Preprints, 4th IFAC Conference on

Advances in Proportional-Integral-Derivative Control

Almería, Spain | June 12-14, 2024

On the hybrid MRAC-PID control: A

comparison study ⋆
Juan D. Gil ∗ Igor M. L. Pataro ∗ J. L. Guzmán ∗
M. Berenguel ∗
∗
Universidad de Almerı́a, Centro Mixto CIESOL, ceia3, Ctra.
Sacramento s/n, 04120, Almerı́a, Spain (e-mail: [Link]@[Link].
ilp428@[Link], [Link]@[Link], beren@[Link]).

Abstract:
This study performs a comparative analysis of three Model Reference Adaptive Control (MRAC)
schemes grounded in Lyapunov theory. Specifically, it compares a pure MRAC controller with
two hybrid MRAC-PID schemes, each integrating the PID component in a distinct manner
within the control structure. The aim is to address a significant gap in the existing literature
related to the lack of comparisons involving the different ways the MRAC and PID control
can be hybridized. The developed controllers undergo extensive testing in simulated scenarios
emphasizing model reference convergence and disturbance rejection. The results underscore
that hybridizing the MRAC mechanism with a PID controller offers notable advantages in both
performance and robustness, especially concerning disturbance rejection, a scenario in which
the hybrid controllers improved by around 70 and 75 % the mean square error obtained with a
pure MRAC controller. In general, these findings highlight the potential benefits of combining
MRAC and PID techniques, providing insight into the improved control system performance
that PID control can bring to conventional MRAC schemes.

Keywords: Adaptive control, PID control, hybrid control structure, process control.

1. INTRODUCTION However, an emerging trend involves hybridizing MRAC

controllers with Proportional, Integral, and Derivative
Model Reference Adaptive Control (MRAC) employs a (PID) controllers, which provides potential advantages in
closed-loop controller with adjustable parameters to op- terms of convergence, as discussed in detail by Zhang
timize the system response and achieve a desired dynamic and Wei (2016). This hybridization strategy represents a
behavior in closed-loop. Specifically, in MRAC controllers, nuanced approach that can potentially improve the adapt-
the adaptation mechanism dynamically modifies the con- ability and performance of MRAC controllers in various
troller parameters to align the process output with the applications.
output of the reference model. This model sets the ex-
In the existing literature, some examples of the combina-
pected (reference) output response that the system aims
tion of PID and MRAC control can be found. Amrane
to track (Åström and Wittenmark, 2013). Thus, the use of and Chaiba (2015) proposed an MRAC-PID for a double-
MRAC controllers provides the advantage of accommodat- feed induction generator. The study conducted by Pawar
ing system deviations from the reference response defined and Parvat (2015) introduced an MRAC-PID controller
by the reference model, whether these deviations arise for an inverted pendulum system; the integration of both
from uncertainties or disturbances (Shekhar and Sharma, control systems (i.e., MRAC and PID) was carried out to
2018). mitigate the limitations associated with traditional MRAC
MRAC control finds prominent application in various do- techniques. Zhou et al. (2016) designed an MRAC-PID
mains, particularly in the control of robotic manipulators controller for an inertial-stabilized aircraft platform fo-
(Zhang and Wei, 2017), motor control (Humaidi et al., cused on disturbance rejection. By and large, these exam-
2017; Nguyen et al., 2018), and diverse applications such ples collectively underscore the diverse applications and
as the stabilization of pendulum systems (Mendez et al., advantages sought through the combination of PID and
2020). It should be noted that, in most cases, the MRAC MRAC methodologies, showcasing the benefits of such
mechanism is applied directly without including additional hybrid control systems in various domains. However, the
auxiliary control schemes to address robustness issues. main gap identified in the literature is the lack of consensus
⋆ This work is a result of the CyberGreen Project, PID2021- regarding the optimal approach to integrate the MRAC
122560OBI00, and the Agroconnect ([Link])
mechanism with a PID controller. Moreover, there is a
facilities, grant EQC2019-006658-P, both funded by dearth of comparisons involving the different ways in which
MCIN/AEI/10.13039/501100011033 and by ERDF A way to MRAC and PID control can be hybridized. To the authors’
make Europe. Igor M. L. Pataro acknowledges the financial support knowledge, the only existing comparison is that proposed
of the National Council for Scientific and Technological Development by Zhang and Wei (2016), which compared a pure MRAC
(CNPq, Brazil) under grant 201143/2019 − 4.

