Application of Artificial Intelligence in Control Systems: Trends, Challenges, and Opportunities

1. Introduction

Automatic control theory and artificial intelligence (AI) have developed side by side since the 1950s. They began with a focus on information science and later grew into control theory. In the 1980s, AI gained important uses in business and industry. It grew at a consistent pace until the end of the twentieth century, with many studies on intelligent controllers [1]. Control theory evolved into modern control as processes needed more complex control. It now combines AI techniques with both classical and contemporary methods [2].

The growing need for advanced solutions highlights the demand for intelligence control techniques. It shows how intelligent controllers manage complex processes. Traditional math models often struggle in these situations. Uncertainties in real systems usually show nonlinear dynamics and changing parameters. This has led to updates in AI theory and its use in controllers. AI and control theory work together. AI helps improve controllers, and they also give each other feedback. This boosts both fields. This article reviews the latest in AI controllers by analyzing 186 scientific papers

The authors organized this paper to provide a straightforward answer to key questions. It presents a clear synthesis of answers from the latest scientific literature. This study seeks to improve our understanding of intelligent controllers powered by artificial intelligence through analysis.

Justification of the review:

A systematic review of AI in control systems helps us to understand technology today. It shows the impact and explores future opportunities. Researchers can identify strengths, weaknesses, opportunities, and threats. This helps guide the evolution of these systems. It leads to better safety, greater effectiveness, and a systematic approach. It also helps build a strong framework for future research in this new field. It helps us to evaluate the latest applications. The lector can see which AI methods are the most successful. It shows the current limits and challenges of AI in control systems. This helps identify ways in which AI can increase efficiency, accuracy, and robustness. Control systems can adapt to new conditions and still perform well.

The systematic review sparks new ideas and methods. Combining various AI methods with traditional control strategies helps us create new architectures. This allows us to tackle challenges that seemed impossible before.

Article Structure:

Here is the framework of the article: Section 2 explains the method, covering data collection and analysis. Section 3 shows a systematic mapping. It shows important relationships and patterns. Section 4 concludes with findings, trends, gaps, and contributions, and provides recommendations. This section also lists the sources used in the references.

2. Methodology

a..

Design of the systematic mapping:

This study has two main steps for the systematic review: (1) establishing the review protocol and (2) conducting the evaluation process.

A review protocol defines a clear strategy for conducting a literature review. This article uses a systematic review process. It uses a simple method to analyze and combine the existing literature on a specific topic.

The process includes:

• Defining clear criteria for choosing relevant studies.
• Do thorough search in academic databases.
• Checking the quality of the chosen items.
• Combining the findings to draw conclusions or identify research gaps.

The aim is to provide a comprehensive and unbiased overview of the current state of knowledge on the subject. To start the review, the authors of this study created a search equation applied to IEEE Xplore, ACM Digital Library, ScienceDirect, Elsevier, Springer, and Google Scholar. This helped them access a wide variety of relevant scientific publications.

This review searches for the chosen bibliographic databases for relevant articles. It includes those published in journals and conference proceedings. They constructed search expressions such as:("Design Intelligent Controllers*") OR ("Control Systems Controllers*" AND "Intelligent*" AND ("Fuzzy" OR "Genetic Algorithms*" OR "PSO" OR "Machine Learning*" OR "Neural Networks*" OR "Hybrid Algorithms*")) OR ("Engineer applications*" AND "intelligent controllers*" AND ("Strategies*" OR "Methods*" OR "Process*")).

This study used the preliminary search equation to find 465 records. Table 1 summarizes these records and includes them in the evaluation process.

After gathering many articles, the review team filtered and selected them according to specific criteria. They checked each article's relevance and quality, discarding those that fell short. Only papers that provided helpful and relevant information for the research questions made the cut.

The inclusion, exclusion, and selection criteria used were:

Inclusion Criteria (CI):

CI-1: Relevant terms appear in the title of the document or abstract.

CI-2: The publication of the article is dated from 2000 to 2025.

CI-3: The abstract of the primary study refers to the problem covered by the corresponding research question.

CI-4: The primary study has an implementation in control engineering.

CI-5: The primary study addresses techniques and strategies for implementing an intelligent controller.

CI-6: Research articles consisting of an evaluation of the research presented (survey, experiment, experience report, and case study).

Exclusion Criteria (EC):

EC-1: Documents published before 2000.

EC-2: Documents presenting a general approach to control systems.

CE-3: The primary study of the candidate is not an article, survey, art study, or review.

CE-4: Documents that present a general approach in applied engineering.

CE-5: The candidate did not write the primary study in English.

CE-6: The primary study is not related to the problem of the corresponding research question.

Selection criteria (SC).

CS-1: Apply inclusion and exclusion criteria to abstracts of articles, surveys, art studies, and reviews.

CS-2: We read the conclusions of the remaining papers to identify the necessary properties that were not evident in the abstracts.

CS-3: After you have obtained a refined list, we will read all of them. To do this, we used the inclusion and exclusion criteria for its content.

CS-4: Documents presenting an overview of intelligent controllers.

CS-5: Research articles that have discussed problems, techniques, and methods of implementing intelligent controllers.

CS-6: Research articles that have discussed the limitations and success factors of intelligence controller deployment processes.

b..

Evaluation of the method:

Once the review protocol was defined, this study went to the evaluation phase of systematic analysis. We adopted an evidence-based approach, reporting the number of records identified, examined, verified for eligibility, and ultimately included in the review. The modified search query retrieved a total of 465 records. The authors analyzed each article's abstract in detail to assess its relevance. They selected articles that discussed control structures, algorithms, optimization techniques, and key parameters for full-text eligibility review, resulting in a final set of 186 articles (see Figure 1).

The authors of this work established and defined the research questions that guided the entire process. These questions aim to address specific aspects related to the application of artificial intelligence in control systems, such as:

Q1: What are the sources that publish the most about smart controllers?

Q2: What are the types of engineering contributions and applications to intelligent controllers?

Q3: What are the types and research approaches of intelligent controller studies?

Q4: What are the current research trends in intelligent controllers?

Q5: What control structures and techniques specify an intelligent controller?

