda_1, \dots, \lambda_n)\) (所有特征值互异或虽有相同特征值但对应足够的线性无关特征向量),则 \(\bar{B} = P^{-1}B\)。系统能控的充要条件是 \(\bar{B}\) 的所有行均不为零向量。 (参考P59,对于对角型,每个输入行非零)
例题 (参考P62) 设系统化为约旦型后, \(\bar{A} = \text{diag}(J_1, J_2, J_3)\),其中 \(J_1 = \begin{bmatrix} -2 & 1 & 0 \\ 0 & -2 & 1 \\ 0 & 0 & -2 \end{bmatrix}\), \(J_2 = \begin{bmatrix} -3 & 1 \\ 0 & -3 \end{bmatrix}\), \(J_3 = \begin{bmatrix} -3 & 1 \\ 0 & -3 \end{bmatrix}\) \(\bar{B}^T = \begin{bmatrix} 1 & 0 & 1 & 0 & 4 & 0 & 7 \\ 0 & 0 & 0 & 1 & 0 & 0 & 1 \\ 0 & 0 & 0 & 0 & 1 & 0 & 0 \end{bmatrix}\) 对应 \(\lambda_1=-2\) 的约旦块 \(J_1\) 的最后一行是第3行,\(\bar{B}\) 的第3行为 \(\begin{bmatrix} 1 & 0 & 0 \end{bmatrix}\) (非零)。 对应 \(\lambda_2=-3\) 的第一个约旦块 \(J_2\) 的最后一行是第5行,\(\bar{B}\) 的第5行为 \(\begin{bmatrix} 4 & 0 & 1 \end{bmatrix}\)。 对应 \(\lambda_2=-3\) 的第二个约旦块 \(J_3\) 的最后一行是第7行,\(\bar{B}\) 的第7行为 \(\begin{bmatrix} 7 & 1 & 0 \end{bmatrix}\)。 由于特征值 \(\lambda_2=-3\) 对应两个约旦块,需要判断 \(\begin{bmatrix} 4 & 0 & 1 \\ 7 & 1 & 0 \end{bmatrix}\) 是否行线性无关。它是行线性无关的。 因此系统能控。
6. 能控性分解 (Controllability Decomposition)
如果系统 \((A,B)\) 不能完全能控,即 \(\text{rank}(Q_c) = n_1 < n\),则可以通过状态变换 \(P\) 将系统分解为能控子系统和不能控子系统: \[ \bar{A} = P^{-1}AP = \begin{bmatrix} A_c & A_{12} \\ 0 & A_{\bar{c}} \end{bmatrix}, \quad \bar{B} = P^{-1}B = \begin{bmatrix} B_c \\ 0 \end{bmatrix} \] \[ \bar{C} = CP = \begin{bmatrix} C_c & C_{\bar{c}} \end{bmatrix} \] 其中 \((A_c, B_c)\) 是 \(n_1\) 维的能控子系统, \(A_{\bar{c}}\) 是 \((n-n_1)\) 维的不能控子系统的系统矩阵。 系统的传递函数仅由能控部分决定:\(G(s) = C(sI-A)^{-1}B+D = C_c(sI-A_c)^{-1}B_c+D\)。 不能控部分的动态 \(\dot{\textbf{x}}_{\bar{c}} = A_{\bar{c}}\textbf{x}_{\bar{c}}\) 不受输入 \(\textbf{u}\) 的影响。
7. 能镇定性 (Stabilizability)
如果一个系统不是完全能控的,但其所有不能控的状态模式 (即 \(A_{\bar{c}}\) 的特征值) 都是稳定的 (即具有负实部),则称该系统是能镇定的 (stabilizable)。 能镇定意味着虽然不能将所有状态任意配置,但可以通过反馈将能控部分的状态任意配置,而不稳定的模式本身就是稳定的,因此整个系统可以通过反馈达到稳定。
三、状态能观测性 (State Observability)
1. 定义
对于线性定常系统 \(\dot{\textbf{x}} = A\textbf{x} + B\textbf{u}\), \(\textbf{y} = C\textbf{x} + D\textbf{u}\),如果在有限时间 \([t_0, t_f]\) 内,能够根据已知的输入 \(\textbf{u}(t)\) 和量测的输出 \(\textbf{y}(t)\) **唯一地确定**系统的初始状态 \(\textbf{x}(t_0)\),则称系统是状态完全能观测的 (completely state observable),简称能观测的。
