AI-Assisted UPQC with Quasi Z-Source SEPIC-Luo Converter for Harmonic Mitigation and Voltage Regulation in PV Applications

Abstract

The intermittent nature of photovoltaic (PV) energy, especially under nonlinear and unbalanced loading situations, has made it more difficult to ensure steady operation as it is increasingly integrated into modern power systems. The Power Quality (PQ) issues cause severe degradation of both system performance and device lifetime. A novel Neural Power Quality Network (NeuPQ-Net) controlled Unified Power Quality Conditioner (UPQC) combined with a Quasi Z-Source Lift (QZSL) converter for PV applications is presented in this research as a novel solution for addressing these issues. The QZSL converter, which is controlled by a Maximum Power Point Tracking (MPPT) algorithm based on Perturb and Observe (P&O), increases the PV source voltage to the necessary DC-link level. A Zebra Optimisation Algorithm tuned PI (ZOA-PI) controller continually adjusts PI gains for quick and accurate regulation, ensuring steady DC-link voltage. Unlike conventional Synchronous Reference Frame (SRF) or Decoupled Double Synchronous Reference Frame (DDSRF)-based reference generation, the proposed NeuPQ-Net operates directly in the abc domain, eliminating Phase-Locked Loop (PLL) dependency and reducing computational complexity. Simulation and hardware prototype validations demonstrate that the proposed system achieves significant improvements in PQ indices, including reduced Total Harmonic Distortion (THD), faster response to transients, and enhanced voltage regulation, while complying with IEEE-519 standards.

1. Introduction

1.1. Importance

The progress over the depletion of fossil fuels and their environmental impact has greatly increased the significance of RESs [1]. Among renewable sources, PV systems have arisen as the most widely utilized form of cleaner energy generation. Despite the numerous benefits, PV systems [2,3] are characteristically intermittent and unpredictable, as their energy output is dependent on external environmental factors, primarily solar radiation levels. This poses challenges when integrating PV systems into grid infrastructure, notably while supplying nonlinear and unbalanced loads. The major issue related with the generation of PV energy is Power Quality (PQ) [4] disturbances, owing to the intermittent nature, which leads to limitations including harmonic distortions, voltage sag, flickers and an unbalanced condition of the electrical grid. These issues negatively impact the system performance and also life span of devices linked to the grid. To mitigate these challenges, deployment of Flexible AC Transmission System (FACTS) [5] devices and advanced controllers are crucial. The following section examines previous research and scholarly works relevant to this specialized filed of interest.

1.2. Literature Review

FACTS devices are progressive power electronic systems designed for improving stability and efficiency of AC power transmission networks [6]. These devices regulate voltage levels, flow of power and system reactance, addressing different PQ issues such as fluctuations in voltage, reactive power imbalances and transmission losses. In the literature [7], the Thyristor Controlled Series Capacitor (TCSC) offers series compensation in a transmission line, improving regulation of voltage and increasing the capability of power transfer. Nevertheless, higher harmonic generation owing to thyristor switching impacts PQ. The Static Synchronous Series Compensator (SSSC) [8] uses passive compensation to regulate line reactance actively under varying load scenarios. However, power loss arises due to additional consumption of power. In [5], a Dynamic Voltage Restorer (DVR) is presented, addressing voltage sag and swell problems owing to disturbances in the grid. Moreover, it injects series voltage in opposition to disturbances, resulting in stabilized voltage. But, for effectual regulation of voltage compensation, DVR necessitates an energy storage system. The Static Synchronous Compensator (STATCOM) [9] provides precise injection of reactive power into the system with better support of voltage, even under heavy grid disturbances. Moreover, it improves fault ride through issues, particularly useful while in integration with RES. Nevertheless, the limitations in compensation of high system voltages restrict the system performance. The Static VAR Compensator (SVC) [10] is a thyristor-based control system offering compensation of reactive power, regulation of voltage, and elimination of sag and swell issues under varying conditions. Yet, reduced effectiveness and slower response occur owing to passive reactive elements. Likewise, the combination of series and shunt FACTS devices, including the Unified Power Quality Conditioner (UPQC) [11], offers simultaneous improvement in voltage and current, making it ideal for comprehensive enhancement of PQ issues. Similarly, the UPFC [12] is another FACTS device, and like UPQC, widely utilized in grid applications and RES-based power systems because of its adaption toward varying power demand and mitigation of power disturbances. However, these devices heavily depend on accurate and efficient reference current generation approaches. This determines the effectiveness of the system in compensating power disturbances, voltage sag/swell issues and fluctuations in reactive power. Generation of reference current is a crucial control mechanism utilized in shunt and series converters of FACTS devices [13] for extraction and injection of suitable compensating current to the grid. The comparative

Table 1. Review of reference current generation methods for power compensation in FACTS devices.

