Documents

Back Stepping

Description
Briefing on Backstepping
Categories
Published
of 8
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
  Backstepping In complicated systems, like aircraft, there are often parameters which are not fully known, or which mighteven occasionally change. To cope with this, you need a controller that adapts to these parameters: anadaptive controller. In this summary we’ll be investigating how such controllers work. In particular, we’llbe investigating the backstepping controller. 1 Basic definitions and theories 1.1 Function definitions Before we’re discussing systems, we start with some formal function definitions. Let’s examine a function V    ( x ). We say that  V    ( x ) is ... ã  positive definite  if   V    ( 0 ) = 0 and  V    ( x )  >  0 with  x  =  0 . ã  positive semidefinite  if   V    ( 0 ) = 0 and  V    ( x ) ≥ 0 with  x  =  0 . ã  negative (semi)definite  if   − V    ( x ) is positive (semi)definite. ã  radially unbounded  if   V    ( x ) →∞  as  | x |→∞ . 1.2 System definitions Let’s examine a system with state  x  and dynamics  ˙x  =  f  ( x ). A function  x ( t ) with initial state  x (0) =  x 0 that satisfies the system dynamics is called a  solution  of the system. A system is called ... ã  stable  if, for given   >  0, there exists a  δ  (  )  >  0 such that all solutions with initial conditions | x (0) |  < δ   satisfy  | x ( t ) |  <   for all  t  ≥  0. More intuitively speaking, all solutions starting near x  =  0  remain bounded. ã  asymptotically stable  (AS) if it is stable and a  δ   can be found such that all solutions with | x (0) | < δ   satisfy  | x ( t ) |→ 0  as  t →∞ . More intuitively speaking, all solutions starting near  x  =  0 are bounded and converge to zero. ã  globally asymptotically stable  (GAS) if it is asymptotically stable for any initial state  x (0). 1.3 The Lyapunov theory Let’s say we have a time-invariant system with state  x  and dynamics  ˙x  =  f  ( x ). How do we prove thatthe system is stable? For that, we can use the Lyapunov theory. The first thing which we need is aLyapunov function  V    ( x ). This function has to be positive definite in a region Γ near  x  =  0 . (It oftenhelps to think of   V    as some kind of energy. It is never negative, and can only be zero in the zero state.)Second, we will examine ˙ V    . We can rewrite this as˙ V    ( x ) =  ∂V    ( x ) ∂t  =  ∂V    ( x ) ∂  x ∂  x ∂t  =  ∂V    ( x ) ∂  x  f  ( x ) .  (1.1)The Lyapunov theory now states that ... ã  if  ˙ V    ( x ) is negative semidefinite in the region Γ, then the solution is stable.1  ã  if  ˙ V    ( x ) is negative definite in the region Γ, then the solution is asymptotically stable. ã  if   V    ( x ) positive definite and radially unbounded for all  x , and if  ˙ V    ( x ) is negative definite for all  x ,then the solution is globally asymptotically stable.The Lyapunov theory is actually quite logical. If you have some function that is always decreasing, thenit must reach zero eventually. So there is no way that the system diverges: it has to be stable. 2 Applying adaptive control 2.1 The problem statement Let’s examine a time-invariant system with an unknown parameter. For example, consider˙ x  =  θx  +  u,  (2.1)where  θ  is unknown. We could use as a controller  u  = − kx . If   θ < k , this will ensure stability. However,if   θ > k , we have an instable system. A possible solution would be to use a nonlinear controller, like u  =  − k 1 x  −  k 2 x 3 . Although this ensures that the system state is bounded, it does not ensure anasymptotically stable system. Instead, it will be better to make our system adapt to  θ . 2.2 Estimation-based adaptive control Let’s suppose that we can measure the state  x  of the system, but not the state derivative ˙ x . Just deriving˙ x  from  x  is in practice unwise, due to measurement noise. Instead, we filter the state  x  and the input  u first. The filtered signals are denoted as x f   =  xs  + 1 and  u f   =  us  + 1 .  (2.2)Using ˙ x  =  sx , we can now rewrite the system to x  = ( θ  + 1) x f   +  u f  .  (2.3)We will also have an  estimated value  of   θ , denoted as ˆ θ . (Initially, we can assume that ˆ θ  = 0.) We canuse ˆ θ  to predict the state  x  of the system. We will findˆ x  = (ˆ θ  + 1) x f   +  u f  .  (2.4)If our estimate ˆ θ  is inaccurate, we will of course get an error in our  prediction  ˆ x . This error is e  =  x − ˆ x  = ( θ − ˆ θ ) x f  .  (2.5)To reduce this  prediction error  e , we should adjust our estimate ˆ θ . One way to do this is by using˙ˆ θ  = − γ  2 ∂  ( e 2 ) ∂  ˆ θ =  γe∂e∂  ˆ θ = − γex f  .  (2.6)By doing this, the prediction error  e  will decrease. If we then also use the controller  u  = − ( k  + ˆ θ ) x , thenour problem is solved. The error will decrease.2  2.3 Lyapunov-based adaptive control Another way in which to stabilize the system is by using the Lyapunov theory. To do this, we first needa Lyapunov function. We could try the function V    ( x,  ˆ θ ) = 12 x 2 + 12  θ − ˆ θ  2 .  (2.7)˙ V    can be found according to˙ V    =  x ˙ x −  θ − ˆ θ   ˙ˆ θ  =  x ( u  +  θx ) −  θ − ˆ θ   ˙ˆ θ  =  xu  + ˆ θ  ˙ˆ θ  +  θ  x 2 −  ˙ˆ θ  .  (2.8)We want to choose our input  u  and our update law ˙ˆ θ  such that ˙ V    is negative definite. However, neither u  nor ˙ˆ θ  can depend on the unknown  θ . To solve this problem, we can first set ˙ˆ θ  =  x 2 . This eliminatesthe term on the right. For the input, we then set  u  =  − ( k  + ˆ θ ) x . This results in ˙ V    =  − kx 2 . If   k >  0,this is evidently negative definite. The system has thus been stabilized. 2.4 A more complicated problem We have seen two techniques now: estimation-based and Lyapunov-based adaptive control. But do thesetechniques always work? We could consider the nonlinear system˙ x  =  θx 2 +  u.  (2.9)Both techniques will use the control law u  = − kx − ˆ θx 2 .  (2.10)But what about the parameter update law? For estimation-based adaptive control it is hard to choosea good parameter update law. If we choose the same law as before, we will get an unstable system.Lyapunov-based adaptive control works better here. Using the same Lyapunov function as before, wefind the update law˙ˆ θ  =  x 3 (2.11)which results in a stable system. This shows the advantage of Lyapunov-based adaptive control. So fromnow on, we will not consider estimation-based adaptive control anymore. Instead, we focus on Lyapunov-based adaptive control. But plenty of challenges await us still. How do we find a Lyapunov function fromwhich we can derive a control law? (Such a function is called a  control Laypunov function  (CLF).)And how do we choose the best control law  u  =  α ( x )? 2.5 Sontag’s formula Let’s suppose that we have a control problem of the form˙ x  =  f  ( x ) +  g ( x ) u,  (2.12)where  f  (0) = 0. Also, we’ve already found a suitable CLF. One way to choose a good control law is byusing  Sontag’s formula u  =  α s ( x ) =  − ∂V  ∂x  f  +   ( ∂V  ∂x  f  ) 2 + ( ∂V  ∂x ) 4 ( ∂V  ∂x  g ) for  ∂V  ∂x  g   = 0 , 0 for  ∂V  ∂x  g  = 0 . (2.13)The resulting control law often uses parts of the system to help stabilize it. This reduces the magnitudeof the control input  u . So, the control law is relatively efficient.3  3 The backstepping algorithm 3.1 A backstepping example The control problem becomes more complicated if the input  u  doesn’t directly influence  x . As an example,we can consider a system with an integrator, like˙ x  =  x − x 3 +  ξ,  (3.1)˙ ξ   =  u.  (3.2)Let’s only examine the first equation, and act as if   ξ   was our control input. A CLF would be  V    =  12 x 2 .It follows that  ξ   =  − ( k  + 1) x . This would stabilize the system. However, we can’t control  ξ   directly.Instead, we specify the desired value ξ  d  =  α ( x ) = − ( k  + 1) x.  (3.3)In this case, we call  ξ   the  virtual control . The function  α ( x ) which it should follow is the  stabilizingfunction . The deviation of   ξ   from  α ( x ) is called the  error state  z . It is defined as z  =  ξ  − ξ  d  =  ξ  − α ( x ) =  ξ   + ( k  + 1) x.  (3.4)We can use the error state to rewrite our system to˙ x  =  x − x 3 +  z  +  α ( x ) = − kx − x 3 +  z,  (3.5)˙ z  = ˙ ξ  −  ˙ α  =  u  + ( k  + 1)˙ x  =  u  + ( k  + 1)( − kx − x 3 +  z ) .  (3.6)We now also need to incorporate  z  in our Lyapunov function. A possible function is V    ( x,z ) = 12 x 2 + 12 z 2 = 12 x 2 + 12 ( ξ   + ( k  + 1) x ) 2 .  (3.7)Taking the derivative gives˙ V    ( x,z ) =  x ˙ x  +  z  ˙ z  =  x ( − kx − x 3 +  z ) +  z  u  + ( k  + 1)( − kx − x 3 +  z )   (3.8)=  − kx 2 − x 4 +  z  x  +  u  + ( k  + 1)( − kx − x 3 +  z )  .  (3.9)As a control law, we can now use u  = − cz − x − ( k  + 1)( − kx − x 3 +  z ) ,  (3.10)which turns the Lyapunov derivative into˙ V    ( x,z ) = − kx 2 − x 4 − cz 2 .  (3.11)This is evidently negative definite, so the system has been stabilized.Now what have we done? We wanted to stabilize  x . To do this, we provided a desired value for  ξ   whichwould accomplish this. We continued by defining an error signal  z , which required to be stabilized aswell. To do this, we set the value for  u . If the system would have more equations, we could continuestepping back through equations like this. This procedure is thus called  backstepping . 3.2 The backstepping algorithm The procedure from the previous paragraph can be generalized. Consider a system of the form˙ x 1  =  f  1 ( x 1 ) +  g 1 ( x 1 ) x 2 ,  (3.12)˙ x 2  =  f  2 ( x 1 ,x 2 ) +  g 1 ( x 1 ,x 2 ) x 3 ,  (3.13)... (3.14)˙ x n  =  f  n ( x 1 ,x 2 ,...,x n ) +  g n ( x 1 ,x 2 ,...,x n ) u.  (3.15)4

A660

Jul 23, 2017
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks