How to Cite:

Aljawad, R. A., & Al-Jilawi, A. S. (2022). Solving multiobjective functions of dynamics optimization
based on constraint and unconstraint non-linear programming. International Journal of Health
Sciences, 6(S1), 5236–5248. https://doi.org/10.53730/ijhs.v6nS1.6041

Solving multiobjective functions of dynamics

optimization based on constraint and
unconstraint non-linear programming

Rabab Abdulsattar Aljawad

Ministry of Education, General Directorate of Education in Karbala
Email: ababaljwad@gmail.com

Dr. Ahmed Sabah Al-Jilawi

College of Education for Pure Sciences, Department of Mathematics, University of
Babylon, Iraq.
Email: ahmed.aljelawy@uobabylon.edu.iq

Abstract---To better understand the connection between nonlinear

optimization problems and differential equations, this study uses the
mathematical models with one or more objective functions that are
constrained by restrictions and therefore take the form of differential
equations called dynamical optimization systems, which mean a
decision-making method that uses differential and algebraic equation
mathematical models to design-wise policies based on forecasts of
future events. This paper focuses on solving mathematical models
systems using the Python programming language.

Keywords---Dynamic Optimization (DO), Mathematical model,

Optimization techniques, and Python language.

Introduction

With differential and algebraic equation mathematical models, decision-makers

may develop intelligent policies based on forecasts of future outcomes using a
technique called "dynamic optimization." This form of analysis may use a wide
variety of tools and methodologies.

Numerous real-world optimization problems have dynamic search spaces due to

the objective function, the number of variables, and the presence of constraints
[1]. Optimizing problems that change over time and must be handled online using
an optimization approach are called dynamic optimization problems (DOPs) [2]. It
is not enough for algorithms to discover excellent solutions when solving DOPs;
they must also be capable of adapting to changing conditions to swiftly find a new
key when the prior one is no longer good enough [3]. Therefore, an appropriate

International Journal of Health Sciences ISSN 2550-6978 E-ISSN 2550-696X © 2022.

Manuscript submitted: 27 Feb 2022, Manuscript revised: 18 March 2022, Accepted for publication: 09 April 2022
5236
 5237

benchmark generator is critical for thoroughly evaluating the efficacy of

algorithms created for DOPs. Thus, the following vital qualities should be
included in a DOP benchmark generator [1], [2]:

 Simple to execute and evaluate; Information on the location of the global

optimum should be made available to scientists doing research and fitness
value, as well as the features of the landscape's components, such as their
evolving interaction structure, form, and intensity of change, among other
things. It is recommended that this information be not used in the
construction or tuning of algorithms and that the problem be treated as
algorithms are hidden in a black box; It enables algorithmic researchers to
explore and evaluate the behavior and performance of algorithms in a
controlled environment. Also, a good benchmark should be easy to use [15].
 Flexibility: The benchmark generator should have a great deal of
adjustability regarding the number of components, component form, size,
change frequency, and degree of change. Additionally, each feature's setting
should be self-contained. This independence enables researchers to
examine the influence of each attribute on algorithm behavior. Additionally,
the ability to generate examples of challenges of increasing complexity is
essential for a solid benchmark. Numerous real-world situations are
complicated and intricate [1], [2], and [4].
 Variety: The benchmark generator should generate problems instances with
a range of characteristics, including modularity (completely separable,
totally non-separable, and partly separable components), heterogeneity,
balanced to very high unbalanced sub-functions, dimensionality ranges
from low to high (largescale), and varying degrees of irregularity.

Numerous DOP benchmarks for various DOP domains have been suggested in the
literature [2], [5], including combinatorial [6], [7], continuous [4], [8], multi-
objective [9]- [11], and restricted DOPs [12] - [14]. This article discusses dynamic
continuous unconstrained, and restricted optimization problems, including
single- and multiobjective. The most extensively widely used and renowned
benchmark generators in this subject are based on the notion of a landscape
made up of multiple components. The width, height, and location of these
components change over time in most published DOP benchmarks [5].