Technique	Description	Advantages	Disadvantages
Instantaneous Reactive Power Theory (p-q Theory) [14]	$\mathrm{Based} \mathrm{on} \alpha -\beta$ transformation, it decomposes active and reactive power components in a three-phase system.	Fast response time and suitable for balanced loads.	Performance degrades under unbalanced and distorted voltage conditions.
Synchronous Reference Frame Theory (SRF Theory—d-q Theory) [15]	Uses Park transformation to extract the fundamental component of current in the synchronous reference frame.	Provides precise fundamental component extraction and is effective under steady state conditions.	Requires a Phase-Locked Loop (PLL) for better performance under dynamic conditions.
Adaptive Filter-Based Approach [16]	Uses adaptive filters to estimate reference currents dynamically.	Appropriate for non-stationary signals and fast tracking capability.	Higher computational complexity and need for high-speed digital processing.
Least Mean Square (LMS) Algorithm [17]	Iteratively updates filter coefficients to estimate reference current.	Suitable for noise rejection and dynamic loads.	Slow convergence in some cases with requirement of extensive parameter tuning.
Synchronous Detection Method [18]	The mechanism of reference current generation is based on recognizing the supply current parts that match the phase of the fundamental voltage.	Works well for harmonic filtering.	Ineffective for unbalanced loads and are not suitable for transient conditions.
Wavelet Transform-Based Method [19]	Uses multi-resolution analysis to extract harmonic components from the signal.	Highly accurate for transient and steady-state analysis with better noise filtering ability.	High computational burden and needs for complex transformation algorithms.
Fuzzy Logic [20]	Uses Fuzzy rules for determination of reference current depending on system condition.	Works well under uncertain conditions.	Necessitates extensive rule-based tuning.

outlines various reference current generation approaches, beginning from conventional methods to recent techniques.

For surmounting these limitations, in this work a Neural Power Quality Network (NeuPQ-Net) controller is proposed for UPQC. In addition to FACTS and compensation methods, the reliance on PV requires [21] efficient energy conversion and regulation. Converters play a crucial role in regulating DC voltage levels. Classical converters [22,23] often fall short, with limited regulation of voltage, poor performance under dynamic load conditions, higher switching, and conduction losses. Recent advancements have presented controllers for mitigation of inter-area oscillations but fail to optimize system performance. Various optimization models including Particle Swarm Optimization Algorithm (PSO) [24], Grey Wolf Optimization (GWO) [25], Marine Predator Algorithm (MPA) [26], and Coot Bird Algorithm (CBA) [26] are some of the closed-loop methods to coordinate with FACTS devices. While these models offer computational efficiency, the systems also struggle in finding the best solution in selecting suitable conditions, limiting their effectiveness.

1.3. Research Gaps

Despite the improvements in UPQC-based power flow control, reactive power compensation, and renewable energy integration, several research gaps remain. Current studies lack adaptive real-time control strategies for dynamically adjusting to grid variations and renewable energy fluctuations. The integration of PV systems with UPQC requires further exploration, particularly in handling voltage stability and reference current optimization. Conventional dq0-based reference current generation methods struggle under nonlinear and unbalanced conditions, highlighting the need for AI-driven approaches. Additionally, economic feasibility and energy efficiency assessments of FACTS devices remain underexplored, limiting their practical implementation. Addressing these gaps requires a self-learning AI-based control strategy and real-time validation to enhance power system resilience, efficiency, and economic viability.

1.4. Contributions

After identifying the challenges in previous research, this paper presents an innovative AI-driven framework for PQ improvement and reactive power compensation. The contributions of this research are as follows:

• Introduces NeuPQ-Net controller for UPQC, which directly operates in the abc domain, eliminating the dependency on SRF, DDSRF or PLL-based transformations.
• NeuPQ-Net concurrently produces reference compensating currents for the shunt active power filter and reference injected voltages for the series compensator, resulting in the complete mitigation of voltage sags/swells, unbalances, harmonics and reactive power disturbances.
• Introduces QZSL converter to efficiently boost the voltage generated by the PV system, making it suitable for grid system.
• Implements a ZOA to fine-tune PI controller parameters and regulate the DC-link voltage.

1.5. Organization

The paper is structured as follow: outline of proposed work in Section 2, detailed modelling of proposed system components in Section 3, results and discussion presented in Section 4, and Section 5 summarizes the conclusion.

2. Description of Proposed System

In grid-connected systems with nonlinear loading conditions, the suggested solution effectively improves PQ by integrating UPQC with a PV-fed QZSL converter, as in Figure 1. The QZSL converter is implemented for increasing the variable DC voltage that the PV source first generates to the necessary level. The converter is controlled by a Perturb and Observe (P&O)-based MPPT algorithm to continuously track the maximum power point and regulate the boosted DC voltage. The ZOA-PI controller, which offers precise DC-link regulation for the UPQC system, then keeps this regulated output at a steady reference level. The stabilized DC-link voltage acts as the common energy interface for both the series Voltage Source Converter (VSC) and the shunt Active Power Filter (APF). On the compensation side, the series VSC is responsible for injecting the reference voltages $V_{Se,a}^{*},V_{Se,b}^{*},V_{Se,c}^{*}$ generated by the NeuPQ-Net controller, thereby mitigating voltage-related disturbances. Simultaneously, the shunt APF handles harmonic elimination and reactive power support by injecting compensating currents $i_{Sh,a}^{*},i_{Sh,b}^{*},i_{Sh,c}^{*}$, also computed by the NeuPQ-Net controller. For powering the VSI modules of the series and shunt branches, these reference signals are transformed into gate pulses using PWM generators. The cooperative action of the series and shunt converters ensures that the load voltages remain balanced and sinusoidal, while the source currents are maintained in phase with the source voltages and free from harmonic distortion.

