反步法Backstepping的数学本质

本文建议有一定反步法知识的同学阅读，欢迎参考笔者另一篇详细介绍反步法Backstepping的文章Backstepping反步法控制四旋翼无人机（2）。

1. 反步法Backstepping简介

假设有一个二阶非线性时变系统
$$\{\begin{array}{l}\begin{array}{rl}{\dot{x}}_1& =x_2\\ {\dot{x}}_2& =f(t,x_1,x_2)+g(t)u\\ y& =x_1\end{array}\end{array}\text{\hspace{1em}(1)}$$其中$f(t,x_1,x_2),g(t)$均为非线性时变函数，状态$x_1$为输出状态。

设期望的状态为$x_{1d}$，则可以写出期望状态与实际状态的误差
$$e_1=x_{1d}-x_1$$根据李雅普诺夫稳定性理论，可以设计李雅普诺夫函数$V_1=\frac{1}{2}{e}_1^2$，则其导数为
$${\dot{V}}_1=e_1{\dot{e}}_1=e_1\left({\dot{x}}_{1d}-{\dot{x}}_1\right)=e_1\left({\dot{x}}_{1d}-x_2\right)$$若存在理想状态量$x_{2v}$，满足
$$x_{2v}={\dot{x}}_{1d}+{\alpha }_1{e}_1$$则李雅普诺夫函数导数即有
$$\begin{array}{rl}{\dot{V}}_1& =e_1\left({\dot{x}}_{1d}-x_{2v}\right)\\ & =e_1\left({\dot{x}}_{1d}-{\dot{x}}_{1d}-{\alpha }_1{e}_1\right)\\ & =-{\alpha }_1{e}_1^2\le 0\end{array}$$然而，实际情况下$x_2\ne x_{2v}$。由此需要引入新的误差$e_2=x_2-x_{2v}$，并修改李雅普诺夫函数：
$$V_2=\frac{1}{2}{e}_1^2+\frac{1}{2}{e}_2^2$$容易观察出：
$$\begin{gathered} e_2=x_2-x_{2v}=x_2-{\dot{x}}_{1d}-{\alpha }_1{e}_1 \\ ⟹{\dot{x}}_{1d}-x_2=-e_2-{\alpha }_1{e}_1 \end{gathered}$$接下来对新的李雅普诺夫函数求导：
$$\begin{array}{rl}{\dot{V}}_2& =e_1{\dot{e}}_1+e_2{\dot{e}}_2\\ & =e_1\left({\dot{x}}_{1d}-x_2\right)+e_2\left({\dot{x}}_2-{\ddot{x}}_{1d}-{\alpha }_1{\dot{e}}_1\right)\\ & =-e_1\left(e_2+{\alpha }_1{e}_1\right)+e_2\left[f(t,x_1,x_2)+g(t)u-{\ddot{x}}_{1d}-{\alpha }_1\left({\dot{x}}_{1d}-x_2\right)\right]\\ & =-e_1\left(e_2+{\alpha }_1{e}_1\right)+e_2\left[f(t,x_1,x_2)+g(t)u-{\ddot{x}}_{1d}+{\alpha }_1\left(e_2+{\alpha }_1{e}_1\right)\right]\end{array}$$此时只要设计$u$为
$$u=\frac{1}{g(t)}\left[-f(t,x_1,x_2)+{\ddot{x}}_{1d}-{\alpha }_1\left(e_2+{\alpha }_1{e}_1\right)+e_1-{\alpha }_2{e}_2\right]\text{\hspace{1em}(2)}$$即可，代回上式可得到此时李雅普诺夫函数的导数为
$${\dot{V}}_2=-{\alpha }_1{e}_1^2-{\alpha }_2{e}_2^2\le 0$$系统即稳定。