Optimization

Optimization has evolved into a flexible technique with industry and information
technology applications to the social sciences. Methods for addressing
optimization issues are many, and they constitute a vast pool of problem-solving
technology. There are many different approaches that it is impossible to fully use
[28], [29]. They are specified in several technical languages and implemented in
various software programs. Many are never implemented. It is difficult to
determine which one is best for a specific situation, and there are far too few
opportunities to mix strategies with complementary qualities. The ideal situation
would be to combine various techniques under one roof so that they and their
combinations could all be used to address an issue. As it turns out, many of them
have a similar problem-solving style on some level [16], [25], and [26]
5238

Differential Equations

Differential equations may be used to model practically any system that is

changing. They permeate all fields of study, including science and engineering,
economics, sociology, biology, business, and healthcare services. Various
mathematicians have studied the nature of these equations for hundreds of years,
and there are numerous well-developed solution methodologies. However, the
systems represented by differential equations are so sophisticated, or the
scenarios described by them are so huge that a solution to the equations using
just analytical methods is impracticable. In these complex systems, computer
simulations and numerical techniques are beneficial. Before the advent of
programmable computers, numerical approximation methods for solving
differential equations were developed. Throughout World War II, in rooms filled
with workers (typically women), mechanical calculators were often used to solve
systems of differential equations numerically for military calculations. Before the
advent of programmable computers, analog computers were often used to
investigate mechanical, thermal, or chemical systems. As the speed and
affordability of programmable computers have grown, it is possible to solve more
complex differential equations on a typical personal computer using simple
programs built for this purpose. Today, the computer on your desk can tackle
issues that were previously unreachable to even the best supercomputers just five
or ten years ago [17], [31].

Nonlinear programming

Problems involving nonlinear programming (NLP) have a nonlinear objective

function or constraints [18]. No efficient approaches exist for tackling the
nonlinear programming issue in its entirety. When it comes to solving problems
with ten variables and hundreds of variables, even the most straightforward
situations can be challenging. As a result, numerous techniques to solve the
general nonlinear programming issue have been developed, which entails some
degree of compromise [19].

Dynamic Optimization

Numerous optimization issues encountered in the real world are dynamic [20].
Additional jobs must should be included the timetable, the fundamental
constituent composition may vary, and new orders may be received, contributing
to the vehicle routing complexity. Due to the unpredictability inherent in dynamic
issues, fixing them is often more difficult than solving their static counterparts.
Additionally, solutions to dynamic issues must be discovered in real time as new
information is received. Even the optimization objective shifts from finding the
best solution to a static issue to monitoring the dynamic problem's shifting
optimum in real time.

Dynamics Optimization Modeling:

The following is a representation of the dynamic optimization model:

𝑚𝑖𝑛𝑖𝑚𝑖𝑧𝑒 𝐽(𝑥, 𝑦, 𝑝)
 5239

𝑑𝑥
𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜 𝑓 ( , 𝑥 , 𝑦 , 𝑝) = 0
𝑑𝑡
𝑑𝑥
0 ≤ 𝑔 ( 𝑑𝑡 , 𝑥 , 𝑦 , 𝑝 )

We note that 𝐽(𝑥, 𝑦, 𝑝) is the objective function of a dynamic optimization system;

where 𝑝 input values, 𝑦 output values, and 𝑥 are the system.

Figure (1): describe the input, output, and system

𝑓 , 𝑔 are constraints of a dynamic optimization system where these constraints are

either equality (𝑓) or inequality (𝑔), and the constraints contain differential
𝑑𝑥
equations ( )
𝑑𝑡

Dynamic optimization benchmarking

The development of benchmark problems is a key subject in the field of dynamic

optimization, fitness landscapes that are varied and dynamic may be represented
using this method. The literature has a large number for dynamic optimization in
the form of test issues. Grefenstette [23] characterizes his dynamic environment
as an assembly of components. The Natural Sciences and Engineering Research
Council of Canada provided funding for this work. Each part is made of a single
time-varying Gaussian peak in n dimensions. Three time-varying characteristics
define each height: center, amplitude, and breadth. Brake [22] proposes a
"Moving Peaks Function" in a similar paper. The author devises for the first time,
a multimodal function with adjustable peak height, breadth, and center was
introduced. In addition, he thoroughly analyzes the literature on standards for
dynamic landscapes. Bear in mind, although the majority of these concerns
involve real space, they are not all applicable to all combinatorial problems.

