Ai-Based Control of Unified Power Quality Conditioners for Power Quality Enhancement in Modern Distribution Networks: A Comprehensive Review

Ai-Based Control of Unified Power Quality Conditioners for Power Quality Enhancement in Modern Distribution Networks: A Comprehensive Review

Ravinder Kaur , Samreet Gosal

GNDEC,Ludhiana

Abstract – The rapid development of renewable-energy generation, electric-vehicle charging infrastructure, distributed energy resources, battery energy-storage systems, and power-electronic loads has fundamentally changed the operating characteristics of modern electrical distribution networks. Although these technologies provide important benefits in terms of energy efficiency, flexibility, and decarbonization, their extensive use has introduced significant power-quality challenges, including voltage sag, voltage swell, harmonic distortion, reactive-power demand, voltage imbalance, current distortion, and rapid load variations. The Unified Power Quality Conditioner (UPQC) is one of the most comprehensive custom-power devices for mitigating these disturbances because it combines the functions of a series active power filter and a shunt active power filter.

Conventional UPQC controllers, particularly proportional-integral (PI), synchronous-reference-frame, and instantaneous- power-theory-based controllers, have demonstrated reliable operation under many conditions. However, their performance may deteriorate when system parameters vary, nonlinear loads change rapidly, renewable-energy generation fluctuates, or multiple disturbances occur simultaneously. Consequently, artificial intelligence (AI) and machine-learning techniques have increasingly been investigated to improve the adaptability and dynamic performance of UPQC systems.

This paper presents a comprehensive review of AI-based UPQC control techniques, with particular emphasis on artificial neural networks (ANN), fuzzy logic control (FLC), adaptive neuro-fuzzy inference systems (ANFIS), optimization-assisted AI, reinforcement learning, and emerging deep-learning approaches. The review examines UPQC configurations, compensation principles, control architectures, reference-generation methods, converter technologies, and applications in renewable-energy-integrated distribution systems. Recent studies demonstrate the growing use of intelligent UPQC controllers in photovoltaic systems, hybrid renewable-energy systems, microgrids, EV charging systems, and other converter-dominated networks. A 2026 scoping review of UPQC research identified 51 studies from 1,700 records and reported that intelligent methods such as fuzzy logic, ANN, and hybrid approaches are increasingly important, although experimental validation and long-term stability remain significant research gaps.

The review concludes that AI can improve UPQC adaptability and disturbance rejection, but practical deployment requires attention to training-data quality, computational latency, stability certification, interpretability, cybersecurity, hardware implementation, and generalization under unseen grid conditions. Future research should therefore move from simulation-only demonstrations toward real-time and hardware-validated, physics-informed, explainable, and safety- constrained AI-UPQC systems.

Keywords: Unified Power Quality Conditioner, UPQC, Artificial Intelligence, Machine Learning, ANFIS, Artificial Neural Network, Fuzzy Logic, Power Quality, Harmonic Compensation, Renewable Energy, Smart Grid, Microgrid.

1. INTRODUCTION

   Electrical distribution systems are undergoing a major transformation due to the increasing penetration of renewable-energy sources, distributed generation, battery energy-storage systems, electric vehicles, data centers, adjustable-speed drives, and other power-electronic loads. These developments are changing both the magnitude and dynamic characteristics of electrical power flows.

   Power-quality problems that were traditionally associated with industrial nonlinear loads are now increasingly influenced by inverter-based renewable generation and electronically controlled loads. Harmonics, voltage sag, voltage swell, reactive power, imbalance, flicker, and fast voltage variations can negatively affect sensitive equipment, increase losses, reduce system efficiency, and compromise reliable operation.

   The Unified Power Quality Conditioner has emerged as an important solution because it can simultaneously compensate voltage- and current-related disturbances. A conventional UPQC consists of two voltage-source converters sharing a common DC link. The series converter compensates supply-voltage disturbances, whereas the shunt converter compensates load-current disturbances.