由于 \(\int_{t_0}^{t_f} e^{A(t_f-\tau)}B\textbf{u}(\tau)d\tau\) 和 \(D\textbf{u}(t)\) 部分是已知的,可以从输出中减去,因此能观测性问题等价于研究对于零输入系统 \(\dot{\textbf{x}} = A\textbf{x}\), \(\textbf{y} = C\textbf{x}\),能否由 \(\textbf{y}(t)\) 确定 \(\textbf{x}(t_0)\)。 即 \(\textbf{y}(t) = Ce^{A(t-t_0)}\textbf{x}(t_0)\),能否反解出 \(\textbf{x}(t_0)\)。
2. 能观测性判据
a. 格拉姆矩阵判据
系统 \((\textit{A, C})\) 状态完全能观测的充分必要条件是:对于任意 \(t_f > t_0\) (通常取 \(t_0=0, t_f=T>0\)),能观测性格拉姆矩阵 (Observability Gramian Matrix) \[ W_o(t_0, t_f) = \int_{t_0}^{t_f} e^{A^T(\tau-t_0)}C^T C e^{A(\tau-t_0)} d\tau \] 或 (更常用的形式,令 \(\sigma = \tau-t_0\)) \[ W_o(T) = \int_{0}^{T} e^{A^T\tau}C^T C e^{A\tau} d\tau \] 是非奇异的 (即满秩的,\(\text{rank}(W_o(T)) = n\))。
- 若 \(W_o(T)\) 非奇异,则初始状态可以由下式确定: \(\textbf{x}_0 = W_o^{-1}(0,T) \int_0^T e^{A^T\tau}C^T \bar{\textbf{y}}(\tau)d\tau\),其中 \(\bar{\textbf{y}}(\tau) = \textbf{y}(\tau) - C\int_0^\tau e^{A(\tau-\sigma)}B\textbf{u}(\sigma)d\sigma - D\textbf{u}(\tau)\)。
b. 卡尔曼能观测性判据
系统 \((\textit{A, C})\) 状态完全能观测的充分必要条件是:\(mn \times n\) 维能观测性矩阵 (Observability Matrix) \[ Q_o = \begin{bmatrix} C \\ CA \\ CA^2 \\ \vdots \\ CA^{n-1} \end{bmatrix} \] 的秩为 \(n\) (即 \(\text{rank}(Q_o) = n\))。
例题 (参考P50,电路图1) \(A = \begin{bmatrix} -1 & 0 \\ 0 & -1 \end{bmatrix}, C = \begin{bmatrix} 0 & 1 \end{bmatrix}\) \(CA = \begin{bmatrix} 0 & 1 \end{bmatrix} \begin{bmatrix} -1 & 0 \\ 0 & -1 \end{bmatrix} = \begin{bmatrix} 0 & -1 \end{bmatrix}\) \(Q_o = \begin{bmatrix} C \\ CA \end{bmatrix} = \begin{bmatrix} 0 & 1 \\ 0 & -1 \end{bmatrix}\) \(\text{rank}(Q_o)=1 < n=2\)。系统不能观测。
例题 (参考P51,电路图2) \(A = \begin{bmatrix} 0 & -1 \\ 1 & -2 \end{bmatrix}, C = \begin{bmatrix} 0 & 1 \end{bmatrix}\) \(CA = \begin{bmatrix} 0 & 1 \end{bmatrix} \begin{bmatrix} 0 & -1 \\ 1 & -2 \end{bmatrix} = \begin{bmatrix} 1 & -2 \end{bmatrix}\) \(Q_o = \begin{bmatrix} C \\ CA \end{bmatrix} = \begin{bmatrix} 0 & 1 \\ 1 & -2 \end{bmatrix}\) \(\text{rank}(Q_o)=2 = n\)。系统能观测。
c. PBH 秩判据
系统 \((\textit{A, C})\) 状态完全能观测的充分必要条件是:对于 \(A\) 的所有特征值 \(\lambda_i\) (包括重根),矩阵
\[ \begin{bmatrix} \lambda_i I - A \\ C \end{bmatrix} \]
的秩均为 \(n\) (即列满秩)。
或者等价地说,不存在 \(A\) 的右特征向量 \(\textbf{v} \neq \textbf{0}\) (即 \(A\textbf{v} = \lambda \textbf{v}\)) 使得 \(C\textbf{v} = \textbf{0}\)。