Basic concept of dynamic optimization

Python is a programming language used to aid with dynamic simulation,

estimation, and control solutions. Numerous tools for modeling and optimizing
dynamic systems are available. Additionally, there are other things to consider
while picking the best instrument for a specific activity. For example, many
factors should be considered when choosing an optimization tool for dynamic
scalings, such as size (scaling with number equations and degrees of freedom),
speed (use in real-time estimation or control), adaptability, extensibility, and
support (commercial vs. open-source community) (ability to embed on a system or
communicate with a particular platform).

*GEKKO combines machine learning and optimization strategies to solve complex

mixed-integer and differential algebraic problems. It is used with solvers for large-
5240

scale linear, quadratic, nonlinear, and mixed-integer programming (LP, QP, NLP,
MILP, MINLP). All types of operation are available, including parametric
regression, data reconciliation, real-time optimization, dynamic simulation, and
nonlinear predictive control. GEKKO is a Python object-oriented library for
executing AP Monitor on a local machine [26].

Algorithms of Dynamic Systems

Any computer procedure that receives an input value or set of values and outputs
an output value or set of values is a formula. Thus, an algorithm is a sequence of
computations that alter the data state. Additionally, an algorithm may be seen as
a tool to solve a well-defined computing issue. The issue statement provides a
comprehensive overview of the intended input/output relationship. The algorithm
defines the computational technique for creating the suitable input/output
connection. For instance, we may need to ascend a series of integers. This is a
common occurrence in reality and gives an excellent opportunity to introduce a
variety of conventional design methodologies and analytical tools [21] and [30].

Figure (2): Dynamic Optimization Algorithm

Optimal control benchmarking

Case 1: This is a standard chemical dynamic optimization problem with an

analytical solution, a mathematical system with no constraints. The state has an
 5241

analytical and global optimum solution due to two state variables. As an example,
consider the following mathematical model:

min 𝑥2 (𝑡𝑓 ) 
𝑢(𝑡)
subject to
𝑑𝑥1
=𝑢
𝑑𝑡
𝑑𝑥2
= 𝑥12 + 𝑢2
𝑑𝑡
𝑥(0) = [2 0]𝑇
𝑡𝑓 = 1
To solve this dynamic optimization system, we use GEKKO python code to find
the optimal solution by minimizing the final state of the nonlinear system with an
unconstraint dynamic such that the value of 𝑢 = −0.5.

Figure (3): Dynamic optimization created the control and state profiles

Case 2: It's a typical chemical dynamic optimization issue with an analytical

solution, a dynamic mathematical system. Including four state variables results
in an analytical and global optimal solution for the state. For example, have a
look at this mathematical formula:

min 𝑥4 (𝑡𝑓 ) 
𝑢(𝑡)
subject to
𝑑𝑥1
= 𝑥2
𝑑𝑡
𝑑𝑥2
= −𝑥3 𝑢 + 16𝑡 − 8
𝑑𝑡
𝑑𝑥3
=𝑢
𝑑𝑡
𝑑𝑥4
= 𝑥12 + 𝑥22 + 0.005(𝑥2 + 16𝑡 − 8 − 0.1𝑥3 𝑢2 )2
𝑑𝑡
𝑥(0) = [0 −1 −√5 1]𝑇
−3 ≤ 𝑢 ≤ 11
5242

𝑡𝑓 = 1

𝑥1 (𝑡) to 𝑥4 (𝑡) are state vectors, while 𝑢(𝑡) is the control vector.
For solving nonlinear systems that have the constraint to minimize the final state,
we use code GEKKO of Python and get:

Figure (4): The control and state profiles were developed using dynamic
optimization

Multiobjective Functions

We will look at optimization problems with a single objective function in this part,
both with and without constraints. Cost reduction, efficiency optimization, and
weight reduction are examples of conventional single-variable goal functions. Two
or more objective functions must be optimized concurrently in multiobjective
optimization situations. For example, the criteria may have cost reduction and
efficiency maximization throughout the product's manufacturing process. The
formal formulation of a multiobjective optimization problem is [24], [25], and [26]