   The increasing complexity of distribution networks, however, creates challenges for conventional UPQC controllers. Fixed-gain controllers are generally designed around specific operating points. Their performance can degrade when the system experiences rapid load changes, nonlinearities, renewable intermittency, parameter uncertainty, or multiple simultaneous disturbances.

   Artificial intelligence offers a possible solution. ANN controllers can learn nonlinear relationships from data, fuzzy controllers can represent expert knowledge through linguistic rules, and ANFIS combines the learning capability of neural networks with fuzzy inference. More recent research has also investigated optimization-assisted AI, reinforcement learning, and deep learning.

   A 2025 review of hybrid renewable-energy systems and UPQC integration specifically identified PI, ANN, fuzzy-PID, ANFIS, and fuzzy-PI controllers as important alternatives for improving the dynamic response of UPQC systems under parameter variation and nonlinear load disturbances.

   More recently, a 2026 scoping review analyzed 51 UPQC studies and found that classical PI and synchronous-reference-frame methods remain common, while intelligent and hybrid methods show promising power-quality improvements. It also highlighted experimental validation, long-term stability, and hybrid microgrid operation as important unresolved areas.

   Therefore, a focused review of AI-based UPQC control is timely.

2. OBJECTIVES OF THE REVIEW

   The main objectives of this review are:

   1. To explain the operating principle and major configurations of UPQC.

   2. To review conventional UPQC control techniques.

   3. To examine AI-based UPQC control strategies.

   4. To compare ANN, FLC, ANFIS, optimization-assisted, and reinforcement-learning approaches.

   5. To investigate UPQC applications in renewable-energy systems.

   6. To examine intelligent UPQC applications in microgrids and EV charging.

   7. To identify limitations of existing AI-UPQC research.

   8. To identify research gaps suitable for future M.Tech/PhD research.

   9. To propose future directions for practical AI-based UPQC implementation.

3. REVIEW METHODOLOGY

   A literature review should not simply list previously published papers. It should identify trends, compare methodologies, evaluate limitations, and determine research gaps.

   For this review, the literature is organized into the following categories:

   • UPQC topology and configuration.

   • Conventional control.

   • Fuzzy control.

   • ANN control.

   • ANFIS control.

   • Optimization-assisted intelligent control.

   • Renewable-energy-integrated UPQC.

   • Microgrid applications.

   • EV charging applications.

   • Hardware and real-time validation.

   • Future AI approaches.

   Recent literature from 2025 and 2026 was given particular attention because of the rapid development of AI-assisted converter control.

   A recent scoping review searched major databases and reported 1,700 initial records, of which 51 studies satisfied its inclusion criteria. The review found that most research remains focused on grid-connected microgrids and that harmonics, voltage sag, and power-factor correction are among the most frequently addressed disturbances.

4. POWER-QUALITY PROBLEMS IN MODERN DISTRIBUTION NETWORKS

   1. Voltage Sag

      Voltage sag is a temporary reduction in RMS voltage. It may be caused by:

      • Short circuits.

      • Large motor starting.

      • Transformer energization.

      • Sudden load increases.

      • Faults in adjacent feeders.

        Voltage sag can interrupt sensitive industrial processes and electronic equipment.

   2. Voltage Swell

      Voltage swell is a temporary increase in RMS voltage. It may result from:

      • Sudden load removal.

      • Single-line-to-ground faults.

      • Capacitor switching.

      • Incorrect voltage regulation.

        The series converter of a UPQC can compensate both sag and swell by injecting an appropriate voltage.

   3. Harmonic Distortion

      Nonlinear loads draw nonsinusoidal current. The current can be represented as:

      [

      i(t)=I_1(t+_1)+

      _{h=2}^{}I_h(ht+_h)

      ]

      The current THD is:

      [

      THD_I =

      ]

      Harmonic distortion can produce:

      • Additional losses.

      • Transformer heating.

      • Neutral-current increase.

      • Resonance.

      • Electromagnetic interference.

      • Reduced equipment lifetime.

   4. Reactive Power

      Inductive loads require reactive power:

      [ Q=VI

      ]

      The shunt converter can provide compensating reactive current and thereby reduce reactive current drawn from the grid.

