Expert Systems With Applications 229 (2023) 120518

Contents lists available at ScienceDirect

Expert Systems With Applications

journal homepage: www.elsevier.com/locate/eswa

Attitude control of a quadrotor using PID controller based on differential

evolution algorithm
Ayhan Gün
Kütahya Dumlupınar University Engineering Faculty Electrical-Electronics Engineering Department, Evliya Çelebi Yerleşkesi Kütahya, Türkiye

ARTICLE INFO ABSTRACT

Keywords: In this study, an energy efficiency study has been carried out through the moment values of a quadrotor
Energy efficiency for quadrotor applied in the position control with the differential evolution algorithm (DE), which is under the evolutionary
Cost function algorithms among meta-heuristic algorithms particle swarm optimization (PSO) which is among swarm-based
Differential evolution
algorithms and gravity search algorithm (GSA) among physics-based algorithms and charged system search
Multi objective optimization PID
(CSS) algorithm, which are all well-known in the literature. The efficient use of energy in electrically powered
Quadrotor
flying mechanisms is important in terms of completing its task. The aim of this study is to obtain proportional–
integral–derivative (PID) coefficients that minimize the energy consumption caused by quadrotor moments.
First of all, 12 PID coefficients calculated with traditional Ziegler Nicholes, and trying-error methods for the
control of the quadrotor have been trained with DE, PSO, GSA and CSS algorithms and their cost values have
been found for this problem. In the study, optimum PID coefficients have been found separately with each
algorithm and torque data giving energy efficiency have been obtained by applying them to the state equations.
Although the cost minimization convergence value obtained from DE does not look good compared to other
algorithms, it has been observed that the moment values of the obtained PID coefficients in the control program
are lower. In the study, it has been shown that the PID coefficients obtained from DE reach the target with
approximately less energy than the maximum-moment value of the PSO, which provides the closest moment
value. It can be thought that this value contributes to energy consumption in each orbital movement of the
quadrotor. The simulation results show that the coefficients obtained from DE provide better energy efficiency
and are superior to other algorithms compared in terms of energy efficiency.

1. Introduction was adjusted by using an independent linear SISO controller (Pounds

et al., 2010). In the study that provided the location determination
In recent years, quadrotor studies have accelerated the studies in of the quadrotor, quaternion-based feedback was developed without
nonlinear process control area. It is indispensable to use high maneuver using inertia matrix knowledge and control was made by paying at-
aircrafts that can conduct research in buildings and danger zones. Our tention to borders in entries (Guerrero-Castellanos et al., 2011). In
expectation from this type of aircrafts is that they fulfill certain tasks another study, a controller based on the usage of PD2 feedback has
such as research, analysis, and data collection. With this aim, quadrotor been proposed to balance Coriolis and gyroscopic moments (Tayebi
forms have been used due to their size, speed and versatility. It is & McGilvray, 2006). Lyapunov guaranteed determination and orbital
important to gather image, sound, and photograph data with regard to performance by using the determination theory was studied to deter-
the equilibrium that is provided in the air with aid of the four motors. mine the optimal performance of a quadrotor (Das et al., 2008). In the
The most important problem in the operation this type of aircrafts is to study by Zhang et al. (2011), a feedback controller estimate was made
use energy efficiently; therefore, the vast majority of the studies con- through sliding mode control to ensure quadrotor location stability
centrated on the moment control. First, the state-space equations have and that discontinuity of control rules in the control of the sliding
been developed in studies and solutions are produced based on these mode was avoided (Zhang et al., 2011). In another study, the problem
equations (Alexis et al., 2012; Bruno et al., 2009; Castillo et al., 2005, resulting from the continuous time model for small-scale quadrotor
2005; Das et al., 2008; Liu et al., 2013; Mian & Daobo, 2008; Zhang dynamics was addressed. To establish control, a continuous time pre-
et al., 2011). mention the necessity of flight and passage dynamic dictor - based subspace approach was added and results were provided
equity for the control of a quadrotor with six-degree freedom. A 4-kg in experimental studies (Bergamasco & Lovera, 2014). In the article
quadrotor that carried a 1-kg load was designed. Also, flight location

E-mail address: [email protected].

https://doi.org/10.1016/j.eswa.2023.120518
Received 1 September 2022; Received in revised form 18 May 2023; Accepted 18 May 2023
Available online 27 May 2023
0957-4174/© 2023 Elsevier Ltd. All rights reserved.
A. Gün Expert Systems With Applications 229 (2023) 120518

by Dong et al. (2015), control analysis for the time-varying behavior With the new algorithm study proposed by Amir Ramezani Dooraki
of drone swarm systems and problems with design were examined and Lee (2020) called Bio-inspired Flight Controller (BFC), no con-
and required and adequate conditions for time-varying behaviors were ventional controller such as PID or MPC has been used to control the
offered. In Cabecinhas et al. (2015) study, the fixed wind effect was quadrotor drone. The proposed BFC algorithm has the ability to fix
given from outside to the quadrotor moving alongside a predetermined the Quadcopter at the desired point and to follow (to keep track of)
orbit. An attempt was made at establishing control with the help of an the waypoints. It has been stated that the algorithm was implemented
experimental global controller, and it was calculated experimentally. (applied) in an AscTec Hummingbird quad helicopter in the Gazebo
In another study, the absolute location of data deficiency of drones simulation program and tested with different scenarios to fully measure
moving inside of a building was studied (Alexis et al., 2012). In this its capabilities, but the BFC algorithm could not overcome difficulties
study, a controller design was performed for robust location control for such as uncertainties in the odometry data and engine speed control
inadequately induced quadrotors. The proposed linear time invariant and that it is also possible to use theoretical approaches to increase the
controller included a robust compensator and a Proportional-Derivative learning speed and ability of the BFC in a real system. Sushama Nagpal
controller. The PD controller was designed to attain the desired orbit, (2017) In the study since the data used for disease prediction contains
and the robust compensator was added to limit the effect of uncertain- many features, GHQ and K-nearest neighbor algorithms have been used
ties (Liu et al., 2013). The Neural Network output feedback control law in the analysis of medical data. In the study, the number of features
for Quadrotor was developed using four control inputs with unmodeled has been reduced by an average of 66 percent and the prediction
dynamics by Dierks and Jagannathan (2010). Önder Efe (2011) pro- accuracy has been significantly increased. The results have also been
posed a PID type controller to prevent model mismatch, wind effects, compared with PSO (Harb & Desuky, 2014) and GSA (Karegowda et al.,
measurement noises and voltage losses in batteries. Krohling et al. 2011) and it has been seen that GSA performed better. In addition, it
(1997) presented a design method for determination of suitable PI/PID has been observed that the accuracy of the predictive model created
parameters on motion control systems. They used a genetic algorithm based on these reduced features has increased in many cases. Gravity
based approach for the minimization of integral of time-squared-error search algorithm (GSA) uses the force of gravity between individuals to
(ITSE) performance criterion and validated the effectiveness of their develop the population. Although GSA is an efficient population-based
method by simulating a PI/PID controlled AC servo drive. Salih et al. algorithm, it shows poor search performance and early convergence.
(2010) proposed a new model design method for the flight control of an To improve these problems, a multi-layered GSA called MLGSA has
autonomous quadrotor. To obtain the flight stability of the quadrotor, been proposed. Inspired by the two-layer structure of GSA, four layers
they first mathematically modeled the under actuated, unmanned air consisting of population, iteration-best, personal-best and global-best
vehicle, which has four thrusting independent propellers as inputs. have been created. Hierarchical interactions between the four layers
Then they designed a PID controller to successfully achieve the desired are dynamically applied at different stages of exploration to greatly
pitch, roll and yaw angles to move it to a desired location. To simplify enhance both the exploration and exploitation capabilities of the pop-
the controller design, they modified the state space representation and ulation. It has been compared with four particle swarm optimization
by applying it to a MATLAB/Simulink model, they showed that PID variants and excellent performance has been obtained (Wang et al.,
controllers are able to robustly stabilize the quadrotor flight with fast 2021). Optimization complications are solved by meta-heuristic meth-
response times. In the article named ‘‘Charged System Search (CSS) ods that make complex data simple, and the computational process
Algorithms for Optimal Tuning of PI Controllers’’, a new optimization is highly attractive due to its concentration on non-linear data, diver-
based on CSS algorithm has been proposed for tuning the Proportional sification, and accurate evaluation/computation behavior. Among the
Integral (PI) controller by Precup et al. (2012). However, it has been different meta-heuristic algorithms available, Tabu Search Optimiza-
stated that the CSS algorithm does not guarantee to provide the global tion (TSO) algorithms have strong performance due to local optima
minimum. The content of the study proposes to find the optimum escaping strategy and extensive extraction capability. Therefore, this
value of the PI controller coefficients of the Charged System Search article (Arandhakar et al., 2022), describes the TSO search technique to
(CSS) algorithms. CSS algorithms solve optimization problems aimed extract parasitic parameters of solar Photovoltaic (PV) modules under
at minimizing the objective functions expressed as the integral of the different climatic conditions. The dual diode design has been developed
squared control error and the squared output sensitivity functions. The and applied from an operational point of view based on the extraction
new CSS algorithm application has been compared with (PSO) Particle issue. Six parameters of the solar cell, namely IPh, I 01, I 02, RS, RP,
Swarm Optimization and (GSA) Gravity Search Algorithm and it has a1 have been emulated, the data obtained have been investigated using
been seen that the CSS algorithm gave better results in this study. In the TSO approach, at the same time the data obtained have been compared
study, it has been seen that PI controller tests were not performed, and with metaheuristic algorithms such as Gravitational Search Algorithm
the evaluation was made only through the cost function. Another work (GSA), Lightning Search Algorithm (LSA), Pattern Search (PS), Particle
presents a Charged System Search (CSS) algorithm to solve robotic Swarm Optimization (PSO) and Genetic Algorithm (GA). When com-
punch path optimization in the printed circuit board (PCB) manufactur- pared to existing algorithms, it has been seen that TSO spent much
ing industries. Most of a Robotic PCB driller’s working time is spent for less time to obtain the six parameters. Roger Miranda-Colorado (2020)
moving the drill bit between holes. An optimized path means minimal presents a new robust controller for PID tuning and trajectory track-
cost of operating the robot. The drill path is formed of a series of ing tasks applied to quadrotor equations. In the proposed structure,
potential locations (positions) where holes will be drilled. The greater the quadrotor position is controlled by the PID (proportional integral
the number of holes required, the greater the complexity of finding derivative) controller, while the directional control is provided via a
the optimized path. Recently developed CSS algorithm has proposed model-based controller (MPC-Model Predictive Control). A controller
to solve this complex problem with minimal computation (calculation) using the cuckoo search algorithm, a power reduction methodology
time. The performance of the CSS algorithm has been tested and that includes gain adjustment, and the minimum jitter trajectory design
validated by four case studies from the literature. The computational phase have been combined. The performance of the new controller
experience conducted in this research shows that the proposed algo- has been evaluated in the case of free distortion and under the in-
rithm is capable of efficiently finding the optimal path for drilling fluence of parametric uncertainty and aerodynamic distortions. The
PCB holes with reasonable computation time. When compared to other new controller has been compared with two linear PID controllers and
metaheuristics PSO, ACS, and FA methods, It ̇ has been seen that CSS one non-linear sliding mode-based controller. Numerical simulations
had less computational cost and reached the optimum with fewer show the superiority of the proposed structure and its robustness
iterations. It has been stated that the performance of the CSS algorithm against different types of perturbations. It has also been proven that
can be improved for larger problem solutions (Kanagaraj et al., 2015). the power demanded by any controller is reduced when the power