© 2024 the authors. Accepted by IFAC for publication 31

under a Creative Commons License CC-BY-NC-ND
IFAC PID 2024
Almería, Spain | June 12-14, 2024

scheme with a hybrid MRAC-PID controller in terms of Given that the main objective of the proposed controller
convergence. However, in that work, only a hybrid scheme is to minimize this error, it is logical to formulate a
is employed, in which the MRAC mechanism adjusts the corresponding differential equation. Using Eqs. (1), (2),
control signal of the PID controller. and (3) yields the following result:
On the basis of the above literature review, this paper con- dem (t)
= −a1,m · em (t) − (a2 · θ2 (t) + a1 − a1,m ) · y(t)
ducts a comparative analysis of three MRAC schemes de- dt
veloped following the Lyapunov theory. It specifically com- + (a2 · θ1 (t) − a2,m ) · uc (t). (5)
pares a pure MRAC controller with two hybrid MRAC- It should be noted that the error goes to zero if the adapt-
PID schemes, each integrating the PID component in a able parameters (that is, θ1 and θ2 ) reach the following
different manner within the control structure. Particularly, equilibrium values:
in the first hybrid controller, the MRAC mechanism is a2,m a1,m − a1
responsible for adjusting the reference of the PID control θ1 = , θ2 = . (6)
loop. On the contrary, in the second hybrid controller, the a2 a2
MRAC mechanism adjusts the PID control signal. The de- To create a mechanism to adjust the parameters θ1 and
sign of both approaches is developed following the princi- θ2 to the previous values, assume a constant term given
ples outlined by Åström and Wittenmark (2013) regarding by a2 γ such that a2 γ > 0 and introduce the following
the Lyapunov theory. Moreover, the corresponding block Lyapunov function (time dependence has been omitted
diagrams are fully depicted, allowing for practical imple- for the sake of clarity) according to the ideas presented
mentation. Finally, the controllers are compared through by Åström and Wittenmark (2013):
simulations, utilizing two distinct scenarios focused on 1  1
model reference convergence and disturbance rejection. V (em , θ1 , θ2 ) = · e2m + · (a2 · θ2 + a1 − a1,m )2
2 a2 γ
The paper is structured as follows: Section 2 is dedicated 1
· (a2 · θ1 − a2,m )2 .

+ (7)
to the design of the different controllers. Section 3 presents a2 γ
and analyzes the main results derived from the application Note that Eq. (7) is differentiated with respect to time,
of the developed controllers in the proposed simulation thus allowing one to determine its Lyapunov stability:
scenarios. Lastly, Section 4 outlines the key conclusions
drawn and offers recommendations for future research. dV dem 1 dθ2
= em · + · (a2 · θ2 + a1 − a1,m ) ·
dt dt γ dt
2. CONTROLLERS’ DESIGN 1 dθ1
+ · (a2 · θ1 − a2,m ) · . (8)
bγ dt
In this section, MRAC-based control solutions are devel- Thus, by introducing Eq. (4), the previous equation be-
oped for the simplest case: a first-order system without comes:
time delay. First, the MRAC mechanism based on the Lya- dV 1
punov theory is depicted. Afterward, two hybrid MRAC- = −a1,m · e2m + · (a2 · θ2 + a1 − a1,m )
dt γ
PID control structures based on the same principles are
dθ2 1
proposed. ·( − γ · y · em ) + · (a2 · θ1 − a2,m )
dt γ
2.1 MRAC based on the Lyapunov theory dθ1
·( + γ · uc · em ), (9)
dt
Take into account the first-order system defined by the so that if the adaptive parameters are updated as:
equation: dθ1 dθ2
dy(t) = −γ · uc · em , = γ · y · em , (10)
= −a1 · y(t) + a2 · u(t), (1) dt dt
dt Eq. (9) results in:
where u(t) represents the input to the process, y(t) is the
dV
observable output of the process, and a1 and a2 are the = −a1,m · e2m . (11)
unknown parameters of the process. dt
Since it is negative semidefinite, it is Lyapunov stable
Now, consider the desired dynamics in closed-loop as (notice that a1,m must be positive for the closed-loop
expressed by the equation: system to be stable). Furthermore, the boundedness and
dym (t) convergence of Eq. (11) can be proven if it is locally
= −a1,m · ym (t) + a2,m · uc (t), (2)
dt Lipschitz (Åström and Wittenmark, 2013). Therefore, by
where ym (t) signifies the desired closed-loop response, a1,m differentiating the equation with respect to time, we get
and a2,m are predetermined gains, and uc (t) represents the the following:
command signal intended as a reference. d2 V dem
= −2 · a1,m · em · . (12)
Let now introduce an adaptive control law to integrate the dt2 dt
2
adaptation mechanism as follows: Now, if Eq. (4) is substituted, it is evident that ddtV2 is
u(t) = θ1 (t) · uc (t) − θ2 (t) · y(t). (3) a function of em , y, and uc , which are bounded signals.
2
This implies that ddtV2 is bounded and dV dt is continuous,
At this point, let define the error between the actual so the system is Lyapunov stable and locally Lipschitz.
process output and the desired behavior as: More details on this issue can be found in (Åström and
em (t) = y(t) − ym (t), (4) Wittenmark, 2013).