   5. Voltage and Current Imbalance

      Unbalanced loads create negative- and zero-sequence components. These components can produce:

      • Motor heating.

      • Unequal phase currents.

      • Increased losses.

      • Converter stress.

        Advanced UPQC controllers can be designed to compensate these components.

5. UNIFIED POWER QUALITY CONDITIONER

   1. Basic Configuration

      The conventional UPQC consists of:

      1. Series voltage-source converter.

      2. Shunt voltage-source converter.

      3. Common DC-link capacitor.

      4. Series coupling transformer/filter.

      5. Shunt coupling filter.

      6. Measurement and control system.

      The series converter is connected in series with the distribution feeder. The shunt converter is connected in parallel with the load.

      The DC link provides the energy exchange mechanism between the converters.

   2. Series Converter

      The series converter generates an injection voltage: [

      V_{inj}=V_{ref}-V_s

      ]

      where:

      • (V_{ref}) = desired load voltage.

      • (V_s) = supply voltage.

        The load voltage becomes:

        [

        V_L=V_s+V_{inj}

        ]

        Therefore, the series converter attempts to maintain: [

        V_LV_{ref}

        ]

   3. Shunt Converter

      The shunt converter generates compensating current: [

      I_c=I_L-I_s^*

      ]

      where (I_s^*) represents the desired source current. The shunt converter therefore compensates:

      • Harmonic current.

      • Reactive current.

      • Negative-sequence current.

      • Other undesirable current components.

6. UPQC TOPOLOGIES

   Several UPQC variants have been investigated.

   1. Conventional UPQC

      The conventional configuration uses independent series and shunt converters connected to a common DC link. Advantages

      • Simultaneous voltage/current compensation.

      • Mature technology.

      • Flexible control.

        Limitations

      • High converter rating.

      • Higher cost.

      • Multiple semiconductor switches.

      • Complex control.

   2. Open-UPQC

      Open-UPQC configurations modify the power-flow structure and can reduce certain converter requirements.

      Recent research has investigated open-UPQC structures using fuzzy and ANFIS controllers in PV-integrated systems for voltage sag, swell, and harmonic compensation.

   3. PV-UPQC

      In PV-UPQC systems, photovoltaic generation can be connected to the common DC link. This provides two functions:

      1. Power-quality compensation.

      2. Renewable-power injection.

      Recent research has investigated PV-UPQC systems using advanced control approaches including ANFIS and sliding-mode control.

   4. UPQC with Battery Storage

      Battery energy storage can support the DC link and improve compensation during renewable-energy fluctuations. The architecture can therefore combine:

      [ PV+BESS+UPQC

      ]

      This is particularly attractive for microgrids.

   5. Multi-Converter UPQC

      Advanced systems may contain multiple converters for:

      • Distributed generation.

      • Storage.

      • EV charging.

      • Power-quality compensation.

        Such architectures increase flexibility but also increase control complexity.

7. CONVENTIONAL UPQC CONTROL METHODS

   1. PI Control

      The PI controller is:

      [

      u(t)=K_pe(t)+K_ie(t)dt

      ]

      Advantages

      • Simple.

      • Low computational cost.

      • Easy implementation.

      • Well understood.

        Limitations

      • Fixed gains.

      • Sensitive to parameter variation.

      • Limited nonlinear adaptation.

      • Requires tuning.

        PI controllers remain widely used in UPQC systems. The 2026 scoping review found classical PI and synchronous-reference- frame approaches to be among the most common control methods in the literature.

   2. Synchronous Reference Frame Control

      The (abc) variables are transformed into (dq) coordinates.

      This separates active and reactive components and simplifies controller design. Advantages

      • Effective for balanced systems.

      • Mature.

      • Relatively simple.

        Limitations

      • Sensitive to synchronization.

      • Performance can deteriorate under distorted/unbalanced conditions.

      • Requires coordinate transformations.