2
A. Gün Expert Systems With Applications 229 (2023) 120518

reduction methodology is used. Paper of the Aristotelis Kapnopoulos control give better results than PSO. Disturbances given at certain
(2022) presents a collaborative PSO approach involving two controllers intervals also being random also affects the control structure. For
to precisely set the parameters of one controller to solve a quadrotor’s correction of the effects of disturbances, limit values are not used in the
path tracking problem. That is, the two control schemes include an software. In the second scenario, the quadrotor needs very large forces
MPC controller responsible for position control and a PID that adjusts to correct both the angular values and the effects of disturbances. It
the behavior of the quadrotor. To deal with the complexity of the has been observed that phi, theta, and psi angles can all make target
optimization problem corresponding to the tuning of the method, a changes of 1 rad. In the third scenario, only 50 m of vertical movement
collaborative PSO approach has been proposed that divides the pa- is considered, while angular changes are not. In Fig. 30, total moment
rameters to be tuned into two separate sets; here the first set contains values are examined only for this motion. According to this scenario,
the parameters of the MPC position controller, the second deals with it was seen that the target was reached with less torque with the
the parameters used by the PID state controller. The two groups col- coefficients obtained from DE PID.
laborate by exchanging information to discover improved solutions.
The presented tuning method has been evaluated on 3 different orbital 2. Mathematical design of a quadrotor
states(cases). In the first two cases, the proposed approach is used to
set the controllers on a given spatial trajectory and then focuses on Certain hypotheses in the definition of the mathematical design
the robustness of the method when evaluating the quadrotor tracking of a quadrotor in Fig. 1 should be taken into consideration (Amir
capability on the same trajectory. The performance of the proposed Ramezani Dooraki & Lee, 2020; Lee et al., 2009; Zuo & Wang, 2014)
methodology has been validated by benchmarking tests against GA and Hypothesis-1: It should be accepted that the structure is symmetric
PSO algorithms as well as a standard PID control scheme. Statistical and straight according to x, y, z - axis due to the nonlinear structure of
tests on simulated results show that the collaborative structure provides the quadrotor. Hypothesis-2: Aerodynamic forces and moments should
tuned parameters with better trajectory tracking ability and robustness be ignored at low speeds (Zuo & Wang, 2014). x𝑏 , y𝑏 , z𝑏 and x, y, z
compared to its competitors. The Capacity Vehicle Routing Problem show earth coordinate and quadrotor axis, respectively. The center of
(CVRP) is a well-known unified problem. In Israel Pereira Souza (2023) gravity of these axes is O. The absolute location of the quadrotor is
paper, a hybrid algorithm based on a discrete adaptation of the Differ- defined with 𝜁 = [𝑥 𝑦 𝑧]𝑇 and its angles 𝛩 = [𝜑 𝜃 𝜓]𝑇 . The quadrotor
[ ]𝑇
ential Development (DE) metaheuristic method designed for continuous has 𝑣 = 𝑣𝑥 𝑣𝑦 𝑣𝑧 orbit speed, 𝛺 = [𝑝 𝑞 𝑟]𝑇 angle speed, and a rigid
problems combined with local search procedures has been proposed to structure with six degrees of freedom (Kuipers, 1999; Lee et al., 2009).
solve CVRP. An individual in the algorithm represents a CVRP solution. At first, the leading edge of the quadrotor is accepted as the direction
The algorithm, instead of arithmetically changing individuals, presents of the x orbit and the frame within which it is located is called the
a separate adaptation for DI, which consists of exchanging individuals- base main frame (B). Aerodynamic friction moments because of angular
customers, taking into account the route positions. It is stated that the speeds according to the B frame occur (Mian & Daobo, 2008). Forces
proposed algorithm, called open source CDELS, is significantly better produced by rotors of the quadrotor are provided by 𝑢𝑖 = 𝐾𝑇 ⋅ 𝜔2𝑖
than high technology methods with a 99 percent confidence level. The equation. K𝑇 stands for lift coefficient, and 𝜔𝑖 stands for angular rotor
study of Noroozi et al. (2022) introduces the Gold Search Optimization velocity. The direction of force that is produced by the impeller speed
Algorithm (GSO), which is an effective population-based optimization in each rotor and the drag moments are in directions that are contrary
algorithm for numerical function optimization. It has been stated that to each other. The slopes of the flap that are attached to the motors
the new algorithm has a simple but effective strategy for solving starting the quadrotor in this design are important. The direction of
complex problems. The proposed algorithm has been compared with 23 rotation of two reciprocal motors and the direction of rotation of the
single-mode, multi-mode and fixed-size functions and the results have other two motors should be different, and the other motors should
been verified by comparing with well-known Gravity Search Algorithm move counter clockwise while two motors move clockwise (Ding &
(GSA), Sine-Cosine Algorithm (SCA), Gray Wolf Optimization. The Yu, 2013; Kuipers, 1999). Quadrotor starts flying due to of pressure
simulation results have shown that GSO is superior and can produce difference; since air pressure under Pal is larger than the air pressure
better optimal solutions compared to other competing algorithms. above. Some scenarios for, respectively, u1 , u2 , u3 , u4 and T1 , T2 , T3 ,
Many optimization problems and solutions have been in the stud- T4 -forces and moments that provide movements of the quadrotor are
ies (Arandhakar et al., 2022; Paez et al., 2021; Roger Miranda-Colorado, provided in Fig. 2 𝜔1 , 𝜔2 , 𝜔3 , and 𝜔4 are the angular velocity of flaps
2020; Tayebi & McGilvray, 2006; Wang et al., 2021). It is seen that attached to the rotor fuselage. Inputs of the vertical force from Fig. 1
Meta-heuristic algorithm studies produce different successful results (moving at the z-direction) are defined in the following manner:
for each problem. In fact, when looking at the comparative results in
data update algorithms in Meta-Heuristic studies, it is seen that the 𝑇1 = 𝑇2 = 𝑇3 = 𝑇4 (1)
results obtained depend on the simulation intervals, code length, the
( )
number of parameters in the problem, the mathematical calculation 𝑢1 = 𝐾𝑇 𝜔21 + 𝜔22 + 𝜔23 + 𝜔24 (2)
methods and the speed of the operating system. Since the algorithms
work for a very long time, the data obtained from the update studies According to the x- axes rolling -𝜑 from Fig. 2a, it is given by;
are used as off-grid in real-time applications. The methods used in 𝑇1 = 𝑇3 ≤ 𝑇2 < 𝑇4 (3)
the implementation process are expected to be effective and powerful.
Since DE algorithm is one of the efficient, easy to adapt and powerful Force of the motor inputs that provide rotation
algorithms, DE-based PID controller design was made in this study. ( )
𝑢2 = 𝐾𝑇 𝜔24 − 𝜔22 (4)
The rest of the work is organized as follows. Mathematical modeling
of the quadrotor is given in Section 2, position control in Section 3, According to the x-axes pitching angle 𝜃 from Fig. 2b
differential evolution algorithm in Section 4, DE-based PID controller
design in Section 5, multi-objective optimization and its results are 𝑇2 = 𝑇4 ≤ 𝑇3 < 𝑇1 (5)
given in Section 6, and the evaluation of the results obtained from Force of the motor inputs that provide pitch from Fig. 2c
each optimization with three different scenarios is done in Section 7. ( )
It is a very difficult problem to determine which of the optimization 𝑢3 = 𝐾𝑇 𝜔21 − 𝜔23 (6)
algorithms is better. In this study, first of all, the same randomly
Yaw angle according to the x-axes from Fig. 2c: 𝜓 counterclockwise
generated matrix pool rnd(15) was used for all optimizations. Although
it is seen that PSO converges more, DE force coefficients for Quadrotor 𝑇1 = 𝑇3 < 𝑇2 = 𝑇4 (7)

3
A. Gün Expert Systems With Applications 229 (2023) 120518

Fig. 1. Quadrotor configuration of roll-pitch-yaw Euler angles.

Fig. 2. Different forces produced by rotors of the quadrotor.

