OPTIMAL AND ADAPTIVE CONTROL SYSTEMS
EPCE6301
Adama Science and Technology University
School of Electrical Engineering and Computing
Department of Electrical Power and Control Engineering
EPCE6301
CHAPTER 5:
MODEL REFERENCE ADAPTIVE CONTROL SYSTEM
2 EPCE6301
Types of Adaptive Control
• There are two major approaches to determine the controller
parameters adaptation.
1. Programmed or Gain Scheduled Adaptive Control
– it adjust the controller parameters based on the measurement of
an auxiliary variable and the knowledge of operating conditions
of process.
2. Self Adaptive Control
– the adaptation of controller parameters is based on the
measurement of closed loop performance and aim to optimize it.
– Model Reference Adaptive Control (MRAC) and Self Tuning
Regulator are two mechanisms that come under self adaptive
control
3 EPCE6301
Introduction to Model Reference Adaptive Control (MRAC)
• MRAC is an important adaptive controller in which the desired
performance is expressed in terms of a reference model, which
gives the desired response to a command signal.
• This model tells how the process output ideally should respond to
the command signal.
• The parameters are changed on the basis of feedback from the error,
which is the difference between the output of the system and the
output of the reference model.
• MRAC strategy is used to design the adaptive controller that works
on the principle of adjusting the controller parameters so that the
output of the actual plant tracks the output of a reference model
having the same reference input.
4 EPCE6301
Introduction to MRAC
• The system has inner loop composed of the process and the
controller in addition to another feedback loop that changes the
controller parameters called the outer loop.
• Controller parameters are adjusted in such a way that the error, the
difference between the plant output 𝑦 and model output 𝑦𝑚 is small.
– If the process out put 𝑦 behaves differently from what the model
prescribes, the controller is re-tuned. Conceptually, MRAC makes
any system behave as desired, but this is not possible in practice.
5 EPCE6301
Introduction to MRAC
• It is assumed that the structure of the controlled system is known,
and only its parameters are unknown.
– The reference model provides the ideal response of the system
that must be achieved through the parameters adaptation.
» The control law must have the ability to follow perfectly or
asymptotically the reference response (trajectory).
– This means that when the system parameters are exactly known,
the respective controller parameters must make the system
output identical with the output of the reference model.
• The adaptation law searches to find the parameter values which
assure that the system response under MRAC is ultimately the same
with the reference model response
– that is to assure convergence of the error between the two
responses to zero.
6 EPCE6301
Designing the Controller’s Adaptation Law
• The mechanism for adjusting the parameters in a MARC can be
obtained in two ways: by using a gradient method or by applying
stability theory.
• The two most popular methods of designing the controller's
adaptation law are as follows:
1. The Steepest Descent Method
2. The Lyapunov Stability Method
7 EPCE6301
Steepest Descent Parameter Adaptation Law
• The MIT rule is the original approach to model-reference adaptive
control. The name is derived from the fact that it was developed at
the Instrumentation Laboratory at MIT.
• To present the MIT rule, we will consider a closed-loop system in
which the controller has adjustable parameter 𝜃.
• The desired response is specified by a model whose output is y𝑚 .
Let 𝑒 be the error between the output 𝑦 of the closed-loop system
and the output 𝑦𝑚 of the model (named tracking error 𝑒 = 𝑦 − 𝑦𝑚 ).
• One possibility is to adjust parameters in such a way that the loss
1
function 𝐽(𝜃) = 𝑒 2 is minimized.
2
• To reduce 𝐽(𝜃), it is logical to vary the parameters in the direction of
the negative gradient of 𝑒 2 that is: (MIT Rule)
𝑑𝜃 𝑑𝐽 𝑑𝑒
= −𝛾 = −𝛾𝑒
𝑑𝑡 𝑑𝜃 𝑑𝜃
8 EPCE6301
Steepest Descent Parameter Adaptation Law
𝑑𝑒
• is called the sensitivity derivative. It indicates how the error is
𝑑𝜃
influenced by the adjustable parameters 𝜃.