Q6: What are the most reported computational tools or algorithms in the literature for the design of intelligent controllers?

Q7: What are the open challenges and research opportunities?

c..

Data extraction:

Data extraction from chosen studies for AI control systems followed a systematic process. This ensured an accurate and complete collection of relevant information. The steps for extracting data from the studies are clear and follow systematic review methods.

We defined the inclusion, exclusion, and selection criteria at the start of the relevant studies. We examined every study to discover the control structure, control techniques, optimization methods, adaptation methods, and algorithms used. We grouped the studies by their categories, based on what they contributed. From each selected article, we extracted key information, including:

• General Information of the Study: Year of publication, authors, title, source of publication.
• The sections —abstract, introduction, and methodology—were reviewed in detail. The abstract sought an overview of the primary technique or method used, while the introduction clarified the objectives and specific applications of the control system.
• In the Methods section, we outline the specific algorithms and procedures utilized in each study, which streamlined the accurate sorting of the articles into those targeting optimization and control system structure
• From the results and discussion, we extracted information on the performance of each technique and the justification for its application in the study.

Specific algorithms for optimization, control, and adaptation were documented, along with their distinctive features, in each article. This document allowed us to pool the studies using similar algorithms.

This study examined the structure of the control system to see how each technique works in specific contexts. This helped it sort the articles into categories such as:

• Neural networks
• Evolutionary algorithms
• Machine learning
• Predictive control
• Metaheuristic optimisation
• Fuzzy logic
• Hybrids
• And more.

We label each item according to the structure of the control system used and the type of optimization or technique.

We compared the extracted data to identify key patterns and differences in the use of control techniques and structures between studies.

To ensure consistency in classification and data extraction, we cross-reviewed the articles. This task involved verifying that we correctly classified items with similar characteristics into the same categories.

Additionally, we used tables and matrices to organize and visualize the data. This structure allowed for efficient comparison between studies and ensured the extracted data was complete and representative of each article's content.

3. Results and Discussions

From identifying sources of publications on smart controllers to assessing research trends and methodological approaches used in this field, this comprehensive analysis provides a detailed view of the current landscape. We analyze the most reported computational tools and algorithms in the literature for the design of these controllers.

3.1. Publication Sources

The sources of intelligent controller publications reveal a marked predominance of publishers, Elsevier being the most prominent, representing 75% of the total of 186 articles reviewed. IEEE and Springer share a similar percentage, each contributing 12.1%. Wiley contributes 8%.

We found that the journals published 82.3% of the reviewed articles. Specialized intelligent controller conferences presented the other 17.7%. These figures highlight publishers' significant role in disseminating scientific articles in this domain. Figure 2 summarizes these values.

3.2. Engineering Contributions and Applications in Smart Controllers

The review of research types has enabled us to identify and reduce three primary areas of contribution in the field of engineering. Firstly, the contribution related to control model strategies stands out, representing 49.19% of the total, which encompasses the development and improvement of new intelligent algorithms for controllers, whether intelligent or conventional, as well as the introduction of new control structures with intelligent algorithms. The category of optimization models constitutes 27.41%, distinguishing itself through the presentation of new intelligent optimization models and the integration of artificial intelligence techniques with other conventional or intelligent ones.

Finally, the contribution of adaptability, with 23.28%, focuses on developing intelligent controllers that exhibit adaptability in their structure to provide control with high-speed response, robustness, and other qualities. These results provide a comprehensive overview of the various engineering contributions and applications in the field of intelligent controllers. Figure 3 presents a summary of the total number of articles according to the contribution made to intelligent controllers.

3.3. Structures and Techniques in Intelligent Controllers

Intelligent controllers use diverse control structures and techniques to optimize efficiency and adaptability across various systems. Artificial neural networks are among the most common structures; they excel at modelling nonlinear behaviors and learning complex patterns. Fuzzy logic and its ability to handle uncertainty offer a crucial option in dynamic environments.

Many specific techniques use meta-heuristics such as particle swarm optimization (PSO) and genetic algorithms (GA) to fine-tune controller parameters for better performance. Predictive Control is particularly good at anticipating and managing changes in a system's dynamics. Multilayer neural networks (MNAs) have adaptable structures that can model complex systems and learn from data. The Adaptive Fuzzy Tracking Algorithm-based Control and Fuzzy MIMO Control prove how fuzzy logic can adapt to many inputs and outputs in complex systems. In e-learning, dynamic parameter adjustment and grammatical evolution are effective strategies. These techniques show how control engineering is moving towards more adaptable solutions, combining fundamental principles with AI strategies to tackle modern system complexities.

3.4. Most Reported Computational Tools and Algorithms

The results of this study offer information on preferences and trends in algorithms and control structures. This comes from an analysis of scientific articles published from 2000 to early 2025. The most common classifications found are:

• Artificial Neural Networks: 9.77%
• Evolutionary or Optimization Algorithms: 15%
• Fuzzy Logic: 19.46%
• Hybrid Methods: 26.6%
• Iterative Learning Control: 4.1%
• Internal Model Control: 3.1%
• Machine Learning: 1.85%
• Metaheuristic Optimization: 11.4%
• Model Predictive Control: 6.02%
• Others: 2.96%

For a better understanding, we provide a brief description of each algorithm mentioned in next subsection.

3.4.1. Artificial Neural Networks (ANNs) replicate the human brain, consisting of layered neurons that process data and refine weights to perform complex tasks. The layers consist of an input layer receiving the data, hidden layers that transform it, and an output layer that produces the result. ANNs employ backpropagation to eliminate errors and increase accuracy in applications such as image recognition and language processing. [129,130,131].

3.4.2. Evolutionary Algorithms (EAs) are methods inspired by biological evolution. They find optimal solutions using selection, crossover, mutation, and survival of the fittest. EAs are great for solving complex problems in large, nonlinear search spaces. Key components of EAs include:

• A population of possible solutions
• Selection processes to identify suitable individuals
• Crossover to create new solutions
• Mutation to maintain diversity
• Survival to maintain the strongest individuals

There are several types of EA. These include genetic algorithms, evolutionary programming, evolutionary strategies, and genetic programming. Researchers use these algorithms for optimization, design, artificial intelligence, and solving complex problems [132,133,134].