   3. Instantaneous Power Theory

      The (p-q) theory calculates instantaneous active and reactive power. For ()-() variables:

      [

      p=v_i_+v_i_

      ] [

      q=v_i_-v_i_

      ]

      Compensating current references can then be calculated. Advantages

      • Fast reference generation.

      • Suitable for nonlinear loads.

      • Widely used in active filtering.

        Limitations

      • Performance depends on voltage conditions.

      • Distorted and unbalanced voltages can complicate compensation.

8. ARTIFICIAL INTELLIGENCE FOR UPQC CONTROL

   Artificial intelligence provides an alternative to fixed mathematical control laws. The principal AI approaches are:

   • Fuzzy logic.

   • ANN.

   • ANFIS.

   • Optimization-assisted AI.

   • Reinforcement learning.

   • Deep learning.

9. FUZZY LOGIC CONTROL

   Fuzzy logic does not require an exact mathematical model. A typical controller uses:

   [

   e=V_{ref}-V

   ]

   and:

   [

   e=e(k)-e(k-1)

   ]

   The inputs are converted into linguistic variables such as:

   • Negative Large.

   • Negative Small.

   • Zero.

   • Positive Small.

   • Positive Large.

     Rules determine the controller output.

     Advantages

   • Good nonlinear behavior.

   • No precise mathematical model required.

   • Robust to parameter variation.

     Limitations

   • Rule design is subjective.

   • Membership-function selection affects performance.

   • Scaling to complicated systems can be difficult.

10. ARTIFICIAL NEURAL NETWORK CONTROL

    ANNs learn the nonlinear relationship between input variables and control actions. A neuron is represented by:

    [

    y=f(_{i=1}^{n}w_ix_i+b)

    ]

    An ANN-UPQC controller can receive:

    • Voltage error.

    • Current error.

    • DC-link error.

    • (d)-axis current.

    • (q)-axis current.

    • Harmonic components.

    The network produces the compensation reference.

    A 2025 study proposed an ANN-based solar-PV-UPQC system using an adaptive leaky least-mean-square switching approach and compared its performance with conventional PI and fuzzy controllers.

11. ADAPTIVE NEURO-FUZZY INFERENCE SYSTEM

    ANFIS is one of the most frequently investigated intelligent methods for UPQC control. It combines:

    Neural learning + fuzzy reasoning

    The typical architecture contains five layers. Layer 1: Fuzzification

    Inputs are converted into membership values. Layer 2: Rule Evaluation

    Fuzzy rules calculate firing strengths. Layer 3: Normalization

    The firing strengths are normalized. Layer 4: Consequent

    Linear consequent functions are calculated.

    Layer 5: Output

    The final control command is generated.

    ANFIS is particularly attractive because the fuzzy rules can be optimized using training data.

12. COMPARISON OF CONVENTIONAL AND AI CONTROLLERS

    Controller	Adaptabilit y	Nonlinearit y	Training Required	Computational Complexity	Typical Application
    PI	Low	Low	No	Low	Conventional UPQC
    SRF-PI	Low medium	Low	No	Lowmedium	Grid-connected systems
    FLC	Medium	High	No	Medium	Dynamic PQ compensation
    ANN	High	High	Yes	Mediumhigh	Renewable systems
    ANFIS	High	High	Yes	Mediumhigh	UPQC/microgrid
    ANFIS + Optimization	Very high	High	Yes	High	Hybrid renewable systems
    Reinforcement Learning	Very high	Very high	Yes	High	Advanced adaptive control
    Deep Learning	Very high	Very high	Yes	Very high	Future intelligent grids

13. ANFIS-BASED UPQC RESEARCH

    Recent research demonstrates increasing interest in ANFIS-based UPQC control.

    A 2025 study investigated an open-UPQC with fuzzy and ANFIS control in a PV-integrated distribution system and employed a nine-level cascaded H-bridge multilevel inverter. The work specifically examined voltage sag, voltage swell, and harmonic mitigation.

    Another 2025 study proposed ANFIS-FBSO for a hybrid renewable-energy system containing PV, wind, and battery storage. The approach combined machine learning with optimization to address power balancing and frequency stabilization.