Yaw angle according to the x-axes from Fig. 2d: 𝜓 clockwise Table 1
Quadrotor unmanned aircraft rotor speed control.
𝑇2 = 𝑇4 < 𝑇1 = 𝑇3 (8) Direction Rotor 1 Rotor 2 Rotor 3 Rotor 4

Yaw force input is defined with Eq. (9). Upward movement ⇑ ⇑ ⇑ ⇑

( ) Pitch ⇑ – ⇓ –
𝑢4 = 𝐾𝑄 𝜔24 + 𝜔22 − 𝜔21 − 𝜔23 (9) Roll – ⇑ – ⇓
Yaw ⇓ ⇑ ⇓ ⇑

⎡cos 𝜃 cos 𝜓 sin 𝜃 cos 𝜓 sin 𝜙 − sin 𝜓 cos 𝜙 sin 𝜃 cos 𝜓 cos 𝜙 + sin 𝜓 sin 𝜙⎤
⎢ ⎥
𝑅 = ⎢ cos 𝜃 sin 𝜓 sin 𝜃 sin 𝜓 sin 𝜙 + cos 𝜓 cos 𝜙 sin 𝜃 sin 𝜓 cos 𝜙 − cos 𝜓 sin 𝜙⎥ equations are obtained to produce moment solutions that are dependent
⎢ − sin 𝜃 cos 𝜃 sin 𝜙 cos 𝜃 cos 𝜙 ⎥
⎣ ⎦ on quadrotor velocity, and velocity equations provided from these
(10) equations are used. The directions of movement for the moment pro-
duced by each rotor are provided in Table 1 (Chen et al., 2009;
Zuo, 2010). Rotation matrix in Eq. (10) is obtained by executing
𝑅 = 𝑅𝑧,𝜓 𝑅𝑦,𝜃 𝑅𝑥,𝜙 (11) Roll-pitch-yaw movements respectively, according to the earth coor-
dination system. If the rotation order changes, the rotation matrix
changes as well, and it is of importance to pay attention to this
Velocity sign in the equations varies according to the desired move order while commenting on rotation (Alexis et al., 2012; Chen &
direction of the quadrotor; Counter - clockwise rotation movement

4
A. Gün Expert Systems With Applications 229 (2023) 120518

Pérez-Arancibia, 2020; Kuipers, 1999; Vallejo-Alarcón et al., 2020)

Differential equations should be obtained, and each parameter should
be solved according to these equations to use forces that will determine 𝐽 𝛺̇ = (−𝛺 × 𝐽 𝛺) + 𝜏𝑓 − 𝜏𝑎 − 𝜏𝑔 (20)
the location of the quadrotor. There are two situations that should [ ]𝑇 [ ]𝑇
be considered while obtaining differential equations; forces on the R, gives the rotation matrix and 𝛺 = 𝑝 𝑞 𝑟 , 𝛩 = 𝜙 𝜃 𝜓
quadrotor, and moments on the quadrotor. Total forces applied to the gives the angular velocities in the locations of the quadrotor. Also, J
quadrotor (Bouadi et al., 2007; Cabecinhas et al., 2015; Das et al., 2008; provides the symmetric moment of inertia and 𝜏𝑓 delivers the moments
Guenard et al., 2008; Kendoul et al., 2007; Mahony et al., 2012; Tayebi that the quadrotor produces according to its fixed frame.
& McGilvray, 2006) are obtained with following equations:
⎡ 𝜏𝑓 𝑥 ⎤
𝜁̇ = 𝑣 ⎫ 𝜏𝑓 = ⎢ 𝜏𝑓 𝑦 ⎥ (21)
⎪ ⎢ ⎥
𝑚𝜁̈ ⎣ 𝜏𝑓 𝑧 ⎦
= 𝐹𝐺 + 𝐹𝑍 + 𝐹𝐷 ⎪
⎬ (12)
𝐹𝑆 = 𝑚𝑣̇ ( )
⎪ 𝜏𝑓 𝑥 = 𝑙 ⋅ 𝐾𝑇 𝜔21 − 𝜔23 (22)
⎪
𝐹𝑆 = 𝐹𝐺 + 𝐹𝑍 + 𝐹𝐷 ⎭
[ ]𝑇 ( )
where 𝜁 = 𝑥 𝑦 𝑧 is the absolute location of the quadrotor, and 𝜏𝑓 𝑦 = 𝑙 ⋅ 𝐾𝑇 𝜔24 − 𝜔22 (23)
[ ]𝑇
𝑣 = 𝑣𝑥 𝑣𝑦 𝑣𝑧 is the orbit velocities of the quadrotor. Also, m
(total mass of the quadrotor) and the force applied in the direction of z ( )
are in the reverse direction, with the force of gravity and external forces 𝜏𝑓 𝑧 = 𝐾𝑑 𝜔21 + 𝜔23 − 𝜔22 − 𝜔24 (24)
[ ]𝑇
preventing take off. Here, 𝐹𝐺 = 0 0 −𝑚𝑔 gravity force, 𝐹𝑍 =
( ) [ ]𝑇 Here K𝑑 : drag coefficient, K𝑇 : lift coefficient, 𝜔𝑖 ∶ angular rotor veloc-
𝑅𝐵 𝜙 𝜃 𝜓 ⋅ 0 0 𝑢1 , force of gravity is obtained by being ity, l: the distance between flaps and center of gravity. Aerodynamic
multiplied by rotation matrix to control movement in the direction of friction moments are obtained by multiplying the square of velocity on
z. each orbit by the environment coefficient (Lee et al., 2009).
⎡cos 𝜃 cos 𝜓 sin 𝜃 cos 𝜓 sin 𝜙 − sin 𝜓 cos 𝜙 sin 𝜃 cos 𝜓 cos 𝜙 + sin 𝜓 sin 𝜙⎤
⎢ ⎥ ⎡𝑘𝑓 𝑎1 0 0 ⎤
𝐹𝑍 = ⎢ cos 𝜃 sin 𝜓 sin 𝜃 sin 𝜓 sin 𝜙 + cos 𝜓 cos 𝜙 sin 𝜃 sin 𝜓 cos 𝜙 − cos 𝜓 sin 𝜙⎥ 𝜏𝑎 = ⎢ 0 𝑘𝑓 𝑎2 0 ⎥ 𝛺2 (25)
⎢ − sin 𝜃 ⎥ ⎢ ⎥
⎣ cos 𝜃 sin 𝜙 cos 𝜃 cos 𝜙 ⎦ ⎣ 0 0 𝑘𝑓 𝑎3 ⎦
⎡0⎤
⎢ ⎥ The moments that occur because of gyroscopic effects
× ⎢0⎥ (Aristotelis Kapnopoulos, 2022; Mahony et al., 2012) are obtained with;
⎢𝑢 ⎥
⎣ 1⎦
(13) ⎡ 0 ⎤
∑
4
𝜏𝑔 = 𝛺 × 𝐽𝑟 ⎢ 0 ⎥ (26)
⎢ ⎥
𝑖=1 ⎣(−1)𝑖+1 𝜔𝑖 ⎦
⎡sin 𝜃 cos 𝜓 cos 𝜙 + sin 𝜓 sin 𝜙⎤ ∑4
𝐹𝑍 = ⎢sin 𝜃 sin 𝜓 cos 𝜙 − cos 𝜓 sin 𝜙⎥ 𝑢 (14) J𝑟 : rotor inertia, differential equations that are obtained according to
⎢ ⎥ 𝑖=1 𝑖
⎣ cos 𝜃 cos 𝜙 ⎦ angular accelerations are obtained in the following way;
⎡−𝑘𝑓 𝑡1 0 0 ⎤ ⎡𝑥̇ ⎤ (𝐼 ) 𝜙̇ 2 ⎫
𝑦 −𝐼𝑧 𝑘
𝐹𝐷 = ⎢ 0 0 ⎥ ⎢𝑦̇ ⎥ 𝐽
−𝑘𝑓 𝑡2 (15) 𝜙̈ = 𝜃̇ 𝜓̇ − 𝐼𝑥 𝜃̇ 𝛺𝑑 + 𝐼𝑙 𝑢2 − 𝑓 𝑎1 ⎪
⎢ ⎥ ⎢ ⎥ (
𝐼𝑥
) 𝑥 𝑥 𝐼𝑥
⎪
⎣ 0 0 −𝑘𝑓 𝑡3 ⎦ ⎣𝑧̇ ⎦ 𝑘𝑓 𝑎2 𝜃 ⎪
̇ 2
𝐼𝑧 −𝐼𝑥 𝐽𝑥 ̇ 𝑙
𝜃̈ = 𝜙̇ 𝜓̇ 𝐼𝑦
+ 𝐼𝑦
𝜙 𝛺𝑑 + 𝐼𝑦
𝑢3 − 𝐼𝑦 ⎬
(27)
where F𝐷 : Drag forces on the x, y, z axes. Force state equations is ⎪
obtained by force equation solution as shown in Eq. (16): ( 𝐼 −𝐼 ) 𝑘𝑓 𝑎3 𝜓̇ 2 ⎪
1
𝜓̈ = 𝜙̇ 𝜃̇ 𝑥 𝑦
𝐼𝑧
+ 𝐼𝑧 4
𝑢 − 𝐼𝑧 ⎪
𝑢1 𝑘𝑓 𝑡1 𝑥̇ ⎫ ⎭
𝑥̈ = (cos 𝜙 sin 𝜃 cos 𝜓 + sin 𝜙 sin 𝜓) 𝑚
− 𝑚 ⎪
𝑘𝑓 𝑡2 𝑦̇ ⎪ Here 𝛺𝑑 is called the effect of noise in the state equations and is given
𝑢1
𝑦̈ = (cos 𝜙 sin 𝜃 sin 𝜓 − sin 𝜙 cos 𝜓) − ⎬ (16) by 𝛺𝑑 = 𝜔1 + 𝜔3 − 𝜔2 − 𝜔4 . State-space form is obtained by assigning
𝑚 𝑚
⎪
𝑧̈ = −𝑔 + (cos 𝜙 cos 𝜃)
𝑢1
−
𝑘𝑓 𝑡3 𝑧̇
⎪ states from Eqs. (16) and (27).
𝑚 𝑚 ⎭
⎡ 𝑥 (7) ⎤
Moment of inertia is a symmetric matrix due to the rigid structure of ⎢ ⎥
the quadrotor. Moments of inertia are in the form of I𝑥 , I𝑦 , I𝑧 symmetric ⎢ 𝑥 (8) ⎥
⎢ ⎥
matrix (Bouadi et al., 2007; Cabecinhas et al., 2015; Das et al., 2008;
⎡ 𝑥̇ ⎤ ⎢ ⎥
𝑥 (9)
Guenard et al., 2008; Mahony et al., 2012; Nicol et al., 2011; Tayebi & ⎢ 𝑦̇ ⎥ ⎢⎢ 𝑥 (10)
⎥
⎥
McGilvray, 2006) ⎢ ⎥ ⎢ ⎥
⎢ 𝑧̇ ⎥ ⎢ 𝑥 (11) ⎥
⎡𝐼𝑥 0 0⎤ ⎢ 𝜙̇ ⎥ ⎢ ⎥
⎢ ̇⎥ ⎢ 𝑥 (12)
𝐽 =⎢0 𝐼𝑦 0⎥ (17) ⎢𝜃 ⎥ ⎢ ⎥
⎢ ⎥ 𝑢1 𝑘𝑓 𝑡1 𝑥̇ ⎥
⎣0 0 𝐼𝑧 ⎦ ⎢𝜓̇ ⎥ ⎢(cos 𝜙 sin 𝜃 cos 𝜓 + sin 𝜙 sin 𝜓) 𝑚 − 𝑚 ⎥
𝑥̇ = ⎢ ⎥ = (28)
Moreover, angular velocity is defined with, ⎢ 𝑥̈ ⎥ ⎢⎢ (cos 𝜙 sin 𝜃 sin 𝜓 − sin 𝜙 cos 𝜓) 𝑢1 − 𝑘𝑓 𝑡2 𝑦̇ ⎥⎥
⎢ 𝑦̈ ⎥ ⎢ 𝑚 𝑚
⎥
⎡𝑝⎤ ⎡1 0 − sin 𝜃 ⎤ ⎡ 𝜙̇ ⎤ ⎢ 𝑧̈ ⎥ ⎢ −𝑔 + (cos 𝜙 cos 𝜃)
𝑢1
−
𝑘𝑓 𝑡3 𝑧̇
⎥
⎢ ⎥ ⎢
𝛺 = ⎢𝑞 ⎥ = ⎢0 cos 𝜃 sin 𝜙 ⎥ ⎢ 𝜃̇ ⎥ 𝑚 𝑚
cos 𝜙 (18) ̈
⎢𝜙⎥ ⎢ ( 𝐼 −𝐼 ) ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥ 𝐽𝑥 ̇ 𝑙 𝑘 𝜙 ̇ 2
⎥
⎢ 𝜃̈ ⎥ ⎢ 𝜃̇ 𝜓̇
𝑦 𝑧 𝑓 𝑎1
⎣ 𝑟 ⎦ ⎣0 − sin 𝜙 cos 𝜙 cos 𝜃 ⎦ ⎣𝜓̇ ⎦ 𝐼
− 𝐼
𝜃 𝛺 𝑑 + 𝐼𝑥
𝑢 2 − 𝐼𝑥 ⎥
⎢ ⎥ ⎢ ( 𝑥 ) 𝑥 ⎥
Moment equations are shown with the Newton–Euler equity (Bruno ⎣𝜓̈ ⎦ 𝑘𝑓 𝑎2 𝜃̇ 2
⎢𝜙̇ 𝜓̇ 𝐼𝑧 −𝐼𝑥
+ 𝐼 𝜙 𝛺𝑑 + 𝐼 𝑢3 − 𝐼 ⎥
𝐽𝑥 ̇ 𝑙
et al., 2009; Lee, 2013). ⎢ 𝐼𝑦 𝑦 𝑦 𝑦 ⎥
⎢ ( 𝐼 −𝐼 ) 𝑘𝑓 𝑎3 𝜓̇ 2
⎥
𝑅̇ = 𝑅 ⋅ 𝑆 (𝛺) (19) ⎢ 𝜙̇ 𝜃̇ 𝑥 𝑦 1
+ 𝐼 𝑢4 − 𝐼 ⎥
⎣ 𝐼𝑧 𝑧 𝑧 ⎦