3. System Modelling

3.1. UPQC System Mathematical Representation

The UPQC is a versatile FACTS device, shown in Figure 2, and is capable of regulating power flow in transmission lines, offering voltage support with improved transient stability and system oscillations. The transmission line connected in series with UPQC [12] supports in reducing the oscillations in grid line power. The magnitude and phase angle of series voltage are determined by $V_{se}$ and ${\phi }_{se}$. The complex power at transmission line is mathematically represented as

$$P_{real}-j{Q}_{reactive}={\overline{V}}_r\times I_{line}={\overline{V}}_r\left\{{\displaystyle \frac{\left({\overline{V}}_r+{\overline{V}}_{se}-{\overline{V}}_r\right)}{j\left(X\right)}}\right\}$$

(1)

$$\mathrm{where} {\overline{V}}_{se}=\left|V_{se}\right|\mathrm{\angle }({\delta }_s-{\phi }_{se})$$

(2)

The real power is expressed by

$$P_{real}={\displaystyle \frac{\left|V_s\right|\left|V_r\right|}{\left(X\right)}}\sin \left(\delta \right)+{\displaystyle \frac{\left|V_s\right|\left|V_r\right|}{\left(X\right)}}\sin \left(\delta -{\phi }_{se}\right)=P_0\left(\delta \right)+P_{se}\left(\delta ,{\phi }_{se}\right)$$

(3)

In the case of $V_{se}=0$, this represents the compensation of real power, whereas the voltage magnitude is regulated between 0 and $V_{se}\left(\mathrm{m}\mathrm{a}\mathrm{x}\right)$, and ${\phi }_{se}$ differs between 0 and $360°$ at any of its power angles. The representation of the UPQC-based controller is expressed as

$$∆P_{UPFC}\left(s\right)=\left({\displaystyle \frac{1}{{\tau }_{UPFC}S+1}}\right)$$

(4)

where the time constant of UPQC is defined as ${\tau }_{UPFC}$.

3.2. Neural Power Quality Network (NeuPQ-Net) Control of UPQC

UPQC control techniques like SRF and DDSRF rely largely on reference frame changes and complicated mathematical formulas. Despite their effectiveness, these approaches have rigidity, delay and sensitivity to PLL mistakes in variable microgrid settings. In order to address these issues, NeuPQ-Net, which is a lightweight, dual-headed neural network controller, is presented. It is intended to process measurements in the abc domain directly, obviating the necessity for synchronous reference frames. The architecture of NeuPQ-Net is provided in Figure 3.

3.2.1. Working of NeuPQ-Net Controller for Series Compensator

For reducing PQ issues including sags, swells and imbalances, the series compensator in a UPQC needs exact reference injected voltages. The NeuPQ-Net controller in the suggested framework, as in Figure 4, performs this function by processing the source and load voltages directly in the abc domain, without SRF or DDSRF conversions. The instantaneous three-phase source voltages $\left[V_{Sa},V_{Sb},V_{Sc}\right]$ and load voltages $\left[V_{La},V_{Lb},V_{Lc}\right]$ form the input vector of the controller, which is mathematically expressed as

$$X(t)={[V_{Sa}(t),V_{Sb}(t),V_{Sc}(t),V_{La}(t),V_{Lb}(t),V_{Lc}(t)]}^T$$

(5)

The trained neural network propagates this vector and uses a nonlinear mapping to produce the series compensator’s reference injected voltages,

$$Y_{se}\left(t\right)=f_{NeuPQ-Net}\left(X(t)\right)={\left[V_{Se,a}^{*}(t),V_{Se,b}^{*}(t),V_{Se,c}^{*}(t)\right]}^T$$

(6)

The function $f_{NeuPQ-Net}\left(·\right)$ specifies the internal transformation of the neural network, which encompasses hidden layers with nonlinear activation functions that learn the relationship between disturbed grid voltages and the ideal sinusoidal references. For ensuring bounded outputs, the final layer uses tanh activation, scaled by the maximum permissible injected voltage, such that

$$V_{se,x}^{*}\left(t\right)=V_{se,max}·\tanh \left(z_x\right),\hspace{1em}x\in \left\{a,b,c\right\}$$

(7)

Here, the pre-activation value from the final neuron is specified as $z_x$. Following generation, the reference voltages $V_{se}^{*}$ are fed into the PWM generator, which transforms them into VSI gating pulses. The load-side voltages are then returned to their rated sinusoidal form when the VSI injects the compensating voltages into the grid. NeuPQ-Net efficiently isolates the load from source disturbances through this direct neural mapping, enabling quick compensation without the delay or PLL dependence of traditional SRF-based controllers.