    These studies demonstrate an important trend: AI is increasingly being combined with optimization rather than being used as a standalone controller.

14. AI-BASED UPQC WITH RENEWABLE ENERGY

    Renewable-energy integration is one of the most important applications of intelligent UPQC.

    1. Solar PV

       PV systems can introduce:

       • Intermittent power.

       • Converter harmonics.

       • DC-link variation.

       • Voltage fluctuations.

         PV-UPQC combines generation and power-quality compensation.

         Recent work on PV-UPQC has investigated ANFIS and sliding-mode control to manage harmonic currents and reactive power while utilizing PV power through the shared DC link.

    2. Wind Energy

       Wind turbines connected through power converters can introduce rapid variations in active and reactive power. UPQC can compensate associated voltage and current disturbances.

    3. Hybrid Renewable Systems A hybrid system may contain:

       [ PV+Wind+BESS+UPQC

       ]

       Such systems have greater control complexity.

       A recent ANFIS-FBSO study specifically investigated UPQC integration with PV, wind turbines, and battery energy storage.

15. AI-BASED UPQC IN MICROGRIDS

    Microgrids contain:

    • Renewable sources.

    • Energy storage.

    • Power converters.

    • Critical loads.

    • EV chargers.

      Consequently, microgrids are particularly suitable for intelligent UPQC systems.

      The 2026 scoping review of UPQC research found that grid-connected microgrids dominate the literature and that intelligent approaches show promise for improved power-quality performance.

      However, microgrid operation introduces additional challenges:

    • Islanding.

    • Grid reconnection.

    • Frequency variations.

    • Bidirectional power flow.

    • Converter interactions.

    • Multiple control layers.

      AI controllers therefore need to be tested under both grid-connected and islanded conditions.

16. AI-BASED UPQC FOR ELECTRIC-VEHICLE CHARGING

    EV chargers are increasingly important nonlinear and dynamic loads. A large EV-charging station can cause:

    • Harmonic currents.

    • Reactive-power demand.

    • Voltage distortion.

    • Peak loading.

      UPQC can simultaneously compensate these disturbances while improving voltage quality at the charging point. An intelligent controller can additionally adapt to changing EV charging demand.

      This makes:

      [ EV+PV+BESS+UPQC

      ]

      a promising architecture for future charging stations.

17. OPTIMIZATION-ASSISTED AI-UPQC

    AI controllers can be combined with optimization algorithms. Examples include:

    • Particle Swarm Optimization.

    • Genetic Algorithm.

    • Grey Wolf Optimization.

    • Firefly-type algorithms.

    • Firebug Swarm Optimization.

    • Differential Evolution.

      The optimization algorithm can tune:

    • ANFIS membership functions.

    • Controller parameters.

    • Converter operating points.

    • Energy-management parameters.

      The ANFIS-FBSO approach reported in 2025 is an example of this direction. The advantage is improved parameter optimization.

      The disadvantage is increased computational complexity.

18. REINFORCEMENT LEARNING

    Reinforcement learning is an emerging approach.

    Instead of explicitly specifying a control rule, an agent learns a control policy through interaction with the environment. The objective may be:

    [

    J=w_1THD+w_2|V_{error}|+w_3|V_{dc,error}|+w_4P_{loss}

    ]

    The controller attempts to minimize (J). Advantages

    • Adaptive.

    • Suitable for nonlinear systems.

    • Can handle changing operating conditions.

      Challenges

    • Training stability.

    • Safety.

    • Exploration risk.

    • Computational requirements.

    • Need for reliable simulation environments.

    • Difficulty proving closed-loop stability.

19. DEEP LEARNING AND TRANSFORMER-BASED METHODS

    Deep learning provides opportunities for more complex power-quality applications. Potential applications include:

    • Disturbance classification.

    • Fault detection.

    • Predictive control.

    • Harmonic identification.

    • Load forecasting.

    • UPQC reference generation.

    Recent AI research beyond UPQC is also moving toward Transformer architectures for power-quality event classification, including operation under noise, DC offset, and amplitude/frequency variation.

