УДК 004.896

Baikadamov S.S.

Kazakh-British Technical University (Almaty, Kazakhstan)

DISTRIBUTION OF FACILITY RESOURCES BASED ON COMBINATION OF AI METHODS AND PID CONTROLLER

Аннотация: this research examined and developed a Supervisory Control and Data Acquisition (SCADA) system for the monitoring of PID process control. In this instance, the test model was designed to be a DC-motor system. The SCADA system offered a methodfor tracking and reducing the control loop's uncertainty by recalibrating the parameters. The material resistance, or the swift changes in the environment, were typical examples of system dynamics changes that led to the control loop's uncertainty. Despite the fact that many PLCs feature tuning processes for adjusting these parameters, PLCs still require instruction on when to tune. Integrating artificial intelligence methods into the allocation of facility resources represents a significant advancement in facility management. By leveraging the fine-grained decision-making capabilities of AI, facilities can achieve higher levels of efficiency and adaptability, ensuring resources are responsive. It can be used optimally to suit the dynamic requirements of your environment. This approach not only improves operational performance, but also contributes to the broader goals of sustainability and user satisfaction.

The effective allocation of resources within a facility is essential for improving operational efficiency and long-term viability. This article examines how genetic algorithms and particle swarm optimization can be used to create and improve Proportional-Integral-Derivative (PID) controllers. These controllers play a crucial role in regulating system behavior and meeting specific performance standards in different locations. Genetic algorithms are used to find the best PID settings by gradually improving solutions through generations, guaranteeing resilience and effectiveness. Particle swarm optimization enhances the solutions by imitating social behaviors observed in nature, assisting in reaching the optimal controller parameters. Comparative studies have shown that the hybrid approach is more effective in achieving quicker response times, reducing errors, and enhancing stability in various simulated situations. This work offers a structured approach for choosing the most appropriate PID controller settings and also adds valuable insights to the study of resource management in automated systems.

Ключевые слова: SCADA, artificial intelligence, genetic algorithm, facility management.

INTRODUCTION.

Although artificial intelligence research has shown encouraging results, the construction industry currently lacks many applications. The efficiency of AI approaches in routine operations and maintenance has not yet been fully tapped into by Facility Management (FM) in the construction industry. An infrequent HVAC problem can result in a significant financial loss because HVAC is such an important aspect of Facility Management and Maintenance (FMM) operations. By reducing energy consumption, planning maintenance, and monitoring equipment, the implementation of AI approaches in FMM can optimize building performance, particularly in predictive maintenance.

In the last ten years, artificial intelligence has advanced dramatically, causing fundamental shifts in technical paradigms across a range of industries, from electronics to medical research. AI techniques can analyze massive amounts of data with accuracy levels so high that they outperform human performance and productivity while also utilizing less time and money. With their applications for autonomous driving and automatically sensing traffic lights, stop signs, obstructions, and other things, those innovations completely changed the automotive industry. AI is also being utilized in automation in the construction sector to detect individuals nearby heavy equipment to increase safety. The adoption of AI techniques in the construction industry has increased due to the high performance levels of computer vision techniques like Deep Learning and Fuzzy Logic.

I Literature review.

The first researchers in the list from University of Bremen [1] also noted the lack of practical application in production control and deal with use of artificial neural networks as control system for shop floor environment. Next one [2] evaluated several machine learning methods and tried to prove that Deep learning is the best suitable for sphere of automation. Research [3] and [4] consist of similar ideas of automation

building re- sources, for Sport facilities and Aquaponics facilities respectfully. This would help us to understand how the type and amount of resources effect our control system performance. The most recent researches like [5] or [6] provide case study of real SCADA control system. AI techniques are being used to solve device diagnostic and preventive maintenance problems using data from field devices.

As it can be seen, there are several excellent researches all over the net, however they are speaking in different languages. Currently, there are no paper about applying above mentioned AI methods to the real working control system. Either its only theoretical or its only regarding manual monitoring of facility resources. In this paper we will try to combine both sides.

The importance of applying intelligent distribution to production facilities by developing complex supersystems with interconnected subsystems could aid in realtime online monitoring and control [7]. Using machine learning technology, energy hubs can efficiently balance energy supply and demand, ensuring optimal use of renewable energy sources while minimizing dependence on traditional energy sources [8]. Facilities management is a set of solutions that help minimize the time and resources spent on real estate management issues and extend the life of buildings and technical systems [9].

The problem statement of the research is formulated as follows: it is necessary to study the PID controller behavior configured by several artificial intelligence methods in order to reach desired value of DC-motor output.

II Research methods.

In this work, first of all, the plan is to come up with control system that collects data from various field equipment such as sensors, tanks, load resistors and willing to pro- vide it some Human-Machine Interface (HMI). Next step is to input that information to AI source methods. Then we will try to train our artificial intelligence to appropriately react with certain output. For example, if the liquid in the tank is getting low, corresponding action should be taken fulfill it back. In this section would

be explained the main three research methods such as genetic algorithm, fuzzy logic and particle swarm optimization.

a) Genetic algorithm description.

The genetic аlgorithm (GА) is an advanced optimization method based on the principles of natural selection and genetic mechanisms. Its goal is to provide approximate solutions to complex problems that may be difficult to solve using traditional methods. The algorithm simulates the evolutionary process by selecting the most suitable individuals for reproduction, thus producing offspring for subsequent generations. The main components of a GA include a set of possible solutions, selection based on fitness, crossover (recombination), and mutation. Through iterative generations, the population gradually evolves toward an optimal solution.

In the Figure represented the various possibilities of genetic algorithm's application:

0

Resource allocation

Optimization problems

Figure 1. Genetic algorithm's application.

From the figure above it is clear that genetic algorithms are very effective in optimizing the intelligent allocation of plant resources such as logistics, production, and service management.

b) Particle swarm optimization.

Particle Swarm Optimization is a metaheuristic optimization algorithm inspired by the social behavior of flocks of birds and schools of fish. It is often used to solve optimization problems, such as those encountered in research on diagnostics of industrial equipment.

Here is the Figure 2, where the simple flowchart of PSO process is shown:

Modification of every searching poinl

NO

Iteration = iteration + 1 С

Ç END )

Figure 2. Particle swarm optimization flowchart.

Below is a detailed description of the advantages of PSO:

- Simple and User-Friendly: PSO is easy to understand and implement and requires fewer lines of code compared to other methods of optimization. Only a few parameters need to be added, such as particles' number, the main coefficients and the inertial weight.

- Versatility: PSO could be used to solve a various type of optimization task including continuous, discrete, and multimodal functions. It can easily be combined with other optimization techniques to improve performance.

- Robustness: Due to the stochastic search process, PSO is less prone to being trapped in a local optimum compared to other methods. It handles noisy and dynamic optimization problems well.

- Parallelism: PSO is inherently parallel and can be efficiently implemented in parallel and distributed computing systems, resulting in faster convergence.

c) PID controller.

PID controller is a fundamental component of feedback control loops and is used to automatically adjust a process variable to maintain it at a desired setpoint. This mechanism is common in approximately 90% of all automatic control systems due to its versatility and efficiency. The PID algorithm calculates a control signal that consists of three different terms: proportional, integral and derivative. These three terms collectively define the corrective actions required to bring the process variable back into an acceptable range, hence the name PID.

Working Mechanism of a PID Controller: to understand the functionality of a PID controller, it is important to understand the dynamics of a feedback system. The heart of this system is a PID controller, which can be a separate device or an algorithm running on a microcontroller. The key parameter to be monitored is called a process variable. This variable can represent temperature, flow rate, pressure, rotation speed, or any other measurable characteristic of the system. A sensor is used to measure a process variable and send that information back to the controller, producing feedback.

CET DHIMT

ЭС 1 rUIW I

PID CONTROLLER

Control signal

ACTUATION PLANT/

DEVICE ■ SYSTEM

Figure 3. Flowchart of the feedback based PID control system.

The controller is programmed with a desired value or setpoint for a process variable, which is the goal the system is trying to achieve.

III Simulation modelling.

A simulation is run with the DC-Motor that has the following specifications:

2hp, 230 v, 8.5 amperes, 1500 rpm

Ra(Armature resistance) = 2.45 H,

La(Armature Inductance) = 0.035 H,

Kb(Back EMF) = 1.2 Vs/rad,

Jm(Moment of Inertia) = 0.022 kgm2,

Bm (Frictional Constant) = 0.5 x 10-3(NmS/rad).

The transfer function of DC-Motor is given by below:

6(s) _ _12_

7а(5) 0.00077S3 + 0.0539S2 + 1.441S

, (4)

Model of DC-Motor in Simulink, which is application of Matlab for modelling, is shows below:

Figure 4. Model of DC-Motor with PID controller in Simulink.

Step - step function which is used for providing input to our system, values between 0 and 1.

K(1,2,3) - coefficients of PID controller, which we going to manipulate using different AI methods, in order to gain better results.

DC-Motor - Transfer function of real DC-Motor, which was described earlier in above sections.