3.4.3. Fuzzy Logic (FL) is based on classical logic dealing with uncertainty and imprecision. Instead of just true or false, it uses truth values ranging from 0 to 1. This approach helps to model unclear situations, such as "high" or "hot." The key components of Fuzzy Logic include:

• Fuzzy sets: These have degrees of membership.
• Membership functions: They assign degrees of membership to items.
• Fuzzy rules: These guide decisions in an "if-then" style.
• Fuzzy inference: This process combines the rules.
• Defuzzification: Converts fuzzy results into clear values.

Solutions apply fuzzy logic in areas such as:

• Control systems
• Medical diagnosis
• Image processing
• Decision-making

This is especially useful when dealing with uncertainty [135,136,137].

3.4.4. Hybrid Methods (HMs) mix different approaches to tackle complex problems better. They use the strengths of each technique. HMs are flexible and can adapt to various issues. They overcome limitations while increasing accuracy, efficiency, and robustness.

Applications include hybrid optimization, hybrid neural networks that integrate FL, and fuzzy-genetic controllers that optimize control decisions. HMs are widely used in robotics, intelligent system design, control engineering, and optimization in economics and finance [138,139,140].

3.4.6. Iterative Learning Control (ILC) is an advanced control technique that improves performance in systems that perform repetitive tasks, adjusting control actions in each cycle to reduce errors based on previous attempts. This approach is helpful in applications that require high precision, such as in industrial automation, robotics, and manufacturing, where iterative learning optimizes the accuracy of the process. Through learning from past mistakes and iterative predictive control, ILC enables continuous improvements, making it ideal for high-precision systems, such as robots and devices such as 3D printers [141,142,143].

3.4.7. Internal Model Control (IMC) is a control technique that utilizes an internal model of the system to predict and compensate for disturbances in real-time, enabling early corrections to be applied and maintaining the accuracy of the results. IMC is exceptionally robust and effective in industrial environments with uncertainties, such as temperature and pressure control in chemical engineering and manufacturing processes. Its internal model allows for precise adjustments and rapid disturbance compensation, making it ideal for complex control systems in the process industry [144,145,146].

3.4.8. Machine Learning (ML) is a branch of AI that allows systems to learn and improve automatically from data, without the need for specific programming for each task. ML analyses patterns in the data to make predictions or decisions. Solutions applied ML in areas such as image recognition, language processing, fraud detection, and personalized recommendations. ML models are trained in large amounts of data to identify patterns, adjust parameters, and generalize new data. It encompasses various types, including supervised, unsupervised, and reinforcement learning, with applications in finance, healthcare, commerce, and autonomous vehicles [147,148,149,150].

3.4.9. Metaheuristic Optimization (MO) is a methodology that seeks approximate solutions to complex optimization problems, where traditional methods are ineffective due to high dimensionality, nonlinearity, or multiple local optima. MO explores the search space without depending on the specific characteristics of the problem. Natural processes inspire MOs approaches, such as biological evolution, flock behavior, or annealing in metallurgy. Used in engineering, logistics, finance, and bioinformatics, it alternates between exploring new solutions and refining current ones to optimize performance [151,152,153,154].

3.4.10. Model Predictive Control (MPC) is an advanced control technique that utilizes mathematical models to forecast the future behavior of a system and optimize control decisions in real-time. MPC calculates and applies optimal actions within a prediction horizon, continuously adapting through new measurements. These actions allow you to manage constraints and adjust decisions at every step, making them ideal for complex nonlinear systems with changing operating conditions. MPC is widely used in industrial processes, energy systems, autonomous vehicles, and finance, where adaptability and robustness are crucial to optimizing performance and minimizing costs [155,156,157,158].

3.5. Outcome of the Studies

Tables 2 to 4 present a classification of the 184 articles analyzed, specifying the type and focus of research that categorizes them into one of these three topics: the control model, the implementation of techniques for parameter optimization, and adaptive controllers. In addition, the classification is based on the type of contribution, exploring various contributions and engineering applications derived from advances in intelligent controllers, including contributions related to control, optimization, and adaptability models. In addition, the structures and control techniques used to specify and design intelligent controllers are analyzed, encompassing both classical and modern control structures, as well as the artificial intelligence techniques employed, such as neural networks, fuzzy logic, and evolutionary algorithms. Finally, the most widely used computational algorithms in the literature are those for designing and implementing controllers that enhance the efficiency and performance of intelligent controllers in various applications and scenarios.

3.5.1. Result of Contributions: Control Models

The integration of AI techniques into smart controllers represents a significant advance in the field of control systems. These new control models combine conventional methods with AI approaches to improve the adaptability, accuracy, and efficiency of control in complex and variable environments. This combination enables us to address challenges that range from variability in operating conditions to the presence of unforeseen disturbances, opening new opportunities to optimize and continuously improve system performance. These articles present a relevant contribution of control models in the following types:

3.5.1.1. Hybrid Controllers

This type of controller combines two or more control techniques or algorithms, typically to leverage the strengths of each and compensate for their weaknesses. Hybrid controllers are becoming increasingly popular in industrial, energy, and complex system applications, particularly in the context of AI, optimization, and classical control.

Advanced control and optimization approaches are evident in the articles, applied to dynamic, nonlinear, and complex systems. Several papers propose hybrid methods that combine techniques such as ANN and FL. At the same time, they apply EAs to improve model identification, controller design and tuning, and robustness against uncertainties and perturbations.

Several works address architectures that integrate ANNs and FL type-1, type-2 or fractional logic for control and compensation tasks in the presence of nonlinearities and uncertainties ([11,16,44,45,67,87,88,96,97,100,105]).

The application of genetic algorithms, differential evolution, evolutionary strategies, particle swarm optimization, harmony search, cuckoo search, and others for the optimization of strategies, design and rules in hybrid controllers is reported ([10,12,26,30,50,57,77,83,97,100,171,205]).).