Minimize Or Maximize 𝑓𝑘 (х) 𝑘 = 1,2, … , 𝐾

𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜: 𝑔𝑖 (х) ≤ 0 𝑖 = 1, 2, … , 𝑚 < 𝑛
ℎ𝑗 (х) = 0 𝑗 = 1, 2, … , 𝑟 < 𝑛
х𝑙 ≤ х ≤ х𝑢

х is a vector that contains the n design variables defined by

 5243
х1
х2
.
х= .
.
[х𝑛 ]

A solution to a multiobjective issue generates a set of Pareto optimum locations in

the objective function space. Figure (5) illustrates a typical Pareto optimum front
for a problem with multiple objectives and two objective functions (efficiency
maximization and cost reduction). The x-axis of this picture includes the first goal
(f1) function, "efficiency," and the second objective (f2) function, "cost," lies on the
y-axis. The principle of dominance is used to produce the Pareto optimum front.
Each answer is compared in this idea to see if it dominates another solution. For
example, suppose the following requirements are met. A solution х 1 is considered
to be superior to another х 2.

 х 1 is not worse than х 2 in all objectives.

 х 1 is superior to х 2 in a minimum of one goal.

Consider points A and C in terms of dominance. As expected, point C is superior

to point A in both objective functions. However, at least one of the points on the
Pareto optimum front is dominated by point C. Non-dominated solutions are the
spots along the Pareto optimum front. For example, the Pareto optimum front is
convex in Figure (5). This front, in contrast, may be concave, somewhat convex
(concave), or discontinuous. Alternatively, the Pareto front's form is determined by
the trade-off between the goal functions [24].

Figure (5): Pareto optimal front

Solving multiple objective function of dynamic optimization

Numerous optimization issues include conflicting goals. These conflicting aims

are a component of the trade-off that characterizes the best solution.
Occasionally, these opposing aims have distinct priorities, with one being
accomplished before the other is considered. This is particularly true in model
predictive control and other forms of dynamic optimization. With an ordered
5244

hierarchy, there are competing aims. If extra degrees of freedom are available, the
highest-level goals are met first, followed by lower-level objectives. The L1-norm
objective is a natural approach to ranking goals explicitly and optimizing many
priorities concurrently using a single optimization problem. For example, consider
the limits or purposes associated with safety, the environment, and economics.
Which are the most critical and why?
Case 3: Create a probable optimum trajectory for the following multiobjective
optimization problem.

Figure (6): Describe the safety, environmental and economic constraints

Discussion of the Results

Each of the instances discussed in this study may demonstrate the final results
after they have been numerically solved using Python scripts, particularly the
library GEKKO. They can explain the final results of each case with a simple
demonstration.

Case1: With a two-variable solution to the dynamic issue, we'll also minimize by
altering the value of 𝑢, which we'll do by changing 𝑥1and 𝑥2. 𝑥1 is equal to 2, and
𝑥2 is equal to 0; therefore, we're going to compute the value until 𝑥1 is equal to 2,
and we'll do the same thing for 𝑥2 until it is equal to 0, as you can see from our
input or variables used to manipulate the X conditions, note Figure (3). Then we
get:
 5245

Number of Iterations: 3
Table (1)
Final Results when Solve Case 1

(scaled) (unscaled)
Objective 3.0347753804326336e+00 3.0347753804326336e+00

Dual infeasibility 8.8817841970012523e-16 8.8817841970012523e-16

Constraint violation 8.8817841970012523e-16 8.8817841970012523e-16

Complementarity 0.0000000000000000e+00 0.0000000000000000e+00

Overall NLP error 8.8817841970012523e-16 8.8817841970012523e-16

The final value of the objective function is 3.03477538043263.

Case 2:
With the four-variable solution to the dynamic issue, we'll also minimize by
altering the value of 𝑢, which we'll do by changing 𝑥1 , 𝑥2 , 𝑥3 and 𝑥4. 𝑥1 is equal to
2, 𝑥2= -1, 𝑥3= √3 and 𝑥4 is equal to 1; therefore, we're going to compute the value
of x's, as you can see from our input or variables being used to manipulate the X
conditions, see figure (4).