The NeuPQ-Net is designed as a supervised deep neural network consisting of a shared feature extraction backbone followed by two independent output heads corresponding to the series and shunt compensators. The shared backbone comprises three fully connected hidden layerscontaining64, 128 and 64 neurons, respectively. Each hidden layer employs the Rectified Linear Unit (ReLU) activation function to enable nonlinear feature learning and improve convergence stability. Batch normalization layers are incorporated after each hidden layer to enhance generalization and accelerate training convergence. The dual-headed structure enables simultaneous prediction of reference injected voltages for the series compensator and reference compensating currents for the shunt compensator. The output layers utilize tanh activation functions, scaled according to permissible voltage and current limits to ensure bounded control signals suitable for PWM generation. A70–15–15% split is adopted for training, validation and testing datasets respectively, ensuring unbiased performance evaluation. Early stopping criteria based on validation loss are implemented to prevent overfitting and guarantee stable generalization under unseen grid disturbances.

3.2.2. Working of NeuPQ-Net Controller for Shunt Compensator

Reactive power support, DC-link stabilization and current harmonic abatement are all handled by the UPQC shunt compensator. The NeuPQ-Net controller in the suggested design, as in Figure 5, accomplishes this functionality by directly processing real-time voltage and current measurements in the abc domain. The networks input vector is generated as

$$X(t)={[V_{Sa}(t),V_{Sb}(t),V_{Sc}(t),I_{La}(t),I_{Lb}(t),I_{Lc}(t)]}^T$$

(8)

Here, nonlinear load current is specified as $\left[I_{La},I_{Lb},I_{Lc}\right]$. The neural network creates the harmonic reference current components for compensation based on this input, which is denoted as

$$I_{habc}\left(t\right)=f_{NeuPQ-Net}\left(X(t)\right)={\left[i_{ha}\left(t\right),i_{hb}\left(t\right),i_{hc}\left(t\right)\right]}^T$$

(9)

Because stable UPQC operation depends on preserving the DC-link voltage, the discrepancy between reference and actual DC-link voltage is calculated as

$$∆V_{dc}\left(t\right)=V_{dc,ref}\left(t\right)-V_{dc}\left(t\right)$$

(10)

The ZOA-PI controller processes this error and adaptively adjusts the PI gains to guarantee quick and reliable DC-link regulation. The corrective current term is the controller output,

$$∆i_{dc}\left(t\right)=f_{ZOA-PI}\left(∆V_{dc}\left(t\right)\right)$$

(11)

which is added to the harmonic reference current generated by NeuPQ-Net. The final shunt reference current is

$$i_{shabc}^{*}\left(t\right)=I_{habc}\left(t\right)+∆i_{dc}\left(t\right)$$

(12)

The PWM generator then receives these reference currents and generates gating signals for the shunt VSI. In order to keep the source currents sinusoidal and in phase with the source voltages, the VSI introduces the necessary compensating currents into the grid. By working together, the ZOA-PI guarantees accurate DC-link regulation, and the NeuPQ-Net handles harmonic and reactive current compensation, thereby improving the PV-UPQC system’s stability and PQ.

3.3. ZOA Optimized PI Controller for DC-Link Voltage Control

The ZOA is a nature-inspired approach that mimics the foraging and defensive behavior of zebras in finding the optimal solution. In this work, ZOA is applied to tune $K_p$ and $K_i$ values of the PI controller to accomplish optimal regulation of the converter with stabilized DC-link voltage. The tuning process is presented in Figure 6.

3.3.1. Initialization of Population

In the initialization phase, random populations of zebras are initialized, each representing a potential solution in search space. Each zebra’s position in the search space corresponds to a pair $\left(K_p,K_i\right)$, defining a candidate PI controller configuration. The population is initialized as

$$X_Z=\left[\begin{array}{c}\begin{array}{cc}{K}_{P_1}& K_{i_1}\end{array}\\ \begin{array}{cc}{K}_{P_2}& K_{i_1}\end{array}\\ \begin{array}{cc}⋮& ⋮\end{array}\\ \begin{array}{cc}{K}_{P_N}& K_{i_N}\end{array}\end{array}\right]$$

(13)

where the population of zebra is defined by $X_Z$, the proportional and integral gain of $i^{th}$ zebra is represented as $K_{p_i}$ and $K_{i_i}$, and the number of zebra in the population as $N$.

3.3.2. Fitness Function Evaluation

A fitness function is evaluated for each zebra utilizing performance criterion such as

$$F_i={\int }_0^T\left[w_1{e\left(t\right)}^2+w_2{\left({\displaystyle \frac{de}{dt}}\right)}^2\right]dt$$

(14)

where a system error that is the difference between desired and actual values is represented as $e\left(t\right)$, rate of change of error as ${\displaystyle \frac{de}{dt}}$, and $w_1$ and $w_2$ define the weighting factors to balance error minimization under dynamic response. The objective is to identify optimal $K_p,K_i$ values that reduce the fitness function by providing improved performance.