3. 能观测标准型 (Observable Canonical Form)
如果系统已经处于可观标准型: \[ A_o = \begin{bmatrix} 0 & 0 & \dots & 0 & -a_0 \\ 1 & 0 & \dots & 0 & -a_1 \\ 0 & 1 & \dots & 0 & -a_2 \\ \vdots & \vdots & \ddots & \vdots & \vdots \\ 0 & 0 & \dots & 1 & -a_{n-1} \end{bmatrix}, \quad C_o = \begin{bmatrix} 0 & 0 & \dots & 0 & 1 \end{bmatrix} \] (注意 \(A_o\) 的形式,有时也定义为 \(A_c^T\)) 则该系统总是状态完全能观测的。
4. 线性变换与能观测性
线性状态变换 \(\bar{\textbf{x}} = P\textbf{x}\) 不改变系统的能观测性。 即,若 \((\textit{A, C})\) 能观测,则 \((\textit{P A P}^{-1}\textit{, C P}^{-1})\) 也一定能观测。 证明:\(\bar{Q}_o = Q_o P^{-1}\)。因为 \(P^{-1}\) 非奇异,所以 \(\text{rank}(\bar{Q}_o) = \text{rank}(Q_o)\)。
5. 按约旦标准型的能观测性判据
将系统矩阵 \(A\) 通过相似变换化为约旦标准型 \(J = P^{-1}AP\),则 \(\bar{C} = CP\)。 系统 \((A,C)\) 能观测的充要条件是:
- 对应于 \(J\) 中每个具有相同特征值 \(\lambda_k\) 的所有约旦块,\(\bar{C}\) 中与这些约旦块的第一列相对应的诸列向量是线性无关的。
- 特别地,如果特征值 \(\lambda_k\) 在 \(J\) 中只出现于一个约旦块中,则条件简化为 \(\bar{C}\) 中与该约旦块第一列对应的列向量非零。
- 如果 \(A\) 可以对角化为 \(\Lambda = \text{diag}(\lambda_1, \dots, \lambda_n)\),则 \(\bar{C} = CP\)。系统能观测的充要条件是 \(\bar{C}\) 的所有列均不为零向量。(参考P59,对于对角型,每个输出列非零)
6. 能观测性分解 (Observability Decomposition)
如果系统 \((A,C)\) 不能完全能观测,即 \(\text{rank}(Q_o) = n_1 < n\),则可以通过状态变换 \(P\) 将系统分解为能观测子系统和不能观测子系统: \[ \bar{A} = P^{-1}AP = \begin{bmatrix} A_o & 0 \\ A_{21} & A_{\bar{o}} \end{bmatrix}, \quad \bar{B} = P^{-1}B = \begin{bmatrix} B_o \\ B_{\bar{o}} \end{bmatrix} \] \[ \bar{C} = CP = \begin{bmatrix} C_o & 0 \end{bmatrix} \] 其中 \((A_o, C_o)\) 是 \(n_1\) 维的能观测子系统。 系统的传递函数仅由能观测部分决定:\(G(s) = C(sI-A)^{-1}B+D = C_o(sI-A_o)^{-1}B_o+D\)。 不能观测部分的动态 \(\dot{\textbf{x}}_{\bar{o}} = A_{21}\textbf{x}_o + A_{\bar{o}}\textbf{x}_{\bar{o}} + B_{\bar{o}}\textbf{u}\) 的状态 \(\textbf{x}_{\bar{o}}\) 不会影响输出 \(\textbf{y}\)。
7. 能检测性 (Detectability)
如果一个系统不是完全能观测的,但其所有不能观测的状态模式 (即 \(A_{\bar{o}}\) 的特征值) 都是稳定的 (即具有负实部),则称该系统是能检测的 (detectable)。 能检测性意味着虽然不能观测所有状态,但未被观测到的状态自身是稳定的,不会发散。
四、对偶原理 (Duality Principle)
能控性与能观测性之间存在对偶关系。
系统 \((A,B)\) **能控**的充分必要条件是系统 \((A^T, B^T)\) **能观测**。
系统 \((A,C)\) **能观测**的充分必要条件是系统 \((A^T, C^T)\) **能控**。
这意味着能控性的判据和理论可以平行地应用于能观测性,只需将 \((A,B)\) 替换为 \((A^T, C^T)\),反之亦然。例如:
\(Q_o(A^T, B^T) = (Q_c(A,B))^T\)。
\(\text{rank}(Q_c(A,B)) = \text{rank}(Q_o(A^T, B^T))\)。