5
A. Gün Expert Systems With Applications 229 (2023) 120518

3. Position control 4.1. Mutation

There is a desire to control the parameters 𝑧, 𝜙, 𝜃, 𝜓 for the quadro- The mutation operation is carried out among at least 3 individuals.
tor (Islam et al., 2015; Lee et al., 2009). The difference between 2 individuals from 3 who are randomly selected
[ ]𝑇 [ ]𝑇 from the pool is multiplied by the mutation factor (F ), and it is added
𝑒 = 𝑧𝑑 𝜙𝑑 𝜃𝑑 𝜓𝑑 − 𝑧 𝜙 𝜃 𝜓 (29)
to the third individual. This operation is carried out with the help of
For each parameter, the error vector is obtained by subtracting quadro- the equation given next:
tor location and angles from the aimed location and angles. The mo- ( )
𝑉𝑖,𝐺+1 = 𝑋𝑎,𝐺 + 𝐹 ∗ 𝑋𝑏,𝐺 − 𝑋𝑐,𝐺 (38)
ments that will be applied to the motors are achieved by multiplying
this error vector by different coefficients. Proportional control;
[ ][ ]𝑇 4.2. Crossover
𝑃𝑒 = 𝑘𝑧𝑝 𝑘𝜙𝑝 𝑘𝜃𝑝 𝑘𝜓𝑝 𝑒𝑧 𝑒𝜙 𝑒𝜃 𝑒𝜓 (30)
The individual who is obtained as a result of the mutation undergoes
Integral control;
the process of crossover with the existing individual. If a random
⎡ 𝑒𝑧 ⎤ value produced for each parameter that belongs to the individual in
[ ] ⎢ ⎥
𝑒 the crossover is bigger than the predetermined crossover (CR) ratio,
𝐼𝑒 = 𝐼𝑒 + 𝑘𝑧𝑖 𝑘𝜙𝑖 𝑘𝜃𝑖 𝑘𝜓𝑖 𝑇𝑠 ⎢ 𝜙 ⎥ (31)
⎢ 𝑒𝜃 ⎥ the parameter is obtained from the candidate individual. If not, the
⎢𝑒 ⎥ parameter is obtained from the existing individual. At the end of the
⎣ 𝜓⎦
crossover, the candidate individual for the new generation will be
and derivative control;
produced. This process is shown in the following equation.
⎡0 − 𝑧̇ ⎤ { }
] ⎢0 − 𝜙̇ ⎥
𝑣𝑖,𝐺+1 𝑖𝑓 𝑟𝑗 ≤ 𝐶𝑅
[ 𝑢𝑖,𝐺+1 = (39)
𝐷𝑒 = 𝑘𝑧𝑑 𝑘𝜙𝑑 𝑘𝜃𝑑 𝑘𝜓𝑑 ⎢ ⎥ (32) 𝑥𝑖,𝐺 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
⎢0 − 𝜃̇ ⎥
⎢0 − 𝜓̇ ⎥ Here, 𝑢𝑖,𝐺+1 represents the candidate individual produced for the next
⎣ ⎦
generation and r𝑗 stands for random number.
Total forces give PID control:
⎡ 𝑢1 ⎤ 4.3. Selection
⎢ ⎥
𝑢
𝑢 = ⎢ 2 ⎥ = 𝑃𝑒 + 𝐼𝑒 + 𝐷𝑒 (33)
⎢ 𝑢3 ⎥ The candidate individual produced at the end of the crossover is
⎢𝑢 ⎥ compared with the existing candidate on the basis of the convenience
⎣ 4⎦
function. The one who has the highest convenience is transformed into
The weight of aircraft and rotation angles are added to u1 force in the
the next generation.
direction of z. ( ) ( )}
{
𝑚g 𝑢 𝑈𝑖,𝐺 𝑖𝑓 𝑓 𝑈𝑖,𝐺 ≤ 𝑓 𝑋𝑖,𝐺
𝑢1 = + (34) 𝑋𝑖,𝐺+1 = (40)
(cos 𝜙 cos 𝜃) (cos 𝜙 cos 𝜃) 𝑋𝑖,𝐺 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
The aim of this is to remove the loss of power due to gravity accel- ( )
Here, 𝑓 𝑢𝑖,𝐺+1 stands for the convenience value of the candidate
eration and angular moments (Mian & Daobo, 2008). In this study, in ( )
individual and 𝑓 𝑥𝑖,𝐺 represents the convenience value of the existing
the beginning, coefficients are approximately determined with the trial- individual.
and-error method. The differential evolution algorithm is applied by
taking the maximum values of used coefficients and optimal coefficients 4.4. Stopping
are found;
[ ] The earlier calculation steps are repeated as the number of pre-
𝑘𝑧𝑝 𝑘𝜙𝑝 𝑘𝜃𝑝 𝑘𝜓𝑝 𝑘𝑧𝑖 𝑘𝜙𝑖 𝑘𝜃𝑖 𝑘𝜓𝑖 𝑘𝑧𝑑 𝑘𝜙𝑑 𝑘𝜃𝑑 𝑘𝜓𝑑
determined iterations. When the iteration number is completed, the
(35) calculation is stopped. At this point, the individual who has the highest
convenience until that point of time is accepted as the solution.
4. Differential evolution algorithm
5. DE based PID controller design
Differential evolution algorithm is a universal algorithm that was
proposed by Storn and Price (1995). It is widely used in various areas Certain studies in the literature related with the differential evolu-
since it has a simple and strong structure (Cheng & Hwang, 2001; tion algorithm are used (Sarker et al., 2014). In one of these studies,
Sarker et al., 2014; Yu et al., 2014). It has a process of calculation ANFN-MODE is proposed for the nonlinear control system (Chen et al.,
directed to new candidate solutions through calculation according to 2009). The 3 PID parameter for six free-graded electrical hydraulic
crossover, mutation and selection principles (Triguero et al., 2011; Yu structure is estimated by the help of the differential evolution algo-
et al., 2014). The steps of DE calculation are provided below: Creating rithm (Luo & Che, 2010). In another study, PID control coefficients are
starting pool Starting pool is created by randomly producing possible adjusted by using the multi-objective differential evolution algorithm.
solutions called individual in the range and number of determined Coefficients obtained through the Ziegler–Nichols method which is a
parameter. Initial population is created by randomly producing possible classical method are minimized (Chiha et al., 2012). PID coefficients
solutions called individual in the range and number of determined are obtained through the differential evolution algorithm for velocity
parameters. control of the permanent magnet linear (PMLM) motor. DE is com-
{ } pared with static and dynamic performance of the traditional PID
𝑋𝑖,𝐺 = 𝑋𝑖,𝐺 1 , 𝑋𝑖,𝐺 2 , … .𝑋𝑖,𝐺 𝑛 , (𝑖 = 1, 2....𝐵𝑆) (36)
controller (Lin et al., 2012). The differential evolution algorithm is
𝑛 𝑛 mathematically provided through dynamic parameter selection for opti-
[𝑋min , 𝑋max ] (37)
mization problems (Sarker et al., 2014). The systematic design method
Here, X𝑖,𝐺 , G, stands for the i. individual in the generation, BS stands for the output voltage regulator of DC-DC boost type convertor and the
for the number of the individual, 𝑋min 𝑛 n. is the minimum value of the studied design method are clearly defined with the genetic algorithm,
𝑛
parameter and 𝑋max n. stands for the maximum value of the parameter. differential evolution, and artificial immune systems (Sundareswaran