    However, deep learning requires considerably more data and computational resources than traditional controllers.

20. CRITICAL COMPARISON OF EXISTING RESEARCH

    The literature indicates several important trends. Trend 1: Simulation Dominance

    Most UPQC studies are still validated primarily through MATLAB/Simulink.

    The 2026 scoping review specifically identified the predominance of simulation-based studies and called for greater experimental and real-time validation.

    Trend 2: Movement from PI to Intelligent Control

    PI remains popular because of its simplicity, but fuzzy, ANN, ANFIS, and hybrid methods are increasingly investigated. Trend 3: Renewable Integration

    PV-UPQC and hybrid renewable UPQC configurations are becoming increasingly important. Trend 4: Hybrid AI

    Researchers increasingly combine AI with optimization algorithms. Trend 5: Hardware Validation

    Real-time platforms such as OPAL-RT and dSPACE are increasingly relevant for demonstrating practical feasibility.

21. RESEARCH GAP

    The literature reveals several significant research gaps.

    1. Lack of Experimental Validation

       Many reported AI-UPQC systems remain simulation-based.

       A controller that performs well in MATLAB may behave differently on real hardware because of:

       • Measurement noise.

       • Switching delays.

       • Dead time.

       • Sensor dynamics.

       • Processor limitations.

       • Parameter uncertainty.

         Therefore, hardware validation is essential.

    2. Lack of Long-Term Stability Studies

       AI controllers can provide excellent transient performance, but long-term stability under changing grid conditions requires further investigation.

       The 2026 scoping review specifically identified long-term stability as an important research gap.

    3. Training-Data Dependency

       ANFIS and ANN controllers depend on representative training datasets.

       If the actual grid condition falls outside the training distribution, performance can degrade.

    4. Lack of Explainability

       A PI controller is easy to understand mathematically. A deep neural network may behave as a black box.

       For safety-critical electrical infrastructure, explainability is important.

    5. Stability Certification

       A major challenge is proving that an AI controller will remain stable under all operating conditions. This is particularly important for:

       • Reinforcement learning.

       • Deep learning.

       • Adaptive controllers.

    6. Cybersecurity

       AI-UPQC controllers may rely on communication and measurement data. Cyberattacks could manipulate:

       • Voltage measurements.

       • Current measurements.

       • Controller inputs.

       • Training data.

         Future research should therefore incorporate cybersecurity.

22. FUTURE RESEARCH DIRECTIONS

    1. Physics-Informed AI

       A promising direction is combining physical equations with machine learning.

       Instead of allowing AI to operate without constraints, the controller can incorporate: [

       Power balance

       ] [

       Voltage limits

       ] [

       Current limits

       ]

       and [

       Converter constraints

       ]

       into the learning process.

    2. Safe Reinforcement Learning

       Future RL-UPQC controllers should incorporate safety constraints. The controller should never generate commands that exceed:

       • Converter current.

       • DC-link voltage.

       • Semiconductor ratings.

       • Grid-code requirements.

    3. Digital Twins

       A digital twin can reproduce the electrical system in real time.

       The AI controller can be trained and evaluated in the digital environment before deployment.

    4. Hardware-in-the-Loop

       The next step after simulation should be:

       MATLAB/Simulink HIL Prototype Grid-connected experiment

       This provides increasing levels of validation.

    5. Multi-Agent AI

       Future distribution systems may contain several UPQCs. Each UPQC could operate as an intelligent agent.

       The agents could coordinate compensation using distributed optimization.

    6. AI for Predictive Compensation

       Instead of responding only after a disturbance occurs, AI could predict:

       • Load changes.

       • PV variations.

       • EV demand.

       • Harmonic trends.

         The UPQC could then prepare compensation in advance.