32
IFAC PID 2024
Almería, Spain | June 12-14, 2024

The MRAC controller can now be displayed using the where:  

previous development, as shown in Fig. 1. In this scheme, 1
Eq. (1) is included in the block System, whereas Eq. (2) is uP ID (t) = Kp · 1 + · g + Td · p · eP ID (t), (14)
Ti
reflected in the block Reference model. d
where p and g have been introduced as the operators dt
Rt
and ζ=0 dζ, respectively, Kp is the proportional gain, Ti
the integral time, and Td is the derivative time of the PID
controller. Moreover, the PID error equation is given by:
eP ID (t) = uc (t) − y(t). (15)
Note that the model reference error is defined as in Eq. (4).
As pointed out in Section 2.1, the objective is to minimize
the error in Eq. (4) so that the differential equation in this
case can be formulated as:
Fig. 1. Schematic diagram of the pure MRAC controller. dem (t)
= −a1,m · em (t) − (a2 · θ2 (t) + a1 − a1,m ) · y(t)
dt
+ a2 · θ1 (t) · uP ID (t) − a2,m · uc (t). (16)
2.2 Hydrid MRAC-PID: Solution 1
By incorporating now Eq. (15) and adding and subtract-
One of the primary limitations of an MRAC structure ing uP ID on the right-hand side, the previous equation
is its relatively weak robustness capability compared to becomes (note that the dependence with time has been
other control structures (Vargas-Martı́nez et al., 2015). To omitted for the sake of clarity):
address this limitation, a conventional PID controller can dem
be incorporated into the MRAC control loop, resulting = −a1,m · em − a2,m · eP ID − (a2 · θ2 + a2,m + a1
dt
in a hybrid MRAC-PID control solution. This approach − a1,m ) · y + (a2 · θ1 + 1) · uP ID − uP ID . (17)
enables us to exploit the power and robustness of PID
controllers within an MRAC mechanism. As can be observed, in this case, the error will go to zero if
the adaptive parameters converge to the following values:
In fact, one of the most straightforward methods of in- −1 a1,m − a1 − a2,m
tegrating the PID and MRAC strategies involves substi- θ1 = , θ2 = . (18)
a2 a2
tuting the system block from the preceding section (see
Fig. 1), with a PID-based feedback control loop, as shown It is also important to mention that the error in this case
in Fig. 2. Note that in this case, the development shown does not depend only on the em -term, but the error of
in the previous section is the same, the only difference is the feedback control loop (i.e., eP ID ) is also involved,
that Eq. (1) reflects closed-loop system dynamics instead as it appears directly in Eq. (17) and through the term
of the open-loop ones. Thus, in this case, the MRAC uP ID as defined in Eq. (14). Thus, it becomes evident that
mechanism is responsible for adjusting the setpoint of the achieving convergence between the model and the desired
PID feedback control loop in order to obtain the desired behavior requires a zero PID error signal (eP ID ). Since
dynamics imposed by the reference model. eP ID (t) = uc (t) − y(t) and em (t) = y(t) − ym (t), at the
convergence point, the system output (y) must be equal
to both the command signal (uc ) and the desired output
(ym ). This condition is achievable only if, considering first-
order reference models, they maintain unity gain, that is,
a1,m = a2,m in Eq. (2), which is typically imposed in
process control for tracking control problems. Note that,
by doing this, at the convergence point, uc (t) = ym (t) =
y(t). This is an important remark that must be taken into
account for the application of this hybrid controller.
At this point, let define a Lyapunov function following the
Fig. 2. Schematic diagram of the hybrid MRAC-PID principles outlined in Section 2.1 as follows:
controller, solution 1.
1  1
V (em , θ1 , θ2 ) = · e2m + · (a2 · θ2 + a2,m + a1
2 a2 γ
2.3 Hydrid MRAC-PID: Solution 2 1
− a1,m )2 + · (a2 · θ1 + 1)2 .