ITAE (Integral Time Absolute Error) - absolute time of when output signal reaches and stabilizes around our set value. Main metric based on which we going to evaluate the result at each generation of current AI method.

Output - scope function which contain and plots the result of current generation in a graph way.

Genetic algorithm implementation.

Coding of GA in Matlab workspace is shown below:

Figure 5. Coding of GA in Matlab working area.

'ga' - main function of Genetic Algorithm,

'no_var' - number of variables which we manipulate during simulation, in order to minimize final output,

'lb' - lower minimal value, if algorithm reaches it simulation would stop, 'ub' - upper maximum value, if algorithm reaches it simulation would stop, 'ga_opt' - parameters of gen_algorithm, in this case its set to 50 generations with 50 population in each,

k,best - would collect final result each time and output the best one at the end of simulation,

b) PSO implementing.

Coding of PSO in Matlab workspace is shown below:

Figure 6. Coding of PSO in Matlab working area.

'particleswarm' - main function of Particle Swarm Optimization, built in

Matlab.

'no_var' - number of variables which we manipulate during simulation, in order to minimize final output,

'lb' - lower minimal value, if algorithm reaches it simulation would stop, 'ub' - upper maximum value, if algorithm reaches it simulation would stop, 'PSO_opt' - parameters of gen_algorithm, in this case its set to 50 generations with 50 population in each,

k,best - would collect final result each time and output the best one at the end of simulation.

IV. Simulation Results. a) Genetic algorithm results

Best value for generic algorithm, which parameters were set as 50 generations with 50 population size in each, are shown below:

- Kp = 0.9112,

- Ki = 0,

- Kd = 0.3868,

- ITAE (best) = 3.6456

Review graph of 50 generations containing fitness function values is shown in Figure below:

Figure 7. Genetic algorithm - 50 generating representation.

The chart shows how an evolutionary algorithm performed over 50 generations. The algorithm quickly reaches a stable solution in the initial generations, with both the highest and average fitness values settling around zero. This means that the algorithm rapidly identifies a solution that is either optimal or very close to optimal, and continues to keep this solution stable with minimal changes in future iterations. Graphical representation of best result for Genetic Algorithm is shown below:

1 ГЗ

I I l I i i i i

1.06

1.04

1.02 1 \

0.98 0.96 0.94 \ /

I—

*i*Не можете найти то, что вам нужно? Попробуйте сервис подбора литературы.

[ 1 2 4 f 3 i 3 10 12 14 16 18 20

|Ready Sample based T=20.000

Figure 8. Genetic algorithm result.

The graph in Figure 8 displays a significant rise in value at the start (approximately at time 2), reaching a maximum just over 1.04. This shows that the Genetic Algorithm promptly identified a solution that greatly enhanced the fitness value. After reaching the highest point, the value decreases and fluctuates.

The initial decrease goes below 0.98 at around time 5, and is then followed by a subsequent increase leading to a smaller peak just above 1 at around time 7. This back and forth movement indicates that the algorithm is continuously fine-tuning the solution in order to find the best possible balance. Stabilization occurs as time goes on, with the value eventually leveling off near 1. Starting at age 10, the value stays relatively constant with very small changes, suggesting that the genetic algorithm has reached a stable and final answer.

The chart shows how the highest achievement in a Genetic Algorithm changes over a period of time. At first, there is a quick progress, but then there are fluctuations as the algorithm fine-tunes the solution. After some time, the value remains constant, indicating that the GA has reached a near-perfect solution through convergence.

The first peak and following fluctuations demonstrate the exploratory aspect of the GA, while the final stabilization shows that it has successfully converged.

b) Particle Swarm Optimization.

Best value for generic algorithm, which parameters were set as 50 iterations with 50 swarms in each, are shown below:

- Kp = 0.9068,

- Ki = 0,

- Kd = 0.3800,

- ITAE (best) = 3.6446

Review graph of 50 generations containing fitness function values is shown in Figure below:

X101S Best Function Value: 3.64457

12 r 10 -8 6 -4 -2

0 10 20 30 40 50

Iteration

Figure 9. PSO 50 generations implementing.

From the Figure 9 is clear, that at the start of iteration 0, the function value is very elevated. This suggests that the original solution or particle swarm possessed a greatly superior fitness value. Starting from the initial iteration, the function value significantly decreases to almost zero. This steep decrease indicates that the PSO promptly discovered much improved solutions, causing a significant drop in the

function value. Graphical representation of best result for PSO is shown below in Figure 9:

Ж]

<1

1 0.8 0.6 0.4 0.2 T

T

T

г

J

0 1

2 4 10 12 14 16 18 20

Ready Sample based T=20.000

Figure 10. The PSO application result.