3.3.3. Foraging Behavior—Update of Position Based on Best Solution

The zebras update their position using the foraging behavior, in which each zebra moves toward the best performing individual. This means each zebra updates it’s $K_p$ and $K_i$ value, moving toward optimal tuning. The expression for position update is given by

$$K_{p_i}^{n+1}=K_{p_i}+a_1·\left(K_{p_{best}}-a_2·K_{p_i}\right)$$

(15)

$$K_{i_i}^{n+1}=K_{i_i}+a_1·\left(K_{i_{best}}-a_2·K_{i_i}\right)$$

(16)

where gains of the best zebra are represented as $K_{p_{best}}$ and $K_{i_{best}}$, random number controlling exploration is represented as $a_1$, and zebra at new position is defined by $a_2$.

3.3.4. Selection of Best Individuals

In this phase, after updating of positions, the new value of PI is evaluated using the fitness function. If the new solution shows improved performance, the previous one will be replaced.

3.3.5. Defensive Strategy—Further Refinement of PI Gains

Zebras adopt a defensive strategy to escape from predators, which translates to additional refinement of gain values depending on stochastic exploration. The expression for defensive position update is given by

$$K_{p_i}^{n+1}=\left\{\begin{array}{cc}{K}_{p_i}+R·\left(2rand-1\right)·(1-t/T_{max})·K_{p_i},\hfill & if a_3\le 0.5\hfill \\ K_{p_i}+rand·\left(K_{p_A}-a_2·K_{p_i}\right),\hfill & if a_3>0.5\hfill \end{array}\right.$$

(17)

where randomly selected zebras with PI gains are defined as $K_{p_A}$, predefined constant controlling exploration as $R$, random number between interval $\left[0,1\right]$ is noted as $rand$, and gradual convergence over iterations as $t/T_{max}$.

3.3.6. Selection of Best PI Gains

After defensive strategy updates, the fitness function is re-examined. If new $K_p$ and $K_i$ values shows improved performance, replace the previous with the best, which ensures optimal PI controller settings. This results in improved converter performance with minimal steady state error, faster response time, minimized overshoot, and stabilized and efficient power system control.

3.4. PV System with QZSL Converter

In this work, PV is utilized as a source of energy for maintaining power supply to the grid. Accurate modelling of PV is essential, where the single diode modelling of PV is widely adopted owing to its beneficial characteristics. The expression is given by

$$I_{pv}=I_{ph}-I_0\left[e^{{\displaystyle \frac{V+I{R}_s}{a}}}-1\right]-{\displaystyle \frac{V+I{R}_s}{R_{sh}}}$$

(18)

$$a={\displaystyle \frac{N_s{n}_{1}k{T}_c}{q}}$$

(19)

where cells connected in series are noted as $N_s$, light generated photocurrent as $I_{ph}$, series and shunt resistance as $R_s$ and $R_{sh}$, Boltzmann constant as $k$, and ideality factor of diode as $a$. Moreover, the electron charge is represented as $q$, ideality factor as $n_1$ and module temperature as $T_c$.

The proposed converter is a combination of Z-Source converter with SEPIC-Luo topology, highly efficient for power conversion, with the ability to regulate voltage dynamically and with improved gain in voltage and reduced switching and conduction losses. The QZSL circuit is presented in Figure 7, with its switching waveform in Figure 8.

Operation of QZSL Converter

When the switch is in the ON state $(0<t<DT)$, shown in Figure 9, it receives positive voltage from the PWM signal, which enters the short circuit state. Inductors $L_a$ and $L_b$ are connected in parallel with switch and voltage source $V_{PV}$. The configuration allows both inductors to store energy simultaneously. At this instance the current across $L_a$ and $L_b$ increases. The capacitor $C_a$ charges using $V_{PV}$, and capacitor $C_b$ using diode $D_c$ ensures energy is stored for the next switching cycle. The diode $D_d$ is reverse biased, while capacitor $C_d$ charges through diode $D_e$, drawing energy from capacitor $C_c$. Both input and output sides remain disconnected, and capacitor $C_0$ maintains stable output voltage to load. The operational mode is expressed by

$$V_{L_a}=V_{L_b}=L_a{\displaystyle \frac{d{i}_{L_a}}{dt}}=L_b{\displaystyle \frac{d{i}_{L_b}}{dt}}=V_{PV}$$

(20)

$$V_{C_a}=V_{C_b}=V_{PV}$$

(21)

$$V_{PV}=V_{C_c}-V_{C_d}$$

(22)