• Procedure
𝑦
Process ∶ 𝐺 𝑠 =
𝑢
𝑦𝑚
Model ∶ 𝐺𝑚 𝑠 =
𝑢𝑐
Control law ∶ 𝑢 𝑡 = 𝑓 𝑢𝑐 , 𝑦
Error ∶ 𝑒 = 𝑦 − 𝑦𝑚
𝑑𝑒 𝑑𝑦
=
𝑑𝜃 𝑑𝜃
𝑑𝜃 𝑑𝑒
MIT Rule ∶ = −𝛾𝑒
𝑑𝑡 𝑑𝜃
9 EPCE6301
Examples - First Order System Adjustment
• Assume
𝑑𝑦 𝑌 𝑏
= −𝑎𝑦 + 𝑏𝑢 𝐺 𝑠 = =
𝑑𝑡 𝑈 𝑠+𝑎
𝑑𝑦𝑚 𝑌𝑚 𝑏𝑚
= −𝑎𝑚 𝑦 + 𝑏𝑚 𝑢𝑐 𝐺𝑚 𝑠 = =
𝑑𝑡 𝑈𝑐 𝑠 + 𝑎𝑚
• The control input structure is given by the following equation
𝑢 𝑡 = 𝜃1 𝑡 𝑢𝑐 𝑡 − 𝜃2 𝑡 𝑦 𝑡
• where 𝜃1 𝑡 and 𝜃2 𝑡 are the adaptive control parameters that are
determined by an adaptive control algorithm.
• The goal of the adaptive control algorithm is to determine the
parameters 𝜃1 𝑡 and 𝜃2 𝑡 such that the system output 𝑦(𝑡) is as
close as possible to the output 𝑦𝑚 𝑡 of the reference model for
some selected reference input function 𝑢𝑐 𝑡 .
10 EPCE6301
Examples - First Order System Adjustment
𝑌 𝑏 𝑌𝑚 𝑏𝑚
𝐺 𝑠 = = 𝐺𝑚 𝑠 = =
𝑈 𝑠+𝑎 𝑈𝑐 𝑠 + 𝑎𝑚
𝑢 𝑡 = 𝜃1 𝑡 𝑢𝑐 𝑡 − 𝜃2 𝑡 𝑦 𝑡
𝑠+𝑎
𝑈 = 𝜃1 𝑈𝑐 − 𝜃2 𝑌 = 𝑌
𝑏
𝑌 𝜃1 𝑏𝜃1
⇒ = =
𝑈𝑐 𝑠 + 𝑎 + 𝜃 𝑠 + 𝑎 + 𝑏𝜃2
𝑏 2
𝑏𝑚 = 𝑏𝜃1 ; 𝑎𝑚 = 𝑎 + 𝑏𝜃2
𝑏𝜃1
𝑒 = 𝑌 − 𝑌𝑚 = 𝑈𝑐 − 𝑌𝑚
𝑠 + 𝑎 + 𝑏𝜃2
𝜕𝑒 𝑏 𝑏
= 𝑈𝑐 = 𝑈𝑐
𝜕𝜃1 𝑠 + 𝑎 + 𝑏𝜃2 𝑠 + 𝑎𝑚
𝜕𝑒 −𝑏 2 𝜃1 −𝑏 𝑏𝜃1 −𝑏
= 𝑈 = 𝑈 = 𝑌
𝜕𝜃2 𝑠 + 𝑎 + 𝑏𝜃2 2 𝑐 𝑠 + 𝑎 + 𝑏𝜃2 𝑠 + 𝑎 + 𝑏𝜃2 𝑐 𝑠 + 𝑎𝑚
11 EPCE6301
Examples - First Order System Adjustment
𝑑𝜃 𝜕𝑒
MIT Rule ∶ = −𝛾 𝑒
𝑑𝑡 𝜕𝜃
𝑑𝜃1 𝜕𝑒 𝑏
= −𝛾1 𝑒 ⇒ 𝑠𝜃1 = −𝛾1 𝑈 𝑒
𝑑𝑡 𝜕𝜃1 𝑠 + 𝑎𝑚 𝑐
𝑑𝜃2 𝜕𝑒 𝑏
= −𝛾2 𝑒 ⇒ 𝑠𝜃2 = 𝛾2 𝑦 𝑒
𝑑𝑡 𝜕𝜃2 𝑠 + 𝑎𝑚
Let : 𝑎 = 1, 𝑏 = 2, 𝑎𝑚 = 8, 𝑏𝑚 = 8
12 EPCE6301
Examples - First Order System Adjustment
• Result at 𝛾1 = 2; 𝛾2 = 0.1 𝛾1 = 1; 𝛾2 = 0.01
13 EPCE6301