Several studies have implemented model-based predictive control (MPC), often combined with ANNs or FL, to speed up calculations or improve tracking and robustness ([38,40,79,82,85,89,116]). The combination of classical control techniques (PI, PID, H∞, Deadbeat, SMC) with intelligent components (ANN, FL) allows for greater accuracy, adaptability, and robustness, even under changing conditions or with incomplete models ([14,21,25,41,42,43,46,57,63,112]).).

The revised approaches are applied and validated in real or simulated systems, such as power plants, turbines, robots, chemical processes, hybrid energy systems, irrigation control, and autonomous vehicles, showing improvements in performance, error reduction, robustness, and energy efficiency compared to traditional methods ([25,57,63,79,171,205]).

3.5.1.2. Fuzzy Logic Controllers

Fuzzy logic controllers (FLC) have proven to be highly effective in various applications. They are known for their ability to handle uncertainty in control systems and for their ease of adaptation to different operating conditions. These controllers use IF-THEN rules and fuzzy sets to adjust system output. They are beneficial in applications where operating conditions constantly change and require an adaptive response.

A Takagi–Sugeno-type incremental fuzzy state model improves stability and error rejection by integrating an FLC-LQR and an optimal observer [33]. In the automotive field, a comparison between PID and fuzzy logic controllers in vehicle stability shows that fuzzy logic is more robust, although PID is more straightforward to implement [36].

For the control of converters and rectifiers, a fuzzy neural type-2 controller (T2FNS) has been proposed, outperforming the traditional PI controller in speed, overshoot, and power quality [47]. Similarly, an FLC designed for a DC–DC converter achieves better response times and greater disturbance rejection compared to conventional PI controllers [17].

In water treatment systems, a Mamdani FLC has been employed for pH control, achieving system stability and reducing contaminants in scenarios where traditional PIDs fail [105]. For electric motors, a Mamdani FLC demonstrates better tracking performance and robustness against disturbances when compared to the PID [107].

In wind energy systems, an adaptive FLC is used to maximize power generation, delivering better dynamic performance than PI controllers [92]. In agricultural applications, an efficient FLC has been developed to regulate temperature and humidity in greenhouses, showing high efficiency and low error under real conditions [113]. Finally, for photovoltaic systems, an optimized FLC for MPPT achieves faster convergence and higher accuracy, while also reducing the number of rules and eliminating steady-state oscillations [121].

3.5.1.3. Predictive Controllers

Model-based predictive controllers (MPCs) predict system output based on an internal model and adjust control actions in real time. They are ideal for applications that require anticipation and continuous correction. A detailed review of the use of MPC in agriculture covers classical approaches as well as robust, nonlinear, distributed, and stochastic variants, reporting successful applications in irrigation, machinery, production, and environmental control in greenhouses [54]. It is worth noting that MPC surpasses traditional and AI-based methods by efficiently handling multivariable, time-delayed, and nonlinear systems. However, there are still challenges for its practical implementation in real agricultural environments.

A Robust Economic Predictive Control (reMPC) method integrates stochastic information on perturbations in time-invariant linear systems and employs a tube-based MPC formulation to ensure feasibility and robustness [34]. This approach makes use of known fault distributions to optimize performance and design suitable terminal costs, and has been validated in a CSTR reactor, demonstrating advantages over conventional robust MPCs.

3.5.1.4. Controllers with Iterative Learning

Applications that require precision in repetitive tasks utilize Iterative Learning Controllers (ILCs). These controllers adjust the control signal based on the errors observed in previous cycles, progressively reducing the error in each repetition. A comparison of three Norm-Optimal ILC schemes in repetitive systems—standard quadratic ILC (QILC), ILC with integrated estimation (E-QILC), and ILC with Kalman filter (K-ILC) in the iteration domain—shows that although the approaches are equivalent under certain conditions, the K-ILC method stands out by combining fast convergence with high robustness against noise [24]. Simulations on a mass-spring-shock absorber system validate its effectiveness.

A robust ILC controller incorporates performance weighting functions to ensure convergence even in the presence of plant uncertainty [73]. This method establishes clear convergence conditions and design criteria to improve tracking, validated through simulations using a mass-spring-shock system.

3.5.1.5. Neural Controllers

Artificial neural network-based controllers (ANNCs) can learn and adapt to changing conditions by adjusting their internal weights. These controllers help handle nonlinear and complex systems. A review of the impact of machine learning on control systems shows that data-driven models can replace or complement classical components of the control loop [89]. A case study demonstrates that a deep neural network enhances the performance of a surface-based autonomous vehicle, improving resilience against modeling and noise errors and outperforming the traditional linearizing controller.

A proposed robust predictive control strategy utilizes ANNC in two stages: one for modeling the plant and the other for residual uncertainty, achieving stability and improved robustness in a pneumatically actuated servomechanism validated experimentally [5]. Another approach introduces the learning feedback linearization (LFL) method, where a single NARMA-type MLP neural network learns to linearize the input of nonlinear systems, enabling subsequent control using a classical PI controller [28]. Simulations validate this method successfully.

A different controller is based on a fuzzy cerebellar model modified with a functional link network (FLMFC), combined with a sliding mode controller, and is used to synchronize chaotic systems and control an inverted pendulum, demonstrating high robustness to perturbations in simulations [8].

In the area of electric motors, an ANNC for a three-phase induction motor shows improved speed tracking and response to disturbances compared to a fuzzy controller [106]. For electric vehicle charging stations, an intelligent adaptive control system based on an ANN-PMC neural controller improves energy management, reducing overshoot and DC bus settling time [121].

Finally, two intelligent control strategies based on reinforcement learning with deep neural networks (RL-DLNN) have been applied to chiller-type HVAC systems, demonstrating superior adaptation, stability, and tracking compared to PID and MPC controllers, particularly in highly nonlinear and disturbed MIMO systems [216].

Table 2 presents the articles that contributed to control models in control systems-based AI.

In summary, Figure 4 presents a classification of the groups of control models that contribute to the application of AI in control systems.

3.5.2. Result of Contributions: Optimization of Parameters

New AI optimization methods have led to significant improvements in performance, adaptability, and efficiency in the design and implementation of intelligent controllers for control systems.