(19)
a2 γ
As mentioned in the scheme outlined in the preceding By differentiating the previous equation with respect to
section, the MRAC mechanism adjusts the reference of time and incorporating Eq. (17) we obtain:
the PID control loop. An alternative approach to hybridize dV 1
MRAC and PID involves the MRAC mechanism directly = −a1,m · e2m − a2,m · em · eP ID + · (a2 · θ2 + a2,m
modifying the control signal computed by the PID con- dt γ
troller, following the ideas presented by Khan and Swamy dθ2 1
+ a1 − a1,m ) · ( − γ · y · em ) + · (a2 · θ1 + 1)
(2016). For this case, consider Eqs. (1) and (2) and the dt γ
following adaptive control law: dθ1
·( + γ · uP ID · em ) − uP ID · em , (20)
u(t) = θ1 (t) · uP ID (t) − θ2 (t) · y(t). (13) dt

33
IFAC PID 2024
Almería, Spain | June 12-14, 2024

so that if the adaptive parameters are updated as: than the open-loop dynamics, resulting in Kp = 0.833 [-]
dθ1 dθ2 and Ti = 1 s. It is important to note that the configuration
= −γ · uP ID · em , = γ · y · em , (21) employed for the PID equation adhered to the ideal form.
dt dt
the following equation is derived: With respect to the tuning of the γ-parameter, this will
be discussed in the following section.
dV
= −a1,m · e2m − a2,m · em · eP ID − uP ID · em , (22)
dt 3.1 Model reference adjustment and convergence
which, again, is a negative semi-definite and, therefore,
Lyapunov stable. Note that the previous equation is neg- The initial simulation experiment assessed the convergence
ative semi-definite since i) it contains the term e2m in the of the different control structures to the reference model.
first term, ii) in the second, it has the multiplication of To achieve this, the reference signal uc was set as a square
em ·eP ID , which are errors with the same sign, since Eq. (2) wave with an amplitude of 2 and a frequency of 0.001 Hz.
must have a unity gain, and iii) it contains the term uP ID · Furthermore, three different γ-values (0.001, 0.01, and 0.1)
em in the third term. In this case, please take into account were examined for each controller to analyze the impact of
the definition in Eq. (14),
h so that the third i term in Eq. (22) this parameter on the controller performance. The results
1 obtained are presented in Fig. 4.
can be written as Kp · 1 + Ti · g + Td · p ·eP ID ·em , which,
again, is a multiplication between the error signals.
MRAC ym y( =0.1) y( =0.01) y( =0.001)
2
Now, as in Section 2.1, the boundedness and convergence
can be proven if it is locally Lipschit. Consequently, by 0
differentiating the equation with respect to time, we obtain
-2
the following Convergence

Hybrid MRAC-PID, solution 1

d2 V dem dem duP ID 2

2
= −2 · a1,m · em · − (uP ID · + em · )
dt dt dt dt 0
deP ID dem
− a2,m · (em · + eP ID · ). (23) -2
dt dt Convergence
Convergence
2 Hybrid MRAC-PID, solution 2
Again, ddtV2 is bounded, and dV dt is continuous, so the
2

system is Lyapunov stable and locally Lipschitz.