Examples - First Order System Adjustment
• Result at 𝛾1 = 0.1; 𝛾2 = 0.01 𝛾1 = 0.2; 𝛾2 = 0.01
14 EPCE6301
Examples - Second Order System Adjustment
• Assume
𝑦 𝑠+𝑏
𝐺 𝑠 = = 2
𝑢 𝑠 + 𝑎1 𝑠 + 𝑎0
𝑦𝑚 𝑠 + 𝑏𝑚
𝐺𝑚 𝑠 = =
𝑢𝑐 𝑠 2 + 𝑎1𝑚 𝑠 + 𝑎0𝑚
• The control input structure is given by the following equation
𝑢 𝑡 = 𝜃1 𝑡 𝑢𝑐 𝑡 − 𝜃2 𝑡 𝑦 𝑡
𝑠 2 + 𝑎1 𝑠 + 𝑎0
𝑦 = 𝜃1 𝑢𝑐 − 𝜃2 𝑦
𝑠+𝑏
𝑦 𝜃1 𝑠 + 𝑏𝜃1
⇒ =
𝑢𝑐 𝑠 2 + 𝑎1 + 𝜃2 𝑠 + 𝑎0 + 𝑏𝜃2
𝑒 = 𝑦 − 𝑦𝑚
𝜕𝑒 𝑠+𝑏
= 2 𝑢𝑐 ≈ 𝐺𝑚 𝑠 𝑢𝑐
𝜕𝜃1 𝑠 + 𝑎1 + 𝜃2 𝑠 + 𝑎0 + 𝑏𝜃2
15 EPCE6301
Examples - Second Order System Adjustment
𝜕𝑒
≈ 𝐺𝑚 𝑠 𝑢𝑐
𝜕𝜃1
𝜕𝑒 −𝜃1 𝑠 + 𝑏 𝑠 + 𝑏
= 2 𝑢
𝜕𝜃2 𝑠 + 𝑎1 + 𝜃2 𝑠 + 𝑎0 + 𝑏𝜃2 2 𝑐
− 𝑠+𝑏 𝑠 + 𝑏 𝜃1 𝑦2
= 2 2
𝑢𝑐 = −
𝑠 + 𝑎1 + 𝜃2 𝑠 + 𝑎0 + 𝑏𝜃2 𝑠 + 𝑎1 + 𝜃2 𝑠 + 𝑎0 + 𝑏𝜃2 𝜃2 𝑢𝑐
𝑑𝜃 𝜕𝑒
MIT Rule ∶ = −𝛾 𝑒
𝑑𝑡 𝜕𝜃
𝑑𝜃1 𝜕𝑒
= −𝛾1 𝑒 ⇒ 𝑠𝜃1 = −𝛾1 𝐺𝑚 𝑢𝑐 𝑒
𝑑𝑡 𝜕𝜃1
𝑑𝜃2 𝜕𝑒 𝑦2
= −𝛾2 𝑒 ⇒ 𝑠𝜃2 = 𝛾2 𝑒 ≈ 𝛾2 𝐺𝑚 𝑠 𝑦𝑒
𝑑𝑡 𝜕𝜃2 𝜃2 𝑢𝑐
16 EPCE6301
Examples - Second Order System Adjustment
𝑠+5 𝑠 + 10
𝐿𝑒𝑡: 𝐺 𝑠 = 2 𝑎𝑛𝑑 𝐺𝑚 𝑠 = 2
𝑠 + 4𝑠 + 3 𝑠 + 8𝑠 + 10
17 EPCE6301
Examples - Second Order System Adjustment
• Result at 𝛾1 = 10; 𝛾2 = 10 𝛾1 = 10; 𝛾2 = 5
18 EPCE6301
Examples - Second Order System Adjustment
• Result at 𝛾1 = 1.2; 𝛾2 = 0.5 𝛾1 = 20; 𝛾2 = 10
19 EPCE6301
Lyapunov-Based Adaptation Law
• This adaptation law assumes right from the beginning that the error
𝑒(𝑡) will really converge to zero. For clarity, we will illustrate this
method for the simple scalar system:
𝑦ሶ 𝑡 = −𝑎𝑦 𝑡 + 𝑏𝑢 𝑡
where 𝑢 𝑡 is the control variable and 𝑦 𝑡 is the measured output.