In the interval $(DT<t<T)$, shown in Figure 9, when the switch is OFF, both the inductors $L_a$ and $L_b$ start discharging, forming a series circuit with capacitors $C_a,C_b$ and $C_c$ along with diode $D_d$. As shown in Figure 9, capacitors $C_a$ and $C_b$ start discharging, while $C_c$ charges, reversing the behavior of the previous mode. Simultaneously, via inductor $L_0$, capacitor $C_0$ receives energy from $V_{PV}$, and capacitor $C_d$ discharges through diode $D_0$, ensuring continuous power delivery to load.

$$V_{PV}=V_{L_a}+V_{L_b}-V_{C_a}-V_{C_b}+V_{C_c}$$

(23)

$$V_{PV}=V_{L_a}+V_{L_b}-V_{C_a}-V_{C_b}-V_{C_d}$$

(24)

$$V_{L_{a,b}}={\displaystyle \frac{3V_{PV}-V_{C_c}}{2}}$$

(25)

On applying voltage second balance principle for inductors, the mathematical expression becomes

$${\int }_0^{DT}{V}_{PV}dt+{\int }_{DT}^T\left({\displaystyle \frac{3V_{PV}-V_{C_c}}{2}}\right)dt=0$$

(26)

Therefore, the voltage across $C_c$ is expressed by

$$V_{C_c}={\displaystyle \frac{3-D}{1-D}}{V}_{PV}$$

(27)

The average voltage across inductors over a complete switching cycle in steady state operating condition is zero, mathematically expressed as

$$V_{L_0}D=V_{L_0}(1-D)$$

(28)

The output voltage is

$$V_0={\displaystyle \frac{5-D}{1-D}}{V}_{PV}$$

(29)

Therefore, the gain of the proposed QZSL converter becomes

$$G={\displaystyle \frac{V_0}{V_{PV}}}={\displaystyle \frac{5-D}{1-D}}$$

(30)

Optimal regulation of the converter is crucial for ensuring stability of the system. In this research two different optimization algorithm-assisted PI controllers are used for ensuring converter regulation.

4. Results and Discussion

MATLAB2024a validation results are explained for the developed system and are examined under different conditions in this section. The specifications of components utilized by the system for efficient functioning are listed in

Table 2. Design specification of proposed system.

Parameter	Ratings	Parameter	Ratings
Solar Photovoltaic System
PV power rating	5 kW	No. of panels connected in series	3
No. of panels connected in parallel	7	Cells per module	36
Open circuit voltage	37.25 V	Voltage at MPP	29.95 V
Short circuit current	8.95 A	Current at MPP	8.35 A
QZSL Converter
C0	2200 $\mathsf{\mu }\mathrm{F}$	La, Lb	22 $\mathrm{m}\mathrm{H}$
Da, Db, Dc, Dd, De	MUR1560	Ca, Cb, Cc, Cd	22 $\mathsf{\mu }\mathrm{F}$
Switching frequency	10 kHz	Duty cycle	0.5
ZOA-PI
Proportional gain (Kp)	0.5–1.5	Integral gain (Ki)	0.01–0.1
Population size	20–40	Maximum iterations	100–300
Learning rate	0.8–0.95	Response time	0.02 s–0.1 s
Settling time	≤0.5 s	Steady state error	≤0.01%
Sampling frequency	20 $\mathrm{k}\mathrm{H}\mathrm{z}$	Control implementation time	40–50 $\mathrm{\mu }\mathrm{s}$
Neural network interface latency	8–12$\mathrm{\mu }\mathrm{s}$	ADC resolution	12 $\mathrm{b}\mathrm{i}\mathrm{t}$

. Moreover, the hardware assessment together with comparative analysis in demonstrating the performance of the proposed work is also discussed.

• Case 1: Normal Condition

The illustration of three-phase source voltage and current is presented in Figure 10. It is seen that the voltage waveform shows balanced and sinusoidal performance, with voltage measuring 380 V, demonstrating stable and undisturbed condition of the grid. Whereas, the current waveform reveals oscillations at the beginning and continues at 25 A, illustrating the stabilized condition.

Under the normal condition, the input voltage and current from PV is utilized as input to the converter, as presented in Figure 11. It is seen that a stabilized voltage of 75 V with current measuring 65 A with initial oscillation is utilized as input to the proposed converter.

Based on the input, the proposed converter demonstrated boosted voltage in assistance with the ZOA-PI controller, as depicted in Figure 12. It is seen that the QZSL converter improved that voltage to 780 V, with ZOA-PI revealing better performance. Under the steady-state scenario, the converter current is noticed to be initially oscillating for a certain period of time and then stabilizes after 0.2 s, maintaining a current measure of 6 A after, as shown in Figure 13.

The assessment of waveform under the normal condition for both series and shunt converters is presented in Figure 14. It is seen that the shunt converter’s actual AC current waveform closely follows the reference signal, indicating accurate reactive power compensation. Similarly, the series converter’s actual AC voltage waveform tracks the reference signal with minimal deviation, demonstrating effective voltage regulation and disturbance mitigation. The smooth and stable waveforms reflect the efficiency of the control technique in maintaining balanced and stable operation, ensuring improved power quality and enhanced system performance.