Below are the articles that contribute to optimization models:

3.5.2.1. Artificial Neural Networks

Researchers utilize artificial neural networks (ANNs) to enhance optimization and learning in controllers, while machine learning facilitates parameter tuning and improves the accuracy of control models in complex systems. ANNs effectively handle nonlinear systems, thereby increasing the accuracy and robustness of controllers and other complex systems. They enable the handling of nonlinear systems, thereby improving the accuracy and robustness of controllers and other complex systems. ANNs are applied in the design of PID controllers, providing an advanced approach to enhance the system response [48].

On the other hand, support vector machines (SVMs) are used for excitation control, improving system stability through supervised learning [7]. Supervised learning neural networks are used for the learning and verification of feedback control systems [49]. Machine learning and predictive control techniques are applied to design an adaptive PID controller [72].

3.5.2.2. Evolutionary Algorithms

Research uses evolutionary algorithms, such as genetic algorithms (GAs), to optimize control parameters, particularly in systems with uncertainty or nonlinearities. These algorithms search for optimal solutions by simulating natural evolutionary processes.

GAs can refine the rule base of a Mamdani FLC [51], improve the performance of a modified PID controller [13], and enhance the adaptability of a self-adjusting PID controller [56].

Multi-objective GAs are crucial to improve controller performance across several competing criteria [78]—for example, a fractional-order PID controller for multivariable systems [58]. In addition, GAs can optimize control structures focusing on structural adaptation [81].

For vibration suppression, GA has been used to fine-tune fuzzy controllers [61]. A hybrid controller based on the Hydra structure has been developed and optimized using a GA-based strategy [52]. Additionally, the Taxi-Cab Evolutionary Algorithm has been used for multi-objective optimization in control systems [29].

3.5.2.3. Metaheuristic Optimization

Metaheuristic algorithms, such as Particle Swarm Optimization (PSO), can be applied to adjust controller parameters in complex systems. These methods are effective in exploring large search spaces to identify optimal solutions.

PSO has been used to tune a PID controller [62], to design a Fuzzy PID controller for a Quasi-Z Source converter [64], to optimize a Fuzzy-PDC controller [65], and to improve fuzzy controllers in distillation processes [99].

Grey wolf optimization optimizes a Fuzzy PID controller for frequency regulation in power systems [71]. The gravitational search algorithm improves the dynamic response of a PID controller [59], while the cuckoo optimization algorithm could manage the energy in a fuzzy controller [68].

Differential evolution algorithms (DEAs) design a fuzzy controller for wireless sensor networks [93]. At the same time, the Firefly Algorithm develops an optimized Fuzzy PID driver [23], and the bee colony algorithm (BCA) is a good method to design a fuzzy controller [75].

3.5.2.4. Fuzzy Controllers

Articles in this category focus on the design of fuzzy controllers, which use fuzzy logic to manage uncertainty and variability in complex systems.

A Fuzzy PID controller can improve continuous industrial processes [74] as a multivariable controller [66] and control underactuated manipulators [70]. The authors of [53] proposed a low-cost servo system based on a fuzzy controller, while [60] proposed a BLDC motor control based on a Fuzzy PID controller.

3.5.2.5. Hybrid Techniques

Studies in this category combine two or more techniques, such as fuzzy logic and metaheuristic optimization, to take advantage of the strengths of each. This combination improves the performance and adaptability of controllers.

Fuzzy control and optimization techniques have been integrated for frequency regulation in electrical networks [69]. Researchers have implemented a hybrid approach GWO-SCA in a Type II fuzzy controller for frequency control in multi-area systems [95]. A combination of harmony search and cuckoo optimization has been applied in a Fuzzy PID driver [98]. A hybrid PSO-GSA strategy has also been employed in a fuzzy sliding-mode controller for wind power systems [84].

3.5.2.6. Predictive Control

In this category, techniques focus on the use of predictive control, which allows us to anticipate changes in the system and adjust the controller based on predictions, improving performance in systems with high variability. Support vector machines (SVMs) are used in a nonlinear predictive controller to anticipate and correct system behavior in multiple steps forward [19].

3.5.2.7. Other Techniques

This category encompasses articles that employ unique or uncommon approaches and that do not fall within the above categories. These studies explore alternative optimization and control techniques. The researchers developed an optimized PID controller using multi-objective techniques, offering a method to manage multiple performance criteria in a control system [55].

Table 3 presents the articles that have contributed to the optimization of artificial intelligence in control systems.

Figure 5 also presents a classification of the groups of optimization strategies that contribute to the application of AI in control systems.

3.5.3. Result of Contributions: Adaptability

The development of adaptive control strategies has gained significant attention in recent decades due to the increasing complexity of dynamic systems, the presence of nonlinearities, time delays, parametric uncertainties, and external disturbances. In this context, several approaches have been proposed that combine classical control structures—such as PID controllers, fuzzy models, and neural networks—with modern techniques of adaptation, optimization, and machine learning.

The reviewed literature has been organized into different categories according to the predominant control technique, which allows a clear comparison of the strengths, limitations, and applications of each approach. The main categories are:

Adaptive PID-based controllers, where the classical PID structure is enhanced through optimization, fuzzy logic, or machine learning.

Adaptive Fuzzy-based controllers (Fuzzy / ANFIS / T-S), which employ fuzzy inference to support decision-making and online parameter tuning.

Adaptive Neural Network-based controllers, where neural models act as universal approximators to handle uncertain nonlinear dynamics.

Hybrid adaptive controllers, which synergistically combine fuzzy, neural, robust, or bio-inspired approaches.

Optimization and adaptive optimal controllers, where evolutionary algorithms, heuristics, and predictive methods are used to tune parameters of the classical or advanced controllers.

3.5.3.1. Adaptive PID-Based Controllers

An Adaptive Particle Swarm Optimization (APSO) to tune linear and nonlinear PID controllers, improving stability and convergence speed compared to classical methods [110].

An adaptive PID controller based on fuzzy predictors is introduced, where fuzzy modelling improves adjustment of online and tracking accuracy in nonlinear processes [111].

A MRAS (Model Reference Adaptive System) approach is applied to tune PID gains for launch vehicles, ensuring robust performance under parameter variations during flight [115].