6
A. Gün Expert Systems With Applications 229 (2023) 120518

et al., 2014). In Paez et al. (2021) study, the terrain that is complex Table 2
Quadrotor unmanned aircraft PID coefficient at the beginning.
and dangerous for rescuers to pass in the event of a disaster has
been considered, and the problem of finding victims in a search and z position x degree y degree z degree

rescue environment has been discussed. Also, time is of the essence P 2 1.5 1 1
I 1e−3 1e−3 1e−3 1e−3
when saving lives after a disaster, given all these challenges, a solution
D 2 0.2 0.15 0.5
has been proposed using a collaborative robot team that speeds up
the process of searching for survivors and avoids risking additional
lives. This article has focused on navigating a swarm of robots that Table 3
Randomly generated pool of 15 x 12 size.
can avoid crashing into static or dynamic obstacles and locate the
place of victims. The method used to navigate the environment is 1 2 3 4 5 6 7 8 9 10 11 12 Y

based on a DPSO Distributed Particle Swarm Optimization, where each P𝑧 P𝜙 P𝜃 P𝜓 I𝑧 I𝜙 I𝜃 I𝜓 D𝑧 D𝜙 D𝜃 D𝜓

1
particle swarm represents a single robot. The interaction between the
⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮
swarms has been demonstrated and the artificial potential function 15
has been used to screen the survivors by avoiding collision. Various
simulation experiments have been conducted to test the navigation
Table 4
algorithm, avoid obstacles and find disaster victims, and the results
Randomly selected three rows.
of the experiments have been found to be successful. Cardenas et al.
1 2 3 4 5 6 7 8 9 10 11 12 Y
(2022) In this paper, the PID control tuning problem for a quadrotor
P𝑧 P𝜙 P𝜃 P𝜓 I𝑧 I𝜙 I𝜃 I𝜓 D𝑧 D𝜙 D𝜃 D𝜓
has been discussed. This task is vital given the variety of applications
1 (M1𝑛 )
a UAV quadrotor can perform. This article focuses on optimum PID ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮
controller setting in ø (pfi angular) axes. It has been proposed to i
use the Multi-Purpose Particle Swarm Optimization (PSO) algorithm ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮
for PID performance improvement. The proposed system has been j
⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮
evaluated in a Parrot Mambo FPV UAV Quadrotor virtual model using
k
MATLAB and Simulink environment. Four different cost functions have ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮
been compared for the multi-objective optimization approach. These 15 ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮
cost functions have focused on improving UAV performance based on
settlement time, maximum overshoot, steady-state error, and control
Table 5
effort. The rating(evaluation) system showed the best performance in Random coefficient and objective function for quadrotor unmanned aircraft.
combining different factors through the multi-objective PSO algorithm. 1 2 3 4 5 6 7 8 9 10 11 12
The study of Zhao et al. (2022), a new bio-inspired optimization
x𝑟×𝑛 x𝑟×1 x𝑟×2 x𝑟×3 x𝑟×4 x𝑟×5 x𝑟×6 x𝑟×7 x𝑟×8 x𝑟×9 x𝑟×10 x𝑟×11 x𝑟×𝑛
algorithm called artificial hummingbird algorithm (AHA) has been ⇕ ⇕ ⇕ ⇕ ⇕ ⇕ ⇕ ⇕ ⇕ ⇕ ⇕ ⇕ ⇕
proposed to solve optimization problems. The AHA algorithm simulates x1×𝑛 x1×1 x1×2 x1×3 x1×4 x1×5 x1×6 x1×7 x1×8 x1×9 x1×10 x1×11 x1×12
specialized flight skills and intelligent foraging strategies of humming-
birds. Three types of flight skills, including axial, transversal, and
omnidirectional flights used in foraging strategies, have been modeled
in nature. Comparisons show that AHA is more competitive than other Since the smallest value is zero, scaling is carried out by multiplying
meta-heuristic algorithms and identifies high-quality solutions with by the biggest values. These values are randomly created values of PID
fewer control parameters. Additionally, AHA’s performance has been coefficients. For each individual in the pool, PID is started to work and
validated on ten challenging engineering design case studies. When the Y cost function is calculated according to results (Sarker et al., 2014).
results have been compared to the Genetic algorithm (GA), one of the
existing optimization techniques in the literature, and Particle swarm Initially, the algorithm given later is applied by 𝑀1𝑛 and is held
optimization (PSO) as a bio-inspired algorithm, it has been stated that steady. As shown in Table 4 three rows such as 𝑀𝑖𝑛 , 𝑀𝑗𝑛 , and 𝑀𝑘𝑛
the superior efficiency of AHA in terms of computational load and solu- except for 𝑀1𝑛 are randomly selected from the 𝑀 × 𝑁 matrix. 𝑥1 𝑛 =
tion precision was demonstrated, and successful results have also been ( )
𝐹 ⋅ 𝑀𝑖𝑛 − 𝑀𝑗𝑛 + 𝑀𝑘𝑛 is calculated for each cell. For n = 1:1:12 PID
obtained from its application in the hydroelectric operation design. In lines, newly obtained values are written in the form of the vector. A
this study, 12 PID coefficients affect the moments that the quadrotor ( )
new vector is formed with 𝑥1 𝑛 = 𝐹 ⋅ 𝑀𝑖𝑛 − 𝑀𝑗𝑛 + 𝑀𝑘𝑛 equation.
produces, so those coefficients that will provide the optimal study Evaluation will be made according to the 1st line of the matrix. Through
should be obtained. Therefore, the differential evolution algorithm was this process, a vector with 12 × 1 is created and this vector is assigned
used in the concerned study. The DE-based PID controller design is as x1𝑛 . (F = 0.5, CR = 0.9 parameters related to differential evolution
executed for the quadrotor with 4 rotors. First of all, moment values are selected) 1 ≤ 𝑛 ≤ 12, newly obtained vector as n is an integer
that will be applied to the motors are calculated through the classic PID number. A random value in the shape of 𝑥𝑟𝑎𝑛 × 𝑛 is selected in the range
controller. After that, the concerned parameters are updated through of 0 to 1 for the value in each cell of this vector in Table 5. We select
the DE algorithm and applied to the system. The block structure of the the value of 0.9 for CR.
created controller is given in Fig. 3. The steps of the applied evolution
algorithm are stated next. The range of PID coefficients determine with
trial-and-error or PID calculation methods, as in Table 2. Minimums If 𝑥𝑟𝑎𝑛 × 𝑛 > 𝐶𝑅, the cell information in M1𝑛 is assigned to x1𝑛 𝑥1𝑛 ←
are accepted as zero (Sarker et al., 2014). Here, PID coefficients have 𝑀1𝑛 If 𝑥𝑟𝑎𝑛 × 𝑛 < 𝐶𝑅, the cell information in x1𝑛 does not change.
the maximum values in the range of 1e−3 and 2. A pool in the size 𝑥1𝑛 ← 𝑥1𝑛 With this process, some of the values in x1𝑛 change and some
of random X = rand(M,N), M = 15 and N = 12 is created as shown of them remain the same. The PID controller is activated by using the
in Table 3. The values in this pool should be in the range of 0 to 1. coefficients in the obtained x1𝑛 vector, and the result created according
After the creation of the pool (starting value of Y -outputs is zero) Y to the x1𝑛 vector becomes Y1𝑛𝑒𝑤 . If 𝑌1𝑛𝑒𝑤 < 𝑌1𝑛 , 𝑀1𝑛 is erased and x1𝑛
cost function is calculated (the values obtained from state equations and Y1𝑛𝑒𝑤 vectors are written instead of it. If 𝑌1𝑛𝑒𝑤 > 𝑌1𝑛 , 𝑀1𝑛 remains
are subtracted from desired values). Each cell in the size of 15 × 12 is the same and the obtained individual is not good. This process is carried
rescaled according to their minimum and maximum values in the table. out until the last line of the 𝑀 ×𝑁 matrix (1 Iteration). The differential
evolution process with a flowchart is given in Fig. 4.

7
A. Gün Expert Systems With Applications 229 (2023) 120518

Fig. 3. Structure of the PID controller based on DE.