23. PROPOSED FRAMEWORK FOR FUTURE AI-UPQC RESEARCH

    A strong future research architecture can be represented as:

    PV + Wind + BESS + EV Loads

    Modern Distribution Network

    Power-Quality Measurement

    AI Disturbance Detection

    ANFIS/Deep Learning Controller

    UPQC Series + Shunt Converters

    Real-Time Compensation

    HIL / Experimental Validation

    The controller can simultaneously optimize: [

    THD

    ] [

    Voltage deviation

    ] [

    Power factor

    ] [

    DC-link error

    ] [

    Converter losses

    ]

24. RECOMMENDED EVALUATION METRICS

    Future research should report standardized metrics rather than only showing waveforms.

    Metric	Purpose
    Voltage THD	Voltage quality
    Current THD	Harmonic-current mitigation
    RMS voltage error	Voltage restoration
    RMS current	Current reduction
    Power factor	Reactive/harmonic compensation
    DC-link error	Converter regulation
    Settling time	Dynamic performance
    Overshoot	Transient quality
    Switching frequency	Converter stress
    Power loss	Efficiency
    Computational time	AI feasibility
    Robustness	Parameter variation
    Generalization	Unseen disturbances

25. OVERALL ASSESSMENT OF AI-UPQC TECHNOLOGY

    Based on the reviewed literature, the maturity of the major techniques can be summarized as follows.

    Technology	Research maturity	Main strength	Main limitation
    PI-UPQC	High	Simplicity	Limited adaptability

    Technology	Research maturity	Main strength	Main limitation
    SRF control	High	Established theory	Sensitivity to disturbances
    Fuzzy-UPQC	Highmedium	Nonlinear control	Rule dependence
    ANN-UPQC	Medium	Learning capability	Training dependency
    ANFIS-UPQC	Mediumhigh	Learning + interpretability	Computational complexity
    Optimization + ANFIS	Medium	Parameter optimization	High computational demand
    RL-UPQC	Emerging	Adaptive decision-making	Stability/safety
    Deep-learning UPQC	Emerging	Complex nonlinear mapping	Data/computation
    Physics-informed AI	Emerging	Physical consistency	Research maturity
    Multi-agent AI-UPQC	Emerging	Distributed coordination	Communication complexity

26. CONCLUSION

This review examined the development of AI-based Unified Power Quality Conditioner systems for modern electrical distribution networks. The increasing penetration of renewable energy, electric vehicles, energy storage, and nonlinear power-electronic loads is creating increasingly complex power-quality problems. Voltage sag, voltage swell, harmonics, reactive power, imbalance, and rapid operating-point changes require flexible and adaptive compensation technologies.

The UPQC is particularly attractive because it integrates series and shunt compensation within a single power-quality device. Conventional PI, synchronous-reference-frame, and instantaneous-power-based controllers have provided effective and practical solutions, but their fixed control characteristics can limit performance under highly nonlinear and rapidly changing conditions.

AI-based techniques provide a promising alternative. Fuzzy logic introduces nonlinear rule-based control, ANN provides data- driven learning, and ANFIS combines the strengths of both. Recent studies have expanded intelligent UPQC applications to PV systems, hybrid renewable-energy systems, microgrids, and EV charging. Research is also moving toward optimization-assisted AI, reinforcement learning, and deep-learning approaches.

However, AI-UPQC research remains at an important transition point. The major challenge is no longer simply demonstrating that an AI controller can reduce THD in a simulation. The more important questions concern robustness, stability certification, computational latency, generalization to unseen operating conditions, cybersecurity, explainability, and real-time implementation.

The recent 2026 scoping review confirms this direction, identifying experimental validation, long-term stability, and complex hybrid microgrid configurations as important areas requiring further research.

Consequently, the next generation of UPQC research should focus on physics-informed, safe, explainable, real-time AI controllers validated through hardware-in-the-loop and laboratory experiments. Such systems could provide adaptive power- quality management for renewable-rich smart grids and contribute to the development of resilient and intelligent future distribution networks.

REFERENCES

1. B. Khan, S. Jadapalli, G. V. Swathi, S. K. Maitra, P. Singh, U. Choudhury, and M. L. Biramo, Power quality improvement of microgrids using unified power quality conditioner (UPQC): A scoping review, Elctric Power Systems Research, vol. 253, 112573, 2026. doi: 10.1016/j.epsr.2025.112573.