A self-adaptive Genetic Algorithm (GA) is presented for PI/PID controllers, optimizing both gains and performance indices for faster and smoother responses [118].

A Fuzzy-PID controller implemented in a PLC is developed for oil pipeline flow regulation, automatically adjusting the PID parameters through fuzzy inference [122].

A Kalman filter-enhanced Fuzzy-PID controller is proposed for liquid level control in conical tanks, where fuzzy logic adjusts PID gains and Kalman filtering improves robustness against noise [124].

A hybrid OENN–OEANFIS system is designed to adapt PID parameters, combining endocrine-inspired neural networks with neuro-fuzzy inference for enhanced adaptability [104].

A Self-Adaptive Differential Evolution (SaDE) algorithm is applied for adjusting PID, dynamically adjusting the crossover and mutation rates to achieve optimal gains [215].

A deep learning-based intelligent PID controller (CNN–LSTM) is proposed for UAVs, where a neural network predicts parameter adjustments to improve stability and energy efficiency [212].

An AIEM-DDPG (Ambient Intelligence Exploration Multi-Delay Deep Deterministic Policy Gradient) is introduced to adapt the PI controller gains for PEM fuel cells, achieving improved stability and reduced overshoot [211].

A TD3 (Twin Delayed Deep Deterministic Policy Gradient)-based tuner is applied to the PI and Lead controllers, providing adaptive adjustment of parameters with improved robustness against uncertainties [214].

3.5.3.2. Adaptive Controllers Based on Fuzzy Logic (Fuzzy / ANFIS / T-S)

Control systems frequently adopt fuzzy logic-based adaptations due to their ability to handle uncertainty and variability under changing operating conditions. Fuzzy logic enables controllers to make precise adjustments without requiring an exact mathematical model of the system, making it well-suited for environments with complex or unpredictable dynamics.

An adaptive fuzzy controller for MPPT in photovoltaic systems, where the output scaling factor is adjusted online to efficiently track the maximum power point under varying irradiance and temperature [6].

An ANFIS controller applied to a PFC boost converter, trained online using data from a conventional fuzzy controller to improve voltage regulation and power factor [18].

An adaptive fuzzy PI controller for coupled tanks, where Mamdani rules and membership functions dynamically tune Kp and Ki gains to handle nonlinearities [31].

A fuzzy T–S adaptive controller with a state observer, designed for nonlinear systems with unknown dead zones, avoiding traditional backstepping complexity [32].

An ANFIS controller for autonomous mobile robots, trained with hybrid learning (least squares + backpropagation) to achieve trajectory tracking and target pursuit [76].

A self-learning fuzzy controller based on Autonomous Adaptive Control (AAC), which generates and updates fuzzy rules online from input–output data without external supervision [102].

An ANFIS optimized with Genetic Algorithms for greenhouse climate control, where GA tunes the membership functions and rules to minimize ISE and improve robustness [108].

A two-level adaptive fuzzy controller for DC motors, where the first level is a Mamdani fuzzy controller and the second is an inverse T–S model that updates the first in real time [123].

3.5.3.3. Adaptive Controllers Based on Bioinspired / Optimal Optimization

An adaptive fuzzy controller optimized with Adaptive Differential Evolution (FADE), applied to synthetic inertia in power systems, where DE dynamically tunes fuzzy membership functions to reduce the frequency nadir and improve stability [120].

An LQR controller optimized with Adaptive Cuckoo Search (ACS), applied to active suspension systems in vehicles, where ACS adjusts the weighting matrices Q and R to reduce vertical acceleration and improve ride comfort [213].

An Adaptive Model Predictive Control (MPC) scheme with set-membership identification, designed for linear systems with parametric uncertainty and disturbances, using tube-based predictions and convex optimization to guarantee constraint satisfaction [125].

3.5.3.4. Adaptive Controllers Based on Neural Networks

An Initial-Training-Free Online Extreme Learning Machine (ITF-OELM) is proposed for the adaptive control of nonlinear discrete-time systems with rapidly varying parameters, allowing online identification without prior training data [109].

A multilayer neural network (MLP) combined with integral sliding mode control is used as a direct adaptive controller to compensate for residual and approximation errors in systems with dead-zone nonlinearities [37].

An adaptive RBF neural network (RBFNN) is applied to nonlinear systems with unknown time delays, acting as a universal approximator with online weight adaptation laws derived from Lyapunov analysis [27].

An adaptive RBFNN integrated with second order sliding mode control estimates unknown dynamics and disturbances, ensuring robust trajectory tracking while eliminating chatter [101].

An RBFNN combined with backstepping and barrier Lyapunov functions (BLF) is developed for vibration suppression in flexible string systems with dead-zone nonlinearities and output constraints [114].

An adaptive RBFNN combined with sliding mode control and a neuro-inspired IAP mechanism (Incentive–Actuator–Preventor) is applied to robotic manipulators, enhancing robustness and trajectory tracking under switching constraints [103].

An observer of the Wavelet Neural Network (WNN) is integrated into an adaptive fuzzy sliding mode stabilizer for multimachine power systems, estimating uncertainties and improving low-frequency oscillation damping [119].

An adaptive neural network controller for DC–DC converters (based on [126]) uses online learning to compensate for model uncertainties and improve voltage regulation under dynamic load changes [126].

3.5.3.5. Adaptive Control based on Hybrid Techniques

A neuro-fuzzy backstepping adaptive controller applied to an induction motor with unified nonlinear friction and unknown model dynamics, where ANFIS is integrated into a robust backstepping design to ensure tracking accuracy [90].

A robust adaptive fuzzy–neural T-S controller for interconnected nonlinear systems, where a Takagi–Sugeno fuzzy model works with neural networks to approximate unknown interconnections, and a compensator handles approximation errors [117].

A hybrid adaptive–fuzzy navigation controller for mobile manipulators, combining adaptive control for dynamic parameter estimation with fuzzy logic (T-S rules) for obstacle avoidance and trajectory tracking in dynamic environments [128].

A robust adaptive controller based on Lyapunov/backstepping techniques (article [127]), designed to handle strong nonlinearities and uncertainties while ensuring global stability through adaptive update laws and robust compensation [127].