6. Multi objective optimization and results Table 6

Simulation and cost function values.
Definitions DE-PID PSO-PID GSA-PID CSS-PID
Each output value is individually calculated through the objective
functions given later. Here, calculated output values are the equations Number of epoch 25 25 25 25
Number of particles 15 15 15 15
that belong to the objective function. These equations are used for the
Number of variables 12 12 12 12
minimization of outputs. As; Minimization values 0.4164 0.4004 0.4537 0.4575
𝑇 𝑜𝑡𝑎𝑙_𝑇 𝑖𝑚𝑒 Simulation time-sec 890.68 916.35 883.46 840.274
𝑁= (41)
𝑆𝑡𝑒𝑝_𝑇 𝑖𝑚𝑒
√
∑𝑁 ( )2
𝑛=0 𝑧𝑑 − 𝑧 (𝑛) presented in Table 7. While the values of the objective functions are
𝑦1 = (42)
𝑁 +1 0.4164, 0.4004, 0.4537, 0.4575 with DE, PSO, GSA and CSS coefficients
√
∑𝑁 ( )2 respectively, the value of the DE based PID controller is 0.4164 as
𝑛=0 𝜑𝑑 − 𝜑 (𝑛) shown in Table 6. Figs. 7, 9, 11, 13, 15 show the change of z positions,
𝑦2 = (43)
𝑁 +1 and x, y, z angle values of the quadrotor for different PID variations.
√
∑𝑁 ( )2
𝑛=0 𝜃𝑑 − 𝜃 (𝑛) 7. Results
𝑦3 = (44)
𝑁 +1
√ For the evaluation of DE PID performance against classical PID and
∑𝑁 ( )2
𝑛=0 𝜓𝑑 − 𝜓 (𝑛) other variations various scenarios were applied and are as explained
𝑦4 = (45)
𝑁 +1 follows.
Total objective function;
7.1. Scenario 1
𝑦 = 0.01 ⋅ 𝑦1 + 𝑦2 + 𝑦3 + 𝑦4 (46)
The PID coefficients given in Table 7 were applied for the change
is given with 𝑦 = 0.01 ⋅ 𝑦1 + 𝑦2 + 𝑦3 + 𝑦4 . In this equation, the
of z-position and angular displacement forces. Results for changes of
effect of y1 is diminished by being multiplied by 0.01. For location
1 rad in the z direction were examined as shown in Figs. 7a to e,
control of the quadrotor, control inputs of the system for four rotors
9a to e, 11a to e, 13a to e and 15a to e. it was seen that the rise
are according to the angular velocities. The addition of equations from
time and maximum overshoot were shorter for DE PID compared to
(1) and (4) and the later velocity equations are provided by subtracting
other variations. Fig. 26a to e show each propellers’ velocities which
from each other. By taking their squares, velocity variables can be
is applied to the rotors. Forces of each rotor were given in Figs. 6, 8,
obtained. The velocity equations given in Eq. (47) are obtained from 1-
10, 12, 14. As applied moments for different algorithms are shown in
2-3-4 equations (Mian & Daobo, 2008). Simulation Results: Simulation
Fig. 27, it can be seen that DE PID provides a moment approximately
results are the obtained real physical parameters that are used in the
gain of 9.7%
source (Derafa et al., 2006). These parameters are shown in Table 8.
First of all, classical PID coefficients are found for the desired direction
7.2. Scenario 2
of the quadrotor with the trial-and-error method in Table 2.
√
𝑢1 +2𝑢3 𝑢 In this scenario when the quadrotor is moving in the z direction
𝜔1 = 4⋅𝐾
− 4⋅𝐾4
𝑇 𝑄 random positive and negative disturbances were applied to the U2
√
𝑢1 +2𝑢2 𝑢4 moment between the 4th and 5th seconds changing the x(𝜙) angle after
𝜔2 = 4⋅𝐾𝑇
− 4⋅𝐾𝑄 disturbance is applied to moment U3 between 7th and 10th seconds
√ (47) changing the y(𝜃) angle and then the attitude of the quadrotor is ex-
𝑢1 −2⋅𝑢3 𝑢4
𝜔3 = 4⋅𝐾𝑇
− 4⋅𝐾𝑄 amined. The reason why the forces are applied randomly is to examine
√ the swaying behavior of the airborne quadrotor according to wind, rain
𝑢1 +2⋅𝑢2 𝑢4
𝜔4 = + and other factors. Figs. 16–18–20–22–24 are shown propellers forces
4⋅𝐾𝑇 4⋅𝐾𝑄
for PID, DE PID, PSO PID, GSA PID and CSS PID respectively. Desired z-
Minimized cost functions with DE, PSO, GSA, and CSS are shown position and angular values are given from Figs. 17–19–21–23 and 25.
in Fig. 5; Cost function results are given in Table 6 and moments, for classical PID, DE PID, PSO PID, GSA PID and CSS PID respectively.
propellers’ max velocities are depicted in Table 9. The multi-objective Angular velocities and Moments of the PID, DE PID, PSO PID, GSA PID
function defined in Eq. (46) is used as a cost function in four different and CSS PID are given Figs. 28 and 29.
optimization based PID controller designs. The minimization of this In the first 4 s, conventional PID coefficients given in Table 7
objective function, which is defined as the sum of 4 error functions were applied for the vehicle to reach the [50, 1, 1, 1] position and
for each optimization, is implemented by the DE, PSO, GSA, and angular values. A disturbance acting on the 𝜙 angle was applied to the
CSS algorithms. The most suitable coefficients which are obtained are quadrotor between 4th and 5th seconds. It is shown in Fig. 16 a that

8
A. Gün Expert Systems With Applications 229 (2023) 120518

Fig. 4. Flowchart of the DE.

Table 7
Classical PID and optimized new DE PID, PSO PID, GSA PID, and CSS PID coefficients of the quadrotor control.
k𝑧𝑝 k𝜙𝑝 k𝜃𝑝 k𝜓𝑝 k𝑧𝑖 k𝜙𝑖 k𝜃𝑖 k𝜓𝑖 k𝑧𝑑 k𝜙𝑑 k𝜃𝑑 k𝜓𝑑
PID 2 1.5 1 1 1e−3 1e−3 1e−3 1e−3 2 0.2 0.15 0.5
DE-PID 1.9872 1.4172 0.428 0.984 1.79e−4 10e−4 0 8.4e−4 1.6545 0.051 0.145 0.0818
PSO-PID 2 1.5 1 1 0 6.57e−4 1.17e−4 0 1.392 0.156 0.125 0.093
GSA-PID 1.889 1.447 0.778 1 8.07e−4 1e−3 9.15e−4 8.24e−4 1.719 0.159 0.145 0.373
CSS-PID 2 1.423 0.849 0.993 9.13e−4 9.62e−4 9.14e−4 9.8e−4 1.985 0.164 0.146 0.4

the disturbance values affect the U2 force in the z direction between and then it was observed that the quadrotor reached its target again as
4th and 5th seconds. It is seen that the F1 and F3 motor forces, as well shown in Fig. 16a to e.
as the F2 and F4 motor forces, produce similar torques. It is known
that the change in power consumption is directly proportional with the 7.3. Scenario 3
change of motor torque values. Between the 7th and 10th seconds, as
shown in Fig. 16c, a disturbance acting on the 𝜃 value was applied, Conventional PID coefficients given in Table 7 were applied to reach
the [50, 0, 0, 0] position and angular values of the vehicle. Although

9
A. Gün Expert Systems With Applications 229 (2023) 120518

Fig. 5. Minimized cost functions.

Table 8 that there was an effective change of force in motors 2 and 4, and less
Physical parameters of the quadrotor.
change in motors 1 and 3 as shown in Figs. 28 and 29.
Parameters Value Unit
M𝑞 0.4 kg
L𝑞 0.205 m
[ ]𝑇 7.5. Efficiency
J𝑞 3.8278 3.8278 7.1345 ⋅ 10−3 N m/rad/s2
[ ]𝑇
K𝑓 𝑎𝑞 5.567 5.567 6.354 ⋅ 10−4 N/rad/s
[ ]𝑇
K𝑓 𝑡𝑞 0.032 0.032 0.048 N/m/s
The coefficients obtained from each optimization algorithm were
J𝑟𝑞 2.83 . 10−5 N m/rad/s2
G𝑞 9.86 m/s2 run in the same simulation and the results were compared. It is seen
K𝑝𝑞 2.98 . 10−5 N m/rad/s that DE PID gives better results without disturbances for this study. The
K𝑑𝑞 3.23 . 10−7 N m/rad/s instantaneous maximum moment values of the quadrotor in motion
are effective in energy consumption. When the graphs of the results
Table 9 obtained were compared, a 9.7% lower moment value was observed.
Moments and velocity values for scenario-1. For example, considering that the quadrotor makes an instantaneous
PID DE-PID PSO-PID GSA-PID CSS-PID change every 30 s in a 30-min mission, according to Fig. 27a, and b, a
Max. moment N m 6.033 5.444 8.811 8.42 6.09 moment gain of 0.59 Nm will be achieved for each movement change,
Propellers velocity (rad/s) 694 695 838 759 711 and 35.4 Nm for total operating time. When the figures from Fig. 10c
to e are compared, the obtained moment gain will be slightly higher.
The quadrotor was stabilized after the 5th second, and a random
disturbance for U3 was applied between the 7th and 10th seconds with
it may seem like a simple scenario, it is a determinative application
the expression Uq(3) = Uq(3) + 2 ∗ (2 ∗ rand − 1). In this case, it was
for evaluating energy efficiency. Figs. 30 and 31 show the speeds and
observed that the y(𝜃) angle of the quadrotor changed. In order for the
torques for the instantaneous energy consumption of quadrotor motors.
quadrotor to reach the desired target, it is seen that the motors 1 and
DE PID needs 2.84606 N m of torque. The nearest Moment value is
3 have more frequency than the motors 2 and 4. Quadrotor reaches its
2.86181 N m. Efficiency is obtained 0.5% in this scenario against other
target after 10th second and the position change graphs are presented
algorithms.
in Fig. 29a to e.