The load characteristics waveform under Case 1 is presented in Figure 15. A pure sinusoidal load voltage of 380 V is attained, indicating no disturbances in the system, with current observed to be initially oscillation and then continuing at 25 A after.

• Case 2: With step magnitude +0.2 (Voltage Swell)

To test the performance of the developed UPQC system, the step magnitude is increased to +0.2, indicating the conditions of voltage sag, as shown in Figure 16. In Case 2, the disturbance corresponds to practical scenarios of load rejection, where a temporary overvoltage is created at the point of common coupling. It is seen that the magnitude is increased between 0.3 s and 0.5 s, with stabilization before and after. In correspondence, the current waveform shows severe oscillation at the first stage and gradually reduces after.

Under the scenario of voltage swell conditions, the input characteristics for the converter are illustrated in Figure 17. PV system voltage is observed to be maintained at a constant 75 V, with a current of 65 A indicating slight oscillations at the initial stage.

In accordance with the input, the proposed converter response with enhanced voltage of 780 Vis shown in Figure 18. Moreover, the current waveform shows slight fluctuations at the starting period and continues at 6 A as in Figure 19.

The waveform analysis under voltage swell conditions is presented in Figure 20; it confirms the effective performance of the proposed system in regulating both current and voltage. The shunt converter’s actual AC current waveform shows a noticeable increase in amplitude during the swell period, while the reference signal remains stable. Similarly, the series converter’s actual AC voltage waveform reflects the swell condition with increased amplitude, but it quickly settles back to a steady state, confirming effective compensation.

Under the voltage swell condition, the load characteristics waveform is presented in Figure 21. It is seen that the utilization of the UPQC system supports mitigation of voltage swell together with the generation of opposite harmonics using the NeuPQ-Net controller. As a result, harmonics are mitigated and stabilized at a load voltage of 380 V, with no more oscillations. In relation, the current waveform demonstrates severe oscillations, and gradually fluctuation reduces and maintains at 25 A after 0.15 s.

• Case 3: With step magnitude −0.2 (Voltage Sag)

Similarly, the UPQC system is tested under voltage sag conditions, as shown in Figure 22. In Case 3, the disturbance emulates events like motor starting, which cause a temporary voltage dip. The decrease in step magnitude of −0.2 is experienced at 0.3 s–0.5 s. Till 0.3 s, the source voltage is noticed to be maintained at 380 V, and after, a reduction in step magnitude is noticed; then, after 0.5 s, the source voltage continues at 380 V. The current waveform demonstrates an oscillatory source current and maintains a constant after 0.6 s.

PV performance under voltage sag (Case 3) is presented in Figure 23, showing astabilized voltage of 75 V, with current at 65 A, with initial oscillations. With respect to the input characteristics, the proposed converter is analyzed using ZOA-PI. The waveform reveals that utilizing the ZOA-PI controller shows effective performance, with voltage measuring 780 V, with slighter oscillations at the beginning as in Figure 24.

In relation to the input and performance of the converter, the corresponding current waveform is presented in Figure 25. With oscillations at the beginning, a current level of 6 A is attained with slight oscillations.

The waveform analysis under voltage sag conditions confirms the effective compensation capability of the proposed system, as shown in Figure 26. The shunt converter’s actual current waveform shows a noticeable reduction in amplitude during the sag period, followed by a rapid recovery, indicating effective reactive power compensation. Similarly, the series converter’s actual voltage waveform reveals sag conditions, with a distinct drop in amplitude, but the system quickly restores it to steady-state levels. The close tracking between reference and actual waveforms demonstrates that the proposed control strategy offers fast and accurate compensation, maintaining stable power flow and consistent power quality under sag conditions.

The condition of voltage sag is minimized, and the resultant load characteristic is presented in Figure 27. It is seen that the utilization of NeuPQ-Net supports the generation of the reference current. Meanwhile, the current shows severe oscillations, followed by stabilization later at 25 A.

The harmonics analysis for all three phases in presented in Figure 28, confirming the system operation under minimal distortion. Under a fundamental frequency of 50 Hz, the R & Y Phase results with a THD of 0.71% and B-Phase of 0.73%, respectively. This lower THD reflects the effective harmonic compensation and stabilized operation, guaranteeing that the system maintains improved power quality and balanced performance.

The proposed UPQC effectively compensates both sag and swell events and current harmonics. The series VSC injects the missing voltage component during sag conditions, restoring the load-side voltage to near-nominal levels. Simultaneously, excess voltage is absorbed during swell conditions, aiding in prevention of overvoltage stress on sensitive loads. Also, during nonlinear loading, the shunt APF assures that the source current remains sinusoidal and balanced. The voltage deviations are reduced with the improved voltage profile regulation along the distribution feeder. Added to this, the harmonic propagation is mitigated, thereby preventing resonance. Approximately 10–40% depth of voltage sag compensation occurs for durations from a few cycles up to several hundred milliseconds. Also, a 10–25% rise of voltage swell mitigation is performed. Along with these, the proposed work is capable of reducing current THD to $\le 5\%$.