2. 反步法的数学本质

把得到的控制量式(2)带回到原系统(1)中：
$$\{\begin{array}{l}\begin{array}{rl}{\dot{x}}_1& =x_2\\ {\dot{x}}_2& =f(t,x_1,x_2)+g(t)u\\ y& =x_1\end{array}\end{array}$$即
$$\begin{array}{rl}{\dot{x}}_2& ={\ddot{x}}_1=f(t,x_1,x_2)+g(t)u\\ & =f(t,x_1,x_2)+g(t)\cdot \frac{1}{g(t)}\left[-f(t,x_1,x_2)+{\ddot{x}}_{1d}-{\alpha }_1\left(e_2+{\alpha }_1{e}_1\right)+e_1-{\alpha }_2{e}_2\right]\\ & ={\ddot{x}}_{1d}-{\alpha }_1\left(e_2+{\alpha }_1{e}_1\right)+e_1-{\alpha }_2{e}_2\end{array}$$代入$e_1,e_2$表达式：
$$\begin{array}{rl}{\dot{x}}_2={\ddot{x}}_1& ={\ddot{x}}_{1d}-{\alpha }_1\left[\left(x_2-{\dot{x}}_{1d}-{\alpha }_1{e}_1\right)+{\alpha }_1{e}_1\right]+e_1-{\alpha }_2\left(x_2-{\dot{x}}_{1d}-{\alpha }_1{e}_1\right)\\ & ={\ddot{x}}_{1d}-{\alpha }_1\left(x_2-{\dot{x}}_{1d}\right)+e_1-{\alpha }_2\left(x_2-{\dot{x}}_{1d}-{\alpha }_1{e}_1\right)\\ & ={\ddot{x}}_{1d}-{\alpha }_1\left({\dot{x}}_1-{\dot{x}}_{1d}\right)+e_1-{\alpha }_2\left({\dot{x}}_1-{\dot{x}}_{1d}-{\alpha }_1{e}_1\right)\\ & ={\ddot{x}}_{1d}+{\alpha }_1{\dot{e}}_1+e_1-{\alpha }_2\left(-{\dot{e}}_1-{\alpha }_1{e}_1\right)\\ & ={\ddot{x}}_{1d}+{\alpha }_1{\dot{e}}_1+e_1+{\alpha }_2{\dot{e}}_1+{\alpha }_1{\alpha }_2{e}_1\end{array}\text{\hspace{1em}(3)}$$将上式变形
$${\ddot{x}}_1-{\ddot{x}}_{1d}={\alpha }_1{\dot{e}}_1+e_1+{\alpha }_2{\dot{e}}_1+{\alpha }_1{\alpha }_2{e}_1$$即
$$-{\ddot{e}}_1=\left({\alpha }_1+{\alpha }_2\right){\dot{e}}_1+\left({\alpha }_1{\alpha }_2+1\right)e_1\text{\hspace{1em}(4)}$$式(4)即为关于$x_1$的误差量$e_1$的一个微分方程。对其做拉普拉斯变换，转换到$s$空间：
$$-s^2{E}_1(s)=\left({\alpha }_1+{\alpha }_2\right)s{E}_1(s)+\left({\alpha }_1{\alpha }_2+1\right)E_1(s)$$若将其视为一个系统，则其传递函数为
$$W(s)=\frac{1}{s^2+\left({\alpha }_1+{\alpha }_2\right)s+\left({\alpha }_1{\alpha }_2+1\right)}\text{\hspace{1em}(5)}$$由于该传递函数是从反步法稳定后的系统得出的，若将稳定后的系统视作某个开环系统$G(s)$的闭环系统，则很容易就能反求出该开环系统：
$$G(s)=\frac{1}{s^2+\left({\alpha }_1+{\alpha }_2\right)s+{\alpha }_1{\alpha }_2}=\frac{1}{\left(s+{\alpha }_1\right)\left(s+{\alpha }_2\right)}\text{\hspace{1em}(6)}$$
由此可以对反步法的数学本质做出解释：

1. 反步法本质上是向系统里引入了两个极点$s_1={\alpha }_1,s_2={\alpha }_2$，使得系统变成两个一阶惯性环节的串联，从而增强稳定性。
2. 同时可以推断出的是，当系统不是二阶系统，而是三阶、n阶时，相应地式(6)必将拥有3个、n个极点。
3. 由于极点$s_1={\alpha }_1,s_2={\alpha }_2$即为反步法的两个调节参数，故也可以得出结论：调节参数${\alpha }_1,{\alpha }_2$可以改善系统表现，当${\alpha }_1,{\alpha }_2$靠近虚轴时系统振荡变强，响应时间变长；当${\alpha }_1,{\alpha }_2$增大时极点远离虚轴，系统响应时间变短，甚至无超调。