The Figure 10 shows how the PSO algorithm performed over a 20-unit time period. At first, the level of fitness increases quickly, showing a rapid enhancement in the solution. Afterward, there is a small back and forth movement as the algorithm refines the solution. After some time, the fitness value settles near 1, indicating that the PSO has reached an optimal or nearly optimal solution and remains stable with minimal changes. This usual conduct of PSO shows its efficiency in quickly identifying and maintaining a top-tier solution.

The chart illustrates a steep rise in the fitness score, starting at zero and peaking just above 1 at approximately time 4. This shows that the PSO algorithm efficiently found a much-improved solution at the beginning of the process.

Movement and balance: oscillation and maintaining stability. Following the initial peak, there is a small fluctuation as the fitness value decreases slightly below 1 at time 6 before leveling off. This conduct indicates that the PSO is improving the

solution and fine-tuning the swarm's location in order to discover the best possible outcome. From approximately age 8 onwards, the fitness value stays fairly consistent with very minimal fluctuations, suggesting that the algorithm has reached a solution. Continued or constant state The fitness value hovers around 1 for most of the time period, from approximately time 8 to 20.

This stable condition indicates that the PSO algorithm has discovered an optimal or very close to optimal solution and is keeping it with minimal variation.

VI Conclusion.

Currently, factory automation uses a PLC-based automatic control system and develops its capabilities as technology advances. To improve the productivity of small and medium-sized factories worldwide, it is crucial to study control systems and SCADA. In this study, we studied so called "nature based" AI algorithms, like Genetic algorithm and Particle Swarm Optimization. We described pros and cons of each method and were able to apply Genetic Algorithm to the DC-motor model. Then we did the same with Particle Swarm Optimization. We haven't tried to combine these two artificial intelligence methods for the particular reason of comparing it to each other.

In conclusion, we can say that SCADA system control can enable the development of an automatic process up to the expertise and technology gained from general program application. And those operations are very sensitive when it comes to PID tuning. Artificial intelligence methods, particularly Genetic Algorithm and Particle Swarm Optimization have shown perspective results in minimizing both time and error values. Fuzzy Logic Controller in its case has shown lower results and need to mention that configuration of Fuzzy Logic itself is complicated and requires high experience.

СПИСОК ЛИТЕРАТУРЫ:

1. Elizabeth Bautista, Melissa Romanus, Thomas Davis, Cary Whitney, and Theodore Kubaska. Collecting, monitoring, and analyzing facility and systems data at

the national energy research scientific computing center // Association for Computing Machinery.- 2019.-Vol. 8.- P. 34-75;

2. Ji-Hyoung Chin, Chanwook Do, and Minjung Kim. How to increase sport facility users' intention to use ai fitness services: Based on the technology adoption model // International Journal of Environmental Research and Public Health.- 2022.-Vol.19.-P. 44 - 53;

3. Mariam Elnour, Yassine Himeur, Fodil Fadli, Hamdi Mohammedsherif, Nader Meskin, Ahmad M. Ahmad, loan Petri, Yacine Rezgui, and Andrei Hodorog. Neural network-based model predictive control system for optimizing building automation and management systems of sports facilities // Applied Energy.- 2022.- P. 31;

4. Sara Masoud, Bijoy Dripta Barua Chowdhury, Young Jun Son, Chieri Kubota, and Russell Tronstad. Simulation based optimization of resource allocation and facility layout for vegetable grafting operations // Computers and Electronics in Agriculture.-2019.- Vol.2.- P. 163;

5. Nico. Mastorakis. Recent researches in circuits, systems, communications and computers // European Conference of Computer Science.- 2022;

6. Teerawat Thepmanee, Sawai Pongswatd, Farzin Asadi, and Prapart Ukakimaparn. Implementation of control and scada system: Case study of allen bradley plc by using wirelesshart to temperature control and device diagnostic // Energy Reports.- 2022.- Vol. 8.- P. 934-941;

7. Anastasiia Shchurenko. Building an ecosystem and infrastructure for smart confectionery production.- 2024.- Vol. 3.- DOI: 10.32370/IAJ.3053;

8. Magdy Tawfik, Ahmed S. Shehata, Amr Ali Hassan, Mohamed A. Kotb. Introducing Optimal Energy Hub Approach in Smart Green Ports based on Machine Learning Methodology.- 2023.- Vol. 1.- DOI: 10.21203/rs.3.rs-3607053;

9. Karolina Viduto. Smart technologies in the field of the facility management // Mokslas - Lietuvos ateitis journal.-2021