• The stable reference model is assumed to be:
𝑦𝑚
ሶ (𝑡) = −𝑎𝑚 𝑦𝑚 (𝑡) + 𝑏𝑚 𝑢𝑐 𝑡 , 𝑎𝑚 > 0
• The control law is selected as:
𝑢(𝑡) = −𝑘1 𝑦(𝑡) + 𝑘0 𝑢𝑐 (𝑡)
• Defining the output error 𝑒 = 𝑦 − 𝑦𝑚 , we get the dynamic equation
for the closed-loop error, obtained using the control law, as:
𝑒ሶ 𝑡 = 𝑎𝑚 𝑒 + (𝑎𝑚 − 𝑎 − 𝑏𝑘1 )𝑦 + 𝑏𝑘0 − 𝑏𝑚 𝑢𝑐
20 EPCE6301
Lyapunov-Based Adaptation Law
𝑒ሶ 𝑡 = 𝑎𝑚 𝑒 + (𝑎𝑚 − 𝑎 − 𝑏𝑘1 )𝑦 + 𝑏𝑘0 − 𝑏𝑚 𝑢𝑐
𝑎𝑚−𝑎
• Clearly, if 𝑎 + 𝑏𝑘1 = 𝑎𝑚 , that is, 𝑘1 = ,
𝑏
𝑏𝑚
and 𝑏𝑘0 − 𝑏𝑚 = 0 𝑖. 𝑒. , 𝑘0 = ,
𝑏
• the closed loop system is identical with the reference model and so
𝑒 𝑡 → 0 for 𝑡 → ∞ .
• To construct an adaptation law that leads the controller parameters
𝑏 𝑎𝑚−𝑎
𝑘0 and 𝑘1 to the above ideal values 𝑘0 = 𝑚 and 𝑘1 = , we use
𝑏 𝑏
the following candidate Lyapunov function:
1 2 1 2
1
𝑉(𝑒, 𝑘0 , 𝑘1 ) = 𝑒 + 𝑏𝑘1 + 𝑎 − 𝑎𝑚 + 𝑏𝑘0 − 𝑏𝑚 2
2 𝑏𝛾 𝑏𝛾
21 EPCE6301
Lyapunov-Based Adaptation Law
1 2 1 2
1 2
𝑉(𝑒, 𝑘0 , 𝑘1 ) = 𝑒 + 𝑏𝑘1 + 𝑎 − 𝑎𝑚 + 𝑏𝑘0 − 𝑏𝑚
2 𝑏𝛾 𝑏𝛾
• This function satisfies the first three properties of Lyapunov functions
and has zero value when 𝑘0 and 𝑘1 have the ideal values.
Differentiating 𝑉, we get:
1 1
ሶ𝑉 = 𝑒𝑒ሶ + 𝑏𝑘1 + 𝑎 − 𝑎𝑚 𝑘ሶ 1 + 𝑏𝑘0 + 𝑏𝑚 𝑘ሶ 0
𝛾 𝛾
1 1
= 𝑒𝑒ሶ + 𝑏𝑘1 + 𝑎 − 𝑎𝑚 𝑘1 − 𝛾𝑦𝑒 + 𝑏𝑘0 + 𝑏𝑚 𝑘ሶ 0 + 𝛾𝑢𝑐 𝑒
ሶ
𝛾 𝛾
Thus, we choose the parameter adaptation (updating) laws as:
𝑘ሶ 0 = −𝛾𝑢𝑐 𝑒
𝑘ሶ 1 = 𝛾𝑦𝑒
Then,
𝑉ሶ = −𝑎𝑚 𝑒 2 < 0 𝑎𝑚 > 0
22 EPCE6301
Lyapunov-Based Adaptation Law
Then,
𝑉ሶ = −𝑎𝑚 𝑒 2 < 0 𝑎𝑚 > 0
• Therefore, by the Lyapunov stability criterion, 𝑒(𝑡) tends
asymptotically to zero as 𝑡 → ∞. Of course, the convergence of 𝑘0
and 𝑘1 to the ideal values is not assured unless some other
appropriate conditions are posed. We observe that the adaptation
laws are of the general form:
𝜃ሶ = 𝛾𝜓𝑒
• where 𝜃 is the parameter vector, 𝑒 is the error between the closed
loop output and the reference output, and 𝜓 is a known function
depending on 𝑢𝑐 and 𝑦.
• The above Lyapunov adaptation method is applicable to MIMO and
nonlinear systems.
23 EPCE6301