0

The schematic diagram of the MRAC controller can now

-2
be depicted in accordance with the previous development, 0 100 200 300 400 500 600
Simulation time [-]
as illustrated in Fig. 3. As in Section 2.1, in this scheme,
Eq. (1) is included in the block System, whereas Eq. (2) is Fig. 4. Outputs of the controller with the different γ-
reflected in the block Reference model. values. Note that ym is the output of the reference
model, while the rest of the signals are the output
with the different γ values for each configuration.

It is worth noting that Fig. 4 exclusively displays the sys-

tem output, a choice made to enhance the clarity of result
visualization with varying γ-values. The examination of
the control signal behavior and a detailed analysis of adap-
tive parameters (θ1 and θ2 ) will be thoroughly discussed
in the subsequent section. However, the moment in which
the adaptive parameters (θ1 and θ2 ) reached convergence
for the best γ-value configuration for each case is marked
Fig. 3. Schematic diagram of the hybrid MRAC-PID in the figure.
controller, solution 2.
As can be seen, in general, low γ-values led to slow con-
vergence in all cases, with special relevance in the MRAC
3. RESULTS and Hybrid MRAC-PID, solution 1. This is because in
these configurations, the adjustment mechanism has an
This section shows the results obtained with the differ- important weight in the controller performance since, in
ent control configurations, that is, i) MRAC, ii) Hybrid the first, it directly computes the control signal, and in
MRAC-PID, solution 1, iii) Hybrid MRAC-PID, solution the second, it is responsible for modifying the reference
2. The controllers were tested in two distinct scenarios, of the PID control loop. In the case of Hybrid MRAC-
focusing on model reference convergence and disturbance PID solution 2, the configuration of the control scheme
rejection. For the different tests, the system in Eq. (1) gives certain advantages since, as can be seen in Eq. (16),
was considered as a1 = 0.5 and a2 = 1, while the de- the PID controller is also involved in the adjustment to
sired behavior in Eq. (2) as a1,m = 2 and a2,m = 2. the reference model, so the convergence was faster for all
Furthermore, the PID controller was configured only with the γ-values. It should also be noted that although higher
proportional and integral terms following the pole-zero values of γ provided a faster adjustment, overshoots were
cancelation method (Åström and Hägglund, 2006). The observed in MRAC and hybrid MRAC-PID, solution 1,
closed-loop time constant was set to be 0.6 times faster which must be considered in practical implementations.

34
IFAC PID 2024
Almería, Spain | June 12-14, 2024

To quantitatively observe the advantages of each controller (1) ym y

in terms of adjustment, Tab. 1 presents the Mean Square 2

Output
Error (MSE) of each controller with different γ-values with 0

respect to the reference model. As can be seen, in all cases, -2

Disturbance
the smallest MSE value was obtained with γ = 0.1, and, (2) u
in particular, the lowest value was provided by the Hybrid 5

Control
signal
MRAC-PID configuration, solution 2, with an MSE of 0

0.0043. Another aspect to highlight is that the MRAC -5

scheme provided a smaller MSE value than Hybrid MRAC- (3) 1 2

2 1.5
PID, solution 1, so the inclusion of the PID following that
1

[-]

[-]
scheme did not provide advantages in terms of adjustment, 1

2
0.5
although it does provide other types of advantages as will 0 0
0 500 1000 1500
be shown in the next section. Simulation time [-]

Table 1. MSE of the different control configu-

rations with the different γ-values with respect Fig. 6. MRAC control system performance during simu-
to the reference model. lation. All the signals are according to the scheme
depicted in Fig. 1.
γ =0.1 γ =0.01 γ =0.001 2

MRAC 0.0050 0.0407 0.3104 1.5

1 (1) ym y

Hybrid MRAC-PID, solution 1 0.0055 0.0527 0.5033

0.5

2 0

Output
Hybrid MRAC-PID, solution 2 0.0043 0.0401 0.2799 0

-2
Disturbance
3.2 Disturbance rejection (2) u PID

2
Control
signal

0
In the second test, the aim was to assess the performance -2

and robustness of the controllers in the presence of distur- -4

bances. Thus, the reference signal (i.e., uc ) was maintained (3) 1 2

as in the previous section. However, a disturbance was 1 0.2

[-]