A fuzzy-PID hybridized with ant colony optimization (ACO) for UAV path planning, where fuzzy logic adjusts PID gains in real time, and improved ACO optimizes pheromone updating to generate shorter and smoother trajectories [210].

Table 4 presents a detailed table of the articles that contributed to the adaptability in artificial intelligence in control systems. Figure 6 also presents a classification of the adaptability groups that contribute to the application of AI in control systems.

3.5.4. Identification of Gaps and Controversies in the Models

Studies on advanced controllers have multiple limitations in their applicability and replicability in real environments, raising doubts about their practical effectiveness. One of the main shortcomings is the reliance on simulations for validation, rather than experimental testing in real-world environments, which limits the evaluation of their performance against factors such as noise, wear, and extreme variability. In addition, the lack of detail in parameter configuration, especially in fuzzy controllers, hinders the replication and adaptation of these methods in other systems, affecting consistency and accuracy under conditions of high uncertainty.

Most studies do not compare these approaches with other advanced methods, such as deep neural networks and modern optimization algorithms, making it difficult to fully assess their competitiveness. Additionally, the high computational complexity of some algorithms—particularly in hybrid implementations—poses a significant challenge for their deployment in embedded systems and critical applications. This issue becomes more severe due to the limited guidance available for parameter tuning and result interpretation, which further hinders their adoption in industrial settings.

In addition, many studies do not evaluate the robustness of controllers under harsh conditions or consider their scalability in large-scale applications. The lack of low-cost hardware testing and the complexity of design in fuzzy logic are additional barriers, as the industry often prefers more straightforward solutions. In conclusion, deficiencies in experimental validation, replicability, and scalability raise questions about the effectiveness of these advanced controllers in real industrial applications.

4. Conclusions

Main Findings

Studies on advanced control strategies report notable improvements in adaptability, accuracy, efficiency, and optimization. Controllers based on Type-2 fuzzy logic, neural networks, and hybrid configurations—such as fuzzy-PID systems or those combined with optimization algorithms such as PSO, GA, and GWO—enable adaptive responses to variability in complex systems. These approaches have proven effective in applications that include robotic manipulators, heat exchangers, greenhouses, and renewable energy systems, where they contribute to maintaining stability and precision in dynamic operating environments.

Optimization using evolutionary and metaheuristic algorithms (gravitational search, Cauchy mutation, PSO, GA) allows automatic and precise tuning of PID and fuzzy controller parameters. These techniques eliminate the need for continuous manual adjustments, facilitating rapid convergence to optimal configurations, especially in nonlinear systems and in applications with high uncertainty.

In terms of energy efficiency and resource management, adaptive and predictive controllers have proven to be effective in renewable energy systems, such as photovoltaics and wind turbines, maximizing generation under variations in irradiance and temperature. In applications such as greenhouses, optimized controllers help maintain stable conditions, minimizing water and energy usage, and optimizing resource utilization.

The validation of these controllers in simulations and controlled environments has shown improvements in stability, reduction of oscillations, and optimization of energy use in the systems studied. However, some studies still lack experimental testing under real-world conditions, it is challenging to evaluate their performance in industrial applications.

Fuzzy drivers, especially in hybrid configurations with optimization algorithms, achieve an optimal balance between stability and accuracy. These methods are highly effective in systems with variability and help improve robustness in power grids and systems with multiple energy sources. Hybrid approaches (such as GWO-SCA and PSO-GA) combine the strengths of different algorithms to overcome limitations and improve performance in exploration and exploitation.

Finally, the integration of optimization techniques with adaptive control (such as the use of genetic algorithms and the Kalman filter) allows for more precise real-time tuning, essential for embedded or real-time systems, such as DC motor control and flow in pipelines. Experimental validation in industrial applications, such as climate control and flow tracking, highlights the practical feasibility and potential for adoption of these advanced controllers in sectors that require high accuracy and adaptability.

Key Trends in Driver Development:

Recent research on advanced controllers has revealed a clear trend toward integrating multiple control techniques to enhance robustness and accuracy in complex, dynamic systems. The combination of fuzzy controllers, neural networks, and PID algorithms seeks to leverage the adaptability, learning capability, and predictability of each approach. These integrated strategies have demonstrated great effectiveness in applications characterized by significant uncertainty, such as robotics, renewable energy systems, and temperature regulation.

The optimization of parameters through evolutionary and metaheuristic algorithms (such as gravitational search, PSO, and Cauchy mutation) is a central technique in these developments, as it allows PID and fuzzy controllers to be adjusted quickly and accurately, optimizing efficiency and adaptability to changes in system conditions. Hybrid approaches, which combine two or more optimization algorithms, are increasingly used to enhance parameter tuning, leading to improved accuracy and stability.

The applications in sustainable energy stand out, with controllers designed to maximize energy capture in photovoltaic and wind systems under changing conditions. In agriculture, diffuse controllers enable the optimization of water and energy consumption in greenhouses, promoting sustainability.

The trend towards robust and adaptive controllers, especially those of the diffuse type II, is key in systems with high uncertainty. These controllers enable the handling of variability in applications that require high precision, such as power systems and robotics. Intelligent adaptive controllers, which integrate fuzzy logic and optimized neural networks, provide real-time responses and continuously adapt, representing a significant advancement for industrial applications where real-time stability and adaptability are crucial.

Simulations and comparisons with traditional methods demonstrate that advanced methods offer improved accuracy, stability, and energy efficiency compared to conventional control methods. In autonomous systems and sensors powered by ambient energy, optimization is crucial for efficiently managing energy, thus prolonging the useful life of IoT and monitoring devices in environments with low energy availability.

Ultimately, interest in nonlinear and high-uncertainty systems is driving the development of robust adaptive controllers, which can maintain stability and optimizing performance under adverse conditions. These systems are especially useful in applications such as industrial process control, mobile robotics, and power systems, where variability is frequent.

Identified Research Gaps:

Advanced controllers have identified several challenges and limitations that hinder their application in real-world conditions. One of the most critical shortcomings is the limited experimental validation in actual operating environments; most evaluations rely on simulations, which restrict insights into practical performance and raise concerns about their effectiveness in industrial scenarios. To confirm the benefits observed in the simulations, researchers must perform tests under real industrial conditions.