7.4. Failure modes

8. Conclusion

The failure modes of the quadrotors differ for the desired target
situations. However, optimal thrust force should be in the range of In this study, the control of the quadrotor with a controller having
1.5 to 2 times the quadrotor weight. Stabilization cannot be achieved 12 different PID coefficients has been carried out with well-known
with thrust force applied more than 2 times the weight as maximum optimization methods such as DE, PSO, GSA, CSS. It has been tested
overshoot will be too high. In this study, suitable PID coefficients were on a sample problem whose angular values in the z-direction are 1 rad
obtained by using the parameters taken from paper (Derafa et al., each. In this study, which has been carried out to use the energy of
2006) Since the quadrotor is assumed to weigh 0.4 kg in the study, the the quadrotor effectively, it has been seen that DE gave better results.
threshold value required for take-off is predicted to be 4 N. In this study Although the cost function of DE was high in the simulation, the
total force is obtained 6.715 N at the steady state condition in Fig. 6e. torque values have been examined according to the PID values obtained
Reasonable ratio is found 1.67. The total moment values required to from each optimization and it has been that DE gave better results in
raise the quadrotor to a height of 50 m were obtained by multiplying terms of energy efficiency. The obtained PID coefficients have been run
the forces applied by each motor with the radii of the propellers. In (simulated) on the same control program, and the results have been
order to stabilize the quadrotor. The condition that the motor velocities transferred to the graphical environment. Moment graphics show that
are not limited can be considered as a failure mode. It was observed DE-PID reaches the target with less energy than other methods (current
that the angle of x𝜙 changed when random disturbances were given to - conventional PID, PSO-PID, GSA-PID, CSS-PID). In future studies, it
the moving quadrotor with the expression Uq(2) = Uq + 2 ∗ (2 ∗ rand is possible to try to increase the energy efficiency by obtaining PID
− 1) between the 4th and 5th seconds. In this case, it was observed coefficients in more optimal ways and using different algorithms.

10
A. Gün Expert Systems With Applications 229 (2023) 120518

Fig. 7. Scenario 1 : z position and desired angular position for traditional PID.

Fig. 6. Scenario 1 : Rotor forces and total force for the traditional PID of the quadrotor
attitude.

11
A. Gün Expert Systems With Applications 229 (2023) 120518

Fig. 8. Rotor forces and total force for the DE PID of the quadrotor attitude. Fig. 9. Scenario 1 : z position and desired angular position for DE PID of the quadrotor
attitude.

12
A. Gün Expert Systems With Applications 229 (2023) 120518

Fig. 10. Scenario 1 : Rotor forces and total force for the PSO PID of the quadrotor Fig. 11. Scenario 1 : z position and desired angular position graphs for PSO PID.
attitude.

13
A. Gün Expert Systems With Applications 229 (2023) 120518

Fig. 12. Scenario 1 : Rotor forces and total force for the GSA PID of the quadrotor Fig. 13. Scenario 1 : The position and angular graphs for the GSA PID of the quadrotor
attitude. attitude.

14
A. Gün Expert Systems With Applications 229 (2023) 120518

Fig. 14. Scenario 1: Rotor forces and total force for the CSS PID of the quadrotor Fig. 15. Scenario 1: The position and angular graphs for the CSS PID of the quadrotor
attitude. attitude.

15
A. Gün Expert Systems With Applications 229 (2023) 120518

Fig. 16. Scenario 2: Forces and total force for the classical PID of the quadrotor Fig. 17. Scenario 2: The position and angular graphs for the classical PID of the
attitude. quadrotor attitude.

16
A. Gün Expert Systems With Applications 229 (2023) 120518

Fig. 19. Scenario 2: The position and angular graphs for the DE PID of the quadrotor
attitude.
Fig. 18. Scenario 2: Forces and total force for the DE PID of the quadrotor attitude.

17
A. Gün Expert Systems With Applications 229 (2023) 120518

Fig. 20. Scenario 2: Forces and total force graphs for the PSO PID of the quadrotor Fig. 21. Scenario 2: The position and angular graphs for the PSO PID of the quadrotor
attitude. attitude.

18
A. Gün Expert Systems With Applications 229 (2023) 120518

Fig. 22. Scenario 2: The forces and total force for the GSA PID of the quadrotor Fig. 23. Scenario 2: The position and angular graphs for the GSA PID of the quadrotor
attitude. attitude.

19
A. Gün Expert Systems With Applications 229 (2023) 120518

Fig. 24. Scenario 2: The forces and total force for the CSS PID of the quadrotor Fig. 25. Scenario 2: The position and angular graphs for the CSS PID of the quadrotor
attitude. attitude.

20
A. Gün Expert Systems With Applications 229 (2023) 120518

Fig. 26. Scenario 1: Velocities graphs without disturbance of the quadrotor attitude. Fig. 27. Scenario 1: Moments graphs without disturbance of the quadrotor attitude.

21
A. Gün Expert Systems With Applications 229 (2023) 120518

Fig. 28. Scenario 2: Moments graphs with disturbance of the quadrotor attitude. Fig. 29. Scenario 2: Moments graphs with disturbance of the quadrotor attitude.

22
A. Gün Expert Systems With Applications 229 (2023) 120518

Fig. 30. Scenario 3: Velocities graphs for vertical movements of the quadrotor attitude. Fig. 31. Scenario 3: Moments graphs for vertical movements of the quadrotor attitude.