[-]
0.1
introduced into the process output at simulation time
1

2
0.5
0

575, as illustrated in Fig. 5. Note that this disturbance 0

0 500 1000 1500
-0.1

consisted of a step of 1 plus band-limited white noise, with Simulation time [-]

a noise power of 5 and a sample time of 60. The results

obtained in this test are presented in Figs. 6, 7, and 8 Fig. 7. Hybrid MRAC-PID controller, solution 1. All the
for the MRAC, hybrid MRAC-PID solution 1, and hybrid signals are according to the scheme depicted in Fig. 2.
MRAC-PID solution 2, respectively. ym
(1) y
2
Disturbance
Output

2
0

-2
1.5 Disturbance
(2) u
2
Control

1
signal

-2
0.5

(3) 1 2
1.5 0.2
0 1 0.1
[-]

[-]

0 750 1500
1

Simulation time [-] 0.5 0

0 -0.1
0 500 1000 1500
Fig. 5. Disturbance signal used as a reference in the Simulation time [-]

simulation test.
Fig. 8. Hybrid MRAC-PID controller, solution 2. All the
As can be seen, until the moment when the disturbance signals are according to the scheme depicted in Fig. 3.
was introduced (see the mark in Figs. 6-(1), 7-(1), and
8-(1)), all controllers almost reached convergence in the In the case of the MRAC controller (see Fig. 6), it can
adaptation parameters (θ1 and θ2 ), see Figs. 6-(3), 7-(3), be seen how it also managed to reject the disturbance,
and 8-(3), observing the same results and trends as those even though it did not include a PID mechanism and
discussed in the previous section. The main differences its control action lies purely in the integrators of the
between the controllers can be seen in the rejection of the adaptation law. However, it can be seen that the con-
disturbance. In general, the two solutions that incorporate vergence of the adaptation parameters was completely
a PID controller (Figs. 7 and 8) exhibited faster rejection altered (see Fig. 6)-(3)). This was quite significant since
of disturbances. This was attributed to the PID controller the parameters’ values showed an upward and oscillating
operating on a faster time scale compared to the MRAC trend, which translated into increasingly aggressive control
mechanism, allowing for a quicker response to reject distur- actions (see Fig. 6)-(2)). It should be noted that the test
bances. This fact also produced an extra advantage, which was stopped at instant 1500 for improved visualization.
was that the convergence of the adaptation parameters was However, this trend persisted over time, leading to pro-
hardly altered in both cases, see Figs. 7-(3) and 8-(3). gressively more aggressive control actions. Consequently,