Another significant challenge is the lack of detail in the configuring and tuning of the parameters. Studies on fuzzy controllers and evolutionary algorithms often lack sufficient descriptions of the configuration of membership functions and fuzzy rules. Additionally, the configuration of parameters in optimization algorithms, such as population size and crossover rates, lacks clear guidelines, which limit replicability and adaptability across different environments.

Limited comparisons with other advanced control methods are another significant gap. Many studies only contrast their methods with conventional controllers, without considering advanced alternatives such as deep neural networks or predictive controllers. Comprehensive comparisons would help to comprehensively assess the advantages and limitations of each approach, highlighting its effectiveness in specific contexts.

Scalability and applicability are also issues that have a waiting list in more complex systems. Although controllers perform well in simulations, their performance in larger and more complicated systems, such as renewable energy plants and large-scale greenhouses, remains inadequately evaluated. Likewise, there is a lack of computational load assessment, a critical aspect for real-time applications and embedded systems.

Additionally, the complexity involved in designing fuzzy logic systems and configuring rule sets creates a barrier to their implementation in industrial applications, which require simplicity and ease of maintenance. Many studies overlook extreme test scenarios, instead focusing on standard operating conditions without assessing how controllers respond to significant disturbances or substantial load variations.

Finally, insufficient technical descriptions and configuration details affect the replicability of studies, especially in neuro-fuzzy and optimization methods. Detailed documentation of parameters and settings is crucial for facilitating the application of these approaches in new areas and ensuring their reproducibility.

Contribution of the study:

The reviewed studies consolidate a solid foundation in the field of advanced controllers, bringing significant advances in several key areas. Hybrid adaptive and predictive control techniques, which combine fuzzy logic, PID controllers, and neural networks, have enhanced adaptability and accuracy in systems characterized by high variability and nonlinearity, such as robotics and renewable energy systems. The introduction of fuzzy type-2 and predictive controllers based on neural networks expands the capabilities to handle uncertainties, setting a framework for future research into more robust and scalable controllers.

Real-time parameter optimization using evolutionary and metaheuristic algorithms (such as PSO, GA, and GWO) enables automatic and accurate tuning, thereby increasing efficiency and response speed in high-variability applications. These advances lay the groundwork for integrating machine learning techniques and creation of adaptive controllers that continuously adjust to changing conditions without human intervention.

In practical applications, researchers have developed optimized controllers that improve energy efficiency in photovoltaic and wind systems, as well as improve resource management in precision agriculture. These studies also emphasize the positive impact on sustainability by reducing water and energy consumption. Future research can extend these efforts to design controllers for other emerging areas, such as advanced agriculture and critical industrial sectors.

Strengthening the theoretical foundations for managing uncertainty through fuzzy logic and neural networks provides a solid basis for advancing adaptive control, promoting the use of artificial intelligence to enhance robustness and adaptability in complex systems. Despite significant progress, essential gaps persist in the experimental validation of these approaches under real-world conditions. Since most studies rely on simulations, conducting tests in actual operating environments remains critical to confirm the effectiveness of these controllers in industrial and commercial applications.

The computational complexity and processing load in real-time applications present an ongoing challenge. Several studies emphasize the need to optimize control algorithms for effective deployment in embedded and low-power systems. By integrating optimization techniques such as genetic algorithms and the Kalman filter, researchers enable controllers to rapidly adapt to environmental changes—an essential capability for industrial sectors that require dynamic responses in real-time.

In Finally, areas for improvement are identified in exhaustive comparisons with other advanced methods and in the robustness of controllers under extreme conditions, highlighting the need for studies that evaluate the comparative effectiveness of optimization approaches and develop controllers capable of operating reliably under significant disturbances. These contributions lay a foundation for future research in intelligent control applications, opening doors to new techniques and methods in sectors such as Industry 4.0 and advanced automation.

Future work:

Research on adaptive, predictive, and optimized controllers has identified several promising directions to improve their applicability, scalability, and robustness in real-world scenarios. A key priority is the experimental validation of these systems under actual operating conditions, as most existing studies rely primarily on simulations. Conducting tests on physical systems—such as renewable energy platforms, industrial robotics, and agricultural applications—would enable researchers to evaluate performance in realistic environments and accelerate the adoption of these controllers in industrial settings.

A detailed comparison with other advanced methods, such as deep neural networks and optimization algorithms, would allow us to determine the effectiveness of these approaches in various contexts and select the most appropriate techniques. In addition, exploring adaptive and evolutionary optimization algorithms for continuous tuning is crucial to enhancing the adaptability of controllers under varying conditions, particularly in applications such as renewable energy and advanced robotics.

The development of fuzzy type-2 controllers combined with machine learning is another key area to improve the handling of uncertainties in systems with high variability, such as industrial processes and collaborative robotics. It also highlights the need to study scalability in large-scale applications, such as power plants and large greenhouses, and to implement distributed control networks to improve the management of complex systems.

To facilitate replicability, this study recommends creating detailed guides and adjustable libraries for driver configuration, particularly for fuzzy logic and PID, as well as accessible libraries for testing in various applications. The application in new fields of control and resource optimization, such as smart cities and advanced manufacturing, highlights the potential of these controllers to improve energy efficiency and sustainability in multiple sectors.

Other areas of development include advanced simulation to evaluate controller performance in critical situations without the risks associated with physical testing, as well as research into low-computational-load controllers for real-time applications and IoT devices. Additionally, enhancing the interpretability and maintainability of controllers is crucial for use in industries where transparency and end-user understanding are paramount.

Researchers emphasize the importance of robustness and resilience under extreme conditions, as well as the use of automation in designing fuzzy logic and configuring rules and integrating prediction techniques to anticipate system changes. These strategies maximize the adaptability and responsiveness of controllers in applications such as advanced robotics and energy systems. Pursuing these research directions will lay the groundwork for significant advances in the adoption of optimized controllers in various industrial and technological sectors, thus expanding their impact in critical and emerging fields.