23
A. Gün Expert Systems With Applications 229 (2023) 120518

CRediT authorship contribution statement Guenard, N., Hamel, T., & Mahony, R. (2008). A practical visual servo control for
an unmanned aerial vehicle. IEEE Transactions on Robotics, 24(2), 331–340. http:
//dx.doi.org/10.1109/TRO.2008.916666.
Ayhan Gün: Conceptualization, Methodology, Software, Data cura-
Guerrero-Castellanos, J., Marchand, N., Hably, A., Lesecq, S., & Delamare, J. (2011).
tion, Writing – original draft, Visualization, Investigation, Supervision, Bounded attitude control of rigid bodies: Real-time experimentation to a quadrotor
Software, Validation, Writing – review & editing. mini-helicopter. Control Engineering Practice, 19(8), 790–797. http://dx.doi.org/
10.1016/j.conengprac.2011.04.004, URL https://www.sciencedirect.com/science/
Declaration of competing interest article/pii/S0967066111000700.
Harb, H. M., & Desuky, A. S. (2014). Selection on classification of medical datasets
based on particle swarm optimization. International Journal of Computer Applications,
The authors declare that they have no known competing finan-
104(5), 14–17.
cial interests or personal relationships that could have appeared to Islam, S., Liu, P. X., & El Saddik, A. (2015). Robust control of four-rotor un-
influence the work reported in this paper. manned aerial vehicle with disturbance uncertainty. IEEE Transactions on Industrial
Electronics, 62(3), 1563–1571. http://dx.doi.org/10.1109/TIE.2014.2365441.
Data availability Israel Pereira Souza, R. E. N. M. (2023). A robust algorithm based on differential
evolution with local search for the capacitated vehicle routing problem. Swarm
and Evolutionary Computation, 77, 1–16.
Data will be made available on request.
Kanagaraj, G., Ponnambalam, S. G., & Loo, C. K. (2015). Proceedings of the 2015
international conference on advanced mechatronic systems. In Charged system search
References algorithm for robotic drill path optimization, vol. 1 (pp. 22–24). Beijing, China.
Karegowda, A. G., Jayaram, M. A., & Manjunath, A. S. (2011). Feature subset selection
Alexis, K., Nikolakopoulos, G., & Tzes, A. (2012). Model predictive quadrotor con- using cascaded GA and CFS: A filter approach in supervised learning. International
trol: attitude, altitude and position experimental studies. IET Control Theory & Journal on Computer Applications (IJCA), 23(2), 1–10.
Applications, 6(12), 1812–1827. Kendoul, F., Lara, D., Coichot, I. F., & Lozano, R. (2007). Real-time nonlinear embedded
Amir Ramezani Dooraki, & Lee, D.-J. (2020). An innovative bio-inspired flight controller control for an autonomous quadrotor helicopter. Journal of Guidance, Control, and
for quad-rotor drones: Quad-rotor drone learning to fly using reinforcement Dynamics, 30, 1049–1061.
learning. Robotics and Autonomous Systems, 135, 1–22. Krohling, R., Jaschek, H., & Rey, J. (1997). Designing PI/PID controllers for a motion
Arandhakar, S., Chaudhary, N., Depuru, S. R., Dubey, R. K., & Bhukya, M. N. (2022). control system based on genetic algorithms. In Proceedings of 12th IEEE international
Analysis and implementation of robust metaheuristic algorithm to extract essential symposium on intelligent control (pp. 125–130). http://dx.doi.org/10.1109/ISIC.1997.
parameters of solar cell. IEEE Access, 10, 40079–40092. 626429.
Aristotelis Kapnopoulos, A. (2022). A cooperative particle swarm optimization approach Kuipers, J. B. (1999). Quaternions and rotation sequences. Princeton: Princeton University
for tuning an MPC-based quadrotor trajectory tracking scheme. Aerospace Science Press, http://dx.doi.org/10.1515/9780691211701.
and Technology, 127, 1270–9638.
Lee, T. (2013). Robust adaptive attitude tracking on SO(3) with an application to a
Bergamasco, M., & Lovera, M. (2014). Identification of linear models for the dynamics
quadrotor UAV. IEEE Transactions on Control Systems Technology, 21(5), 1924–1930.
of a hovering quadrotor. IEEE Transactions on Control Systems Technology, 22(5),
http://dx.doi.org/10.1109/TCST.2012.2209887.
1696–1707. http://dx.doi.org/10.1109/TCST.2014.2299555.
Lee, D., Kim, J., & Sastry, H. S. (2009). Feedback linearization vs. adaptive sliding mode
Bouadi, H., Bouchoucha, M., & Tadjine, M. (2007). Sliding mode control based on
control for a quadrotor helicopter. International Journal of Control, Automation and
backstepping approach for an uav type-quadrotor. World Academy of Science,
Systems, 7(3), 419–428. http://dx.doi.org/10.1007/s12555-009-0311-8.
Engineering and Technology, International Journal of Mechanical, Aerospace, Industrial,
Lin, J., Huang, J., & Chen, G. (2012). Design of DE-tuning PID controller for linear
Mechatronic and Manufacturing Engineering, 1(2), 60–65.
Bruno, S., Lorenzo, S., Luigi, V., & Oriolo, G. (2009). Robotics modelling, planning and servo system. In 2012 IEEE fifth international conference on advanced computational
control (1st ed.). Springer. intelligence (ICACI) (pp. 437–440). http://dx.doi.org/10.1109/ICACI.2012.6463201.
Cabecinhas, D., Cunha, R., & Silvestre, C. (2015). A globally stabilizing path following Liu, H., Bai, Y., Lu, G., & Zhong, Y. (2013). Brief paper robust attitude control of
controller for rotorcraft with wind disturbance rejection. IEEE Transactions on uncertain quadrotors. IET Control Theory & Applications, 7(11), 583–1589.
Control Systems Technology, 23, 708–714. Luo, Y., & Che, X. (2010). Tuning PID control parameters on hydraulic servo control
Cardenas, J. A., Carrero, U. E., Camacho, E. C., & Calderon, J. M. (2022). Optimal system based on differential evolution algorithm. In 2010 2nd International confer-
PID phi axis control for UAV quadrotor based on multi-objective PSO. IFAC Papers ence on advanced computer control, vol. 3 (pp. 348–351). http://dx.doi.org/10.1109/
OnLine, 55, 101–106. ICACC.2010.5486842.
Castillo, P., Lozano, R., & Dzul, A. (2005). Advances in industrial control, Modelling and Mahony, R., Kumar, V., & Corke, P. (2012). Multirotor aerial vehicles: Modeling,
control of mini flying machines. Springer. estimation, and control of quadrotor. IEEE Robotics & Automation Magazine, 19(3),
Chen, C.-H., Lin, C.-J., & Lin, C.-T. (2009). Nonlinear system control using adaptive 20–32. http://dx.doi.org/10.1109/MRA.2012.2206474.
neural fuzzy networks based on a modified differential evolution. IEEE Transactions Mian, A. A., & Daobo, W. (2008). Nonlinear flight control strategy for an underactuated
on Systems, Man, and Cybernetics, Part C (Applications and Reviews), 39, 459–473. quadrotor aerial robot. In 2008 IEEE international conference on networking, sensing
Chen, Y., & Pérez-Arancibia, N. O. (2020). Controller synthesis and performance and control (pp. 938–942). http://dx.doi.org/10.1109/ICNSC.2008.4525351.
optimization for aerobatic quadrotor flight. IEEE Transactions on Control Systems Nicol, C., Macnab, C. J. B., & Ramirez-Serrano, A. (2011). Robust adaptive control of
Technology, 28(6), 2204–2219. http://dx.doi.org/10.1109/TCST.2019.2919819. a quadrotor helicopter. Mechatronics, 21, 927–938.
Cheng, S.-L., & Hwang, C. (2001). Optimal approximation of linear systems by a Noroozi, M., Mohammadi, H., Lashgari, E. E. A., Eslami, M., & Khan, B. (2022). Golden
differential evolution algorithm. IEEE Transactions on Systems, Man, and Cybernetics search optimization algorithm. IEEE Access, 10, 37515–37532.
- Part A: Systems and Humans, 31(6), 698–707. http://dx.doi.org/10.1109/3468. Önder Efe, M. (2011). Neural network assisted computationally simple PID control of
983425. a quadrotor uav. IEEE Transactıons on Industrial Informatics, 7, 357–361.
Chiha, I., Ghabi, J., & Liouane, N. (2012). Tuning PID controller with multi-objective
Paez, D., Romero, J. P., Noriega, B., Cardona, G. A., & Calderon, J. M. (2021).
differential evolution. In 2012 5th International symposium on communications, control
Distributed particle swarm optimization for multi-robot system in search and rescue
and signal processing (pp. 1–4). http://dx.doi.org/10.1109/ISCCSP.2012.6217801.
operations. IFAC PapersOnLine ScienceDirect, 54, 1–6.
Das, A., Subbarao, K., & Lewis, F. (2008). Dynamic inversion of quadrotor with zero-
Pounds, P., Mahony, R., & Corke, P. (2010). Modelling and control of a large quadrotor
dynamics stabilization. In 2008 IEEE International conference on control applications
robot. Control Engineering Practice, 18(1), 691–699.
(pp. 1189–1194). IEEE.
Derafa, L., Madani, T., & Benallegue, A. (2006). Dynamic modelling and experimental Precup, R., David, R., Petriu, E., Preitl, S., & Radac, M. B. (2012). Proceedings of the 1st
identification of four rotors helicopter parameters. In 2006 IEEE international IFAC conference on embedded systems, computational intelligence and telematics
conference on industrial technology (pp. 1834–1839). http://dx.doi.org/10.1109/ICIT. in control - CESCIT. In Charged system search algorithms for optimal tuning of PI
2006.372515. controllers, vol. 45 (pp. 115–120). Würzburg Germany.
Dierks, T., & Jagannathan, S. (2010). Output feedback control of a quadrotor UAV Roger Miranda-Colorado, L. T. A. (2020). Robust PID control of quadrotors with power
using neural networks. IEEE Transactions on Neural Networks, 21(1), 50–66. http: reduction analysis. ISA Transactions 98, 98, 47–62.
//dx.doi.org/10.1109/TNN.2009.2034145. Salih, A. L., Moghavvemi, M., Mohamed, H. A. F., & Gaeid, K. S. (2010). Modelling
Ding, X., & Yu, Y. (2013). Motion planning and stabilization control of a multipro- and PID controller design for a quadrotor unmanned air vehicle. In 2010 IEEE
peller multifunction aerial robot. IEEE/ASME Transactions on Mechatronics, 18(2), international conference on automation, quality and testing, robotics (AQTR), vol. 1
645–656. http://dx.doi.org/10.1109/TMECH.2011.2182202. (pp. 1–5). http://dx.doi.org/10.1109/AQTR.2010.5520914.
Dong, X., Yu, B., Shi, Z., & Zhong, Y. (2015). Time-varying formation control Sarker, R. A., Elsayed, S. M., & Tapabrata, R. (2014). Differential evolution with
for unmanned aerial vehicles: Theories and applications. IEEE Transactions on dynamic parameters selection for optimization problems. IEEE Transactions on
Control Systems Technology, 23(1), 340–348. http://dx.doi.org/10.1109/TCST.2014. Evolutionary Computation, 18(5), 689–707. http://dx.doi.org/10.1109/TEVC.2013.
2314460. 2281528.

24
A. Gün Expert Systems With Applications 229 (2023) 120518

Storn, R., & Price, K. (1995). Differential evolution-a simple and efficient adaptive Wang, Y., Gao, S., Zhou, M., & Yu, Y. (2021). A multi-layered gravitational search
scheme for global optimization over continuous spaces: (1st ed.). Technical Report algorithm for function optimization and real-world problems. IEEE/CAA Journal of
No. TR-95-012, Berkeley, California, USA: International Computer Science Institute Automatica Sinica, 8(1), 94–109. http://dx.doi.org/10.1109/JAS.2020.1003462.
(ICSI). Yu, W.-J., Shen, M., Chen, W.-N., Zhan, Z.-H., Gong, Y.-J., Lin, Y., Liu, O., &
Sundareswaran, K., Devi, V., Sankar, S., Nayak, P. S. R., & Peddapati, S. (2014). Zhang, J. (2014). Differential evolution with two-level parameter adaptation. IEEE
Feedback controller design for a boost converter through evolutionary algorithms. Transactions on Cybernetics, 44(7), 1080–1099. http://dx.doi.org/10.1109/TCYB.
IET Power Electronics, 7(4), 903–913. 2013.2279211.
Sushama Nagpal, S. D. (2017). Feature selection using gravitational search algorithm
Zhang, R., Quan, Q., & Cail, K. Y. (2011). Brief Paper Attitude control of a quadrotor
for biomedical data. In Procedia computer science, vol. 115 (pp. 258–265). Cochin,
aircraft subject to a class of time-varying disturbances. IET Control Theory &
India.
Applications, 1140–1146.
Tayebi, A., & McGilvray, S. (2006). Attitude stabilization of a VTOL quadrotor aircraft.
Zhao, W., Mirjalilib, S., & Wanga, L. (2022). Artificial hummingbird algorithm: A
IEEE Transactions on Control Systems Technology, 14(3), 562–571. http://dx.doi.org/
10.1109/TCST.2006.872519. new bio-inspired optimizer with its engineering applications. Computer Methods in
Triguero, I., García, S., & Herrera, F. (2011). Differential evolution for optimizing Applied Mechanics and Engineering, 388, 1–45.
the positioning of prototypes in nearest neighbor classification. Pattern Recogni- Zuo, Z. (2010). Trajectory tracking control design with command-filtered compensation
tion, 44(4), 901–916. http://dx.doi.org/10.1016/j.patcog.2010.10.020, URL https: for a quadrotor. IET Control Theory & Applications, 4(11), 2343–2355.
//www.sciencedirect.com/science/article/pii/S0031320310005042. Zuo, Z., & Wang, C. (2014). Adaptive trajectory tracking control of output constrained
Vallejo-Alarcón, M. A., Velasco-Villa, M., & Castro-Linares, R. (2020). Robust backstep- multi-rotors systems. IET Control Theory & Applications, 8(13), 1163–1174.
ping control for highly demanding quadrotor flight. Journal of Control Engineering
and Applied Informatics, 22(1), 51–62.

25