35
IFAC PID 2024
Almería, Spain | June 12-14, 2024

in practical implementations, particularly for systems with and/or time delay with input constraints and obtaining
input constraints, applying such aggressive controls could practical rules for tuning the γ-parameter.
be impossible or pose significant challenges. In addition,
this behavior indicated instability, which is one of the REFERENCES
problems stated by Åström and Wittenmark (2013) when
a pure MRAC controller is subjected to large amplitude Amrane, F. and Chaiba, A. (2015). Comparative study
changes either in the reference or caused by disturbances, on the performance of Fuzzy-PID and MRAC-PID con-
as in this case. trollers based on DPC with SVM for DFIG using MPPT
strategy. In International Conference on Automatic
Finally, Tab. 2 shows the MSE of each of the controllers control, Telecommunications and Signals (ICATS15), 1–
with respect to the reference model, as well as the Control 6.
Effort Total Variation (CETV), which respresents the sum Åström, K.J. and Wittenmark, B. (2013). Adaptive con-
of the absolute values of the increments of the control trol. Courier Corporation.
signal. As observed, the magnitude of MSE obtained by Åström, K.J. and Hägglund, T. (2006). Advanced PID
the schemes that incorporate PID is significantly lower, control. ISA-The Instrumentation, Systems and Au-
improving by around 70 and 75 % the performance of the tomation Society.
MRAC controller for hybrid MRAC-PID solution 1 and Humaidi, A.J., Hameed, A.H., and Hameed, M.R. (2017).
solution 2, respectively. Robust adaptive speed control for DC motor using
Table 2. MSE of the different control configu- novel weighted E-modified MRAC. In 2017 IEEE
rations with respect to the reference model. International Conference on Power, Control, Signals
and Instrumentation Engineering (ICPCSI), 313–319.
MSE CETV IEEE.
MRAC 0.0188 228.13 Khan, A. and Swamy, M.S. (2016). Modified MRAC based
Hybrid MRAC-PID, solution 1 0.0058 171.02 on Lyapunov theory for improved controller efficiency.
Hybrid MRAC-PID, solution 2 0.0044 153.35 In 2016 International Conference on Automatic Control
and Dynamic Optimization Techniques (ICACDOT),
4. CONCLUSION
989–995. IEEE.
This paper provides a comparative analysis of three Mendez, E., Baltazar-Reyes, G., Macias, I., Vargas-
MRAC schemes grounded in the Lyapunov theory. Specifi- Martinez, A., de Jesus Lozoya-Santos, J., Ramirez-
cally, it contrasts a pure MRAC controller with two hybrid Mendoza, R., Morales-Menendez, R., and Molina, A.
MRAC-PID schemes, each incorporating the PID compo- (2020). ANN based MRAC-PID controller implemen-
nent in a distinct manner within the control structure. The tation for a Furuta pendulum system stabilization. Ad-
controllers were tested in two different scenarios focused vances in Science, Technology and Engineering Systems
on model reference convergence and disturbance rejection. Journal, 5(3), 324–333.
The results obtained allow us to draw the following con- Nguyen, A.T., Rafaq, M.S., Choi, H.H., and Jung, J.W.
clusions: (2018). A model reference adaptive control based speed
controller for a surface-mounted permanent magnet syn-
• All controllers demonstrated satisfactory adjustment chronous motor drive. IEEE Transactions on Industrial
to the reference model. In particular, the proposed Electronics, 65(12), 9399–9409.
hybrid MRAC-PID solution 2 (with the MRAC mech- Pawar, R. and Parvat, B. (2015). Design and implemen-
anism adjusting the control signal of the PID con- tation of MRAC and modified MRAC technique for
troller) exhibited improved MSE values, underscoring inverted pendulum. In 2015 International Conference
the advantages of incorporating a PID into the adap- on Pervasive Computing (ICPC), 1–6. IEEE.
tation mechanism according to this specific scheme Shekhar, A. and Sharma, A. (2018). Review of model
for the adjustment of the reference model. reference adaptive control. In 2018 International Con-
• Regarding disturbance rejection, the schemes incor- ference on Information, Communication, Engineering
porating PID exhibited markedly superior perfor- and Technology (ICICET), 1–5. IEEE.
mance. This can be attributed to the PID mechanism Vargas-Martı́nez, A., Minchala Avila, L.I., Zhang, Y.,
operating on a faster time scale compared to the Garza-Castañón, L.E., and Badihi, H. (2015). Hybrid
MRAC one. Thus, the hybrid MRAC-PID solution adaptive fault-tolerant control algorithms for voltage
1 and solution 2 improved by around 70 and 75 % and frequency regulation of an islanded microgrid. In-
the MSE performance of the MRAC controller, re- ternational Transactions on Electrical Energy Systems,
spectively. Moreover, the inclusion of the PID played 25(5), 827–844.
a fundamental role in the robustness of the MRAC Zhang, D. and Wei, B. (2016). Convergence performance
controller facing large amplitude changes caused by comparisons of PID, MRAC, and PID+MRAC hybrid
disturbances. controller. Frontiers of Mechanical Engineering, 11,
213–217.
In general, the findings of this work underscore the poten-
Zhang, D. and Wei, B. (2017). A review on model refer-
tial advantages of integrating MRAC and PID techniques,
ence adaptive control of robotic manipulators. Annual
providing insights on the improved performance of the
Reviews in Control, 43, 188–198.
control system that PID control can bring to traditional
Zhou, X., Yang, C., Cai, T., et al. (2016). A model
MRAC schemes.
reference adaptive control/PID compound scheme on
Future work will be aimed at analyzing the performance of disturbance rejection for an aerial inertially stabilized
the different control schemes for systems of higher orders platform. Journal of Sensors, 2016.

36