2. N. Samala and C. Bethi, Harnessing synergy: a holistic review of hybrid renewable energy systems and unified power quality conditioner integration,

   Journal of Electrical Systems and Information Technology, vol. 12, article 4, 2025. doi: 10.1186/s43067-025-00193-1.

3. M. I. Abu Bakar, S. M. Uddin, and M. H. H. Rozlan, Performance analysis of an open-unified power quality conditioner with fuzzy logic and ANFIS controllers for enhanced PQ in distribution system, Ain Shams Engineering Journal, vol. 16, no. 12, 103836, 2025. doi: 10.1016/j.asej.2025.103836.

4. C. L. Babu, T. Gowri Manohar, and M. P. S., An Artificial Neural Network Control Based Unified Power Quality Conditioner for a Renewable Energy Coordinated Distribution System, NIPES Journal of Science and Technology Research, vol. 7, no. 3, pp. 1930, 2025. doi: 10.37933/nipes/7.3.2025.2.

5. M. Singh and L. Singh, Enhancing Power Quality in Grid-Integrated Hybrid Renewable Energy System using ANFIS-FBSO, Power Electronics and Drives, vol. 10, no. 1, pp. 189209, 2025. doi: 10.2478/pead-2025-0014.

6. L. Saihi and B. Berbaoui, Advanced Control for Photovoltaic-Unified Power Quality Conditioner Systems: A Simulation Approach, Revue Roumaine des Sciences Techniques Série Électrotechnique et Énergétique, 2025. doi: 10.59277/RRST-EE.2025.4.18.

7. A. K. Sahu and N. Vishwakarma, Power Quality Enhancement in Renewable-Integrated Smart Grids Using Unified Power Quality Conditioners: A Review,

   SMART MOVES JOURNAL IJOSCIENCE, vol. 11, no. 12, 2025.

8. P. K. Rajesh, S. Chapala, R. M. and L. Reddy, A Review of Emerging Techniques for Power Quality Improvement in Renewable Energy Integration, 2025.

9. S. Rao Sura, Artificial Intelligence Integrated Unified Power Quality Conditioner for Renewable Energy Systems, International Journal of Advanced Research in Electrical, Electronics and Instrumentation Engineering, vol. 15, no. 5, 2026. doi: 10.15662/IJAREEIE.2026.1505020.

10. P. Lin, Y. Gao, Y. Tang, M. W. Qaisar, P. Hui, C. Zhang, M. Zhu, and P. Wang, Artificial Intelligence for Power-Converter-Rich Electrical Systems: A Review, 2026.

11. IEEE Standards Association, IEEE Std 519-2022: IEEE Standard for Harmonic Control in Electric Power Systems, IEEE, 2022.

12. A. Mohammad Saber, A. Youssef, D. Svetinovic, H. Zeineldin, D. Kundur, and E. El-Saadany, Enhancing Power Quality Event Classification with AI Transformer Models, 2024.

Suggested Figures for Publication

Figure 1: Classification of power-quality disturbances.

Figure 2: Basic UPQC topology.

Figure 3: Series and shunt converter compensation principle.

Figure 4: Classification of UPQC control techniques. Figure 5: Conventional PI-based UPQC controller. Figure 6: Fuzzy logic UPQC controller.

Figure 7: ANN-based UPQC controller.

Figure 8: ANFIS architecture for UPQC.

Figure 9: AI-UPQC integrated with PV/BESS/EV system.

Figure 10: Comparison of AI-based UPQC methods.

Figure 11: Research gaps and future directions.

Figure 12: Proposed framework for physics-informed AI-UPQC. Recommended Novelty for Your Own Research

A review paper should not merely repeat existing reviews. The strongest angle for your paper is:

This is stronger than a generic UPQC review because the recent literature already contains general UPQC and renewable-UPQC reviews. The AI-control taxonomy + critical comparison + research gaps + future physics-informed/safe AI framework can provide the reviews distinctive contribution.