Number of Iterations: 19
Table (2):
Final Results when Solve Case 2

(scaled) (unscaled)
Objective 2.5121002922896420e+00 2.5121002922896420e+00

Dual infeasibility 6.2326704832482986e-07 6.2326704832482986e-07

Constraint violation 4.4408920985006262e-15 4.4408920985006262e-15

Complementarity 1.4447122070015801e-10 1.4447122070015801e-10

Overall NLP error 6.2326704832482986e-07 6.2326704832482986e-07

The final value of the objective function is 2.51210029228964.

Case 3: Figure (6) shows how to solve a multiobjective optimization problem in

model predictive control or dynamic optimization. In many cases, we might have
high-priority items like safety constraints. For example, priority two might be
something like an environmental constraint or regulation, and priority three
might be to make the process as profitable as possible.
When using python code to solve case 3, we get:
5246

Number of Iterations: 28
Table (2)
Final Results when Solve Case 3

(scaled) (unscaled)
Objective 1.2277417296878346e+04 3.6832251890635038e+04

Dual infeasibility 7.1054273576010019e-13 2.1316282072803006e-12

Constraint violation 9.4368957093138306e-16 9.4368957093138306e-16

Complementarity 1.1738233691638086e-11 3.5214701074914258e-11

Overall NLP error 1.1738233691638086e-11 3.5214701074914258e-11

The final value of the objective function is 36832.2518906350.

Conclusion

This paper presented a technique for tackling single- and multiobjective,

continuous, restricted, and unconstrained dynamic optimization problems. Please
remember that many real-world problems may be optimized and solved using
Python to arrive at the ideal answer.

References

[1] T. T. Nguyen, "Continuous dynamic optimization using evolutionary

algorithms," Ph.D. dissertation, University of Birmingham, 2011.
[2] T. T. Nguyen, S. Yang, and J. Branke, "Evolutionary dynamic optimization:
A survey of state of the art," Swarm and Evolutionary Computation, vol. 6,
pp. 1–24, 2012.
[3] D. Yazdani, "Particle swarm optimization for dynamically changing
environments with particular focus on scalability and switching cost," Ph.D.
dissertation, Liverpool John Moores University, Liverpool, UK, 2018.
[4] C. Li, T. T. Nguyen, S. Zeng, M. Yang, and M. Wu, "An open framework for
constructing continuous optimization problems," IEEE Transactions on
Cybernetics, pp. 1–15, 2018.
[5] C. Cruz, J. R. Gonz'alez, and D. A. Pelta, "Optimization in dynamic
environments: a survey on problems, methods, and measures," Soft
Computing, vol. 15, no. 7, pp. 1427–1448, 2011.
[6] S. Yang, Y. Jiang, and T. T. Nguyen, "Metaheuristics for dynamic
combinatorial optimization problems," IMA Journal of Management
Mathematics, vol. 24, no. 4, pp. 451–480, 2013.
[7] M. Mavrovouniotis, F. M. Muller, and S. Yang, "Ant colony optimization with
local search for dynamic traveling salesman problems," IEEE Transactions
on Cybernetics, vol. 47, no. 7, pp. 1743–1756, 2017.
[8] J. Branke, "Memory enhanced evolutionary algorithms for changing
optimization problems," in IEEE Congress on Evolutionary Computation.
IEEE, 1999, pp. 1875–1882.
 5247