五、卡尔曼结构分解 (Kalman Decomposition Theorem)
任何线性定常系统可以通过非奇异线性变换分解为四个子系统,分别对应于:
- 能控且能观测 (co - controllable and observable) 的部分 (\(\Sigma_{co}\))
- 能控但不能观测 (c\(\bar{o}\) - controllable but unobservable) 的部分 (\(\Sigma_{c\bar{o}}\))
- 不能控但能观测 (\(\bar{c}\)o - uncontrollable but observable) 的部分 (\(\Sigma_{\bar{c}o}\))
- 既不能控也不能观测 (\(\bar{c}\bar{o}\) - uncontrollable and unobservable) 的部分 (\(\Sigma_{\bar{c}\bar{o}}\))
变换后的系统矩阵和输入/输出矩阵具有如下结构 (参考P76): \[ \bar{A} = \begin{bmatrix} A_{co} & 0 & A_{13} & 0 \\ A_{21} & A_{c\bar{o}} & A_{23} & A_{24} \\ 0 & 0 & A_{\bar{c}o} & 0 \\ 0 & 0 & A_{43} & A_{\bar{c}\bar{o}} \end{bmatrix}, \quad \bar{B} = \begin{bmatrix} B_{co} \\ B_{c\bar{o}} \\ 0 \\ 0 \end{bmatrix}, \quad \bar{C} = \begin{bmatrix} C_{co} & 0 & C_{\bar{c}o} & 0 \end{bmatrix} \]
系统的传递函数仅由其能控且能观测的部分决定: \(G(s) = C(sI-A)^{-1}B+D = C_{co}(sI-A_{co})^{-1}B_{co}+D\)
六、最小实现 (Minimal Realization)
1. 定义
对于一个给定的传递函数矩阵 \(G(s)\),如果存在一个状态空间表达式 \((A,B,C,D)\) 使得 \(C(sI-A)^{-1}B+D = G(s)\),则称 \((A,B,C,D)\) 是 \(G(s)\) 的一个实现 (realization)。 在所有可能的实现中,状态空间维数 \(n\) 最小的实现称为最小实现 (minimal realization)。
2. 最小实现的条件
状态空间表达式 \((A,B,C,D)\) 是传递函数矩阵 \(G(s)\) 的最小实现的充分必要条件是:系统 \((A,B)\) **既状态完全能控又状态完全能观测**。
3. 传递函数与零极点对消
- 如果一个系统 \((A,B,C,D)\) 是最小实现,则其传递函数 \(G(s)=C(sI-A)^{-1}B+D\) 中,分子多项式和分母多项式之间没有共同的因子,即不存在零极点对消。
- 反之,如果传递函数中存在零极点对消,则其对应的状态空间实现一定不是最小的(即系统要么不能控,要么不能观,或者两者都不是)。
- 对消掉的极点对应于系统中不能控或不能观测的模态。
例题 (参考P89) \(G(s) = \frac{s+2}{(s+2)(s-1)}\) 如果实现为:\(\dot{x} = \begin{bmatrix} 1 & 0 \\ 0 & -2 \end{bmatrix}x + \begin{bmatrix} 1 \\ 1 \end{bmatrix}u, y = \begin{bmatrix} 1 & 0 \end{bmatrix}x\) 这个系统是能控的 (\(Q_c = \begin{bmatrix} 1 & 1 \\ 1 & -2 \end{bmatrix}\) 满秩),但是不能观测的 (\(Q_o = \begin{bmatrix} 1 & 0 \\ 1 & 0 \end{bmatrix}\) 秩为1)。极点 \(s=-2\) 被零点 \(s=-2\) 对消了,这个模态 \(e^{-2t}\) 是不能被观测到的。 其最小实现是 \(G(s) = \frac{1}{s-1}\),对应一阶系统 \(\dot{x}=x+u, y=x\)。
4. 获得最小实现的方法
- 对任意一个非最小实现 \((A,B,C,D)\) 进行卡尔曼分解。
- 其能控且能观测部分 \((A_{co}, B_{co}, C_{co}, D)\) 就是 \(G(s)\) 的一个最小实现。
七、总结
- 能控性判断系统输入能否完全控制系统状态。
- 能观测性判断系统状态能否通过输出完全反映出来。
- 掌握卡尔曼秩判据和PBH秩判据是核心。
- 对偶原理联系了能控性和能观测性。
- 卡尔曼分解揭示了系统的内部结构。
- 最小实现对应于既能控又能观测的系统,其传递函数无零极点对消。
这两个概念是后续设计状态反馈控制器和状态观测器的理论基础。