[9] S. B. Gee, K. C. Tan, and H. A. Abbass, "A benchmark test suite for dynamic
evolutionary multiobjective optimization," IEEE Transactions on
Cybernetics, vol. 47, no. 2, pp. 461–472, 2017.
[10] S. Jiang, M. Kaiser, S. Yang, S. Kollias, and N. Krasnogor, "A scalable test
suite for continuous dynamic multiobjective optimization," IEEE
Transactions on Cybernetics, pp. 1–13, 2019.
[11] S. Jiang and S. Yang, "Evolutionary dynamic multiobjective optimization:
Benchmarks and algorithm comparisons," IEEE Transactions on
Cybernetics, vol. 47, no. 1, pp. 198–211, 2017.
[12] T. T. Nguyen and X. Yao, "Continuous dynamic constrained optimization—
the challenges," IEEE Transactions on Evolutionary Computation, vol. 16,
no. 6, pp. 769–786, 2012.
[13] C. Bu, W. Luo, and L. Yue, "Continuous dynamic constrained optimization
with ensemble of locating and tracking feasible regions strategies," IEEE
Transactions on Evolutionary Computation, vol. 21, no. 1, pp. 14–33, 2017.
[14] Y. Wang, J. Yu, S. Yang, S. Jiang, and S. Zhao, "Evolutionary dynamic
constrained optimization: Test suite construction and algorithm
comparisons," Swarm and Evolutionary Computation, vol. 50, p. 100559,
2019.
[15] Yazdani, D, Omidvar, MN orcid.org/0000-0003-1944-4624, Cheng, R et al.
(3 more authors) (2020) Benchmarking Continuous Dynamic Optimization:
Survey and Generalized Test Suite. IEEE Transactions on Cybernetics. ISSN
2168-2267 https://doi.org/10.1109/TCYB.2020.3011828.
[16] INTEGRATED METHODS FOR OPTIMIZATION by JOHN N. HOOKER.
[17] introduction to numerical methods of differential equations.
[18] INTRODUCTION TO APPLIED OPTIMIZATION, Second Edition. By URMILA
DIWEKAR, Vishwamitra Research Institute, Clarendon Hills, IL, USA.
[19] Convex Optimization, Stephen Boyd, Department of Electrical Engineering,
Stanford University, Lieven Vandenberghe, Electrical Engineering
Department, University of California, Los Angeles.
[20] L. Bianchi, Notes on dynamic vehicle routing - state of the art - Technical
report idsia 05-01, Italy, 2000.
[21] Introduction to Algorithms. Third Edition, Thomas H. Cormen, Charles E.
Leiserson, Ronald L. Rivest, Clifford Stein.
[22] J. Branke, Evolutionary Optimization in Dynamic Environments. Kluwer
Academic Publishers, 2002.
[23] J. Grefenstette, Evolvability in Dynamic Fitness Landscapes: A Genetic
Algorithm Approach. In Proc. 1999 Congress on Evolutionary Computation
(CEC 99), Washington, DC. IEEE Press, pp. 2031-2038.
[24] OPTIMIZATION Algorithms and Applications Rajesh Kumar Arora, Senior
Engineer, Vikram Sarabhai Space Centre, Indian Space Research
Organization, Trivandrum, India.
[25] Al-Jilawi, A. S., & Abd Alsharify, F. H. (2022). Review of Mathematical
Modelling Techniques with Applications in Biosciences. Iraqi Journal For
Computer Science and Mathematics, 3(1), 135-144.
[26] Alridha, A., Wahbi, F. A., & Kadhim, M. K. (2021). Training analysis of
optimization models in machine learning. International Journal of Nonlinear
Analysis and Applications, 12(2), 1453-1461.
[27] Kadhim, M. K., Wahbi, F. A., & Hasan Alridha, A. (2022). Mathematical
optimization modeling for estimating the incidence of clinical diseases.
5248

International Journal of Nonlinear Analysis and Applications, 13(1), 185-

195.
[28] Alridha, A., Salman, A. M., & Al-Jilawi, A. S. (2021, March). The
Applications of NP-hardness optimizations problem. In Journal of Physics:
Conference Series (Vol. 1818, No. 1, p. 012179). IOP Publishing.
[29] Salman, A. M., Alridha, A., & Hussain, A. H. (2021, March). Some Topics on
Convex Optimization. In Journal of Physics: Conference Series (Vol. 1818,
No. 1, p. 012171). IOP Publishing.
[30] Alridha, A., & Al-Jilawi, A. S. (2021, March). Mathematical Programming
Computational for Solving NP-Hardness Problem. In Journal of Physics:
Conference Series (Vol. 1818, No. 1, p. 012137). IOP Publishing.
[31] Alridha, A. H., & Al-Jilawi, A. S. (2022). Solving NP-hard problems using a
new relaxation of approximate methods. International Journal of Health
Sciences, 6(S3), 523–536. https://doi.org/10.53730/ijhs.v6nS3.5375