4.1. Hardware Assessment

The experimental prototype in Figure 29 demonstrates the UPQC system using a rooftop solar PV system with QZSL Converter (KCP Eco Energy, Chennai, India) for enhanced power quality and dynamic compensation. The system includes a three-phase AC supply, input transformer, and an FPGA controller for real-time control and monitoring. The series and shunt converters work together to regulate active and reactive power flow, while an LC filter ensures harmonic reduction. The FPGA-based controller programmed with an optimization algorithm ensures efficient coordination between converters, providing stable voltage and current output under varying load conditions.

The waveform presenting three-phase source characteristics is presented in Figure 30. It is seen that a stabilized sinusoidal voltage is accomplished, with current demonstrating balanced performance across all phases. This confirms effective sharing of load and compensation of reactive power under steady-state operation.

The voltage utilized as input to the proposed converter is presented in Figure 31. It is seen that with oscillations at the beginning, a stabilized voltage of 75 V is attained from the PV system, necessitating a converter for further performance.

The performance of the proposed converter is examined using ZOA-PI controllers, as shown in Figure 32. It is seen that, by utilizing the controller, a boosted voltage of 780 V is accomplished.

The load voltage waveform of Figure 33 reveals that irrespective of changes in load condition, the proposed system continues to offer improved voltage. The balanced three-phase voltage indicates the effective regulation of load voltage. This demonstrates the effectiveness of the proposed QZSL converter in offering consistent voltage regulation.

The demonstration of source current THD presented in Figure 34 reveals the effectiveness of the proposed UPQC system with reduced harmonics and improved power quality. It is seen that the THD value of R-Phase is 3.12%, for Y-Phase it is 2.58%, and for B-Phase it is 3.55%, where the THD of all three phases lies within the acceptable limits of the grid system. This confirms the effectiveness of the proposed system in harmonic compensation and stabilized operation under varying conditions of load.

4.2. Comparative Analysis

The analysis of the existing converter with the QZSL converter is presented in

Table 3. Comparison of existing converter with the proposed QZSL converter.

Converter Analysis	Converter of Ref [27]	Converter of Ref [28]	Converter of Ref [29]	Converter of Ref [30]	Converter of Ref [31]	Proposed QZSL
Total No. Components	14	16	15	12	12	16
Voltage Stress on Switch	${\displaystyle \frac{V_0}{1+\left(n+1\right)D}}$	${\displaystyle \frac{V_0(1-D)}{2\left(1+D\right)}}$	${\displaystyle \frac{V_0(1+D)}{\left(1-D\right)}}$	${\displaystyle \frac{V_0(1+D)}{\left(1-D\right)}}$	${\displaystyle \frac{V_0}{\left(2+D\right)}}$	${\displaystyle \frac{2V_0}{\left(1-D\right)}}$
Voltage Gain	${\displaystyle \frac{1+3D}{1-D}}$	${\displaystyle \frac{2(1+D)}{(1-D)}}$	${\displaystyle \frac{nD(1+D)}{(1-D)}}$	${\displaystyle \frac{(1+D)}{(1-D)}}$	${\displaystyle \frac{(2+D)}{(1-D)}}$	${\displaystyle \frac{(5-D)}{(1-D)}}$

, in terms of component count, stress across switch and voltage gain. The QZSL converter has 16 components, which is higher than some existing designs but ensures enhanced performance. It is seen that stress across switches is minimized and results in better switching efficiency. Moreover, in contrast to existing converters, the QZSL ranks with higher voltage gain.

The comparison of voltage gain and efficiency is shown in Figure 35 for QZSL vs. the existing designs across various duty ratios. The voltage gain of the proposed converter reaches maximum gain, which is significantly higher than the other converters of [27,28,29,30,31], reflecting its enhanced voltage conversion capability. In terms of efficiency, the QZSL achieves nearly 96%, outperforming the other designs of references [27,28,29,30,31]. The superior performance is due to the optimized control strategy and improved design of the QZSL converter, ensuring better power conversion and higher energy efficiency.

The results presented in

Table 4. Evaluation of control techniques for UPQC in active power flow regulation.

Performance Indices	PI	FLC	MPC	SMC	GA	PSO	NeuPQ-Net
Peak Time (s)	0.29	0.2793	0.279	0.278	0.2777	0.2801	0.255
Peak (pu)	11.22	11.0661	11.062	11.071	11.0486	11.0235	10.01
Undershoot (pu)	0	0	0	0	0	0	0
Overshoot (pu)	2.304	0.6714	0.637	0.723	0.4965	0.2449	0.182
Settling Max	11.22	11.0661	11.062	11.071	11.0486	11.0235	9.01
Settling Min	9.972	9.9677	9.971	9.964	9.972	9.965	9.559
Setting time (s)	0.416	0.3976	0.379	0.381	0.4173	0.3973	0.265
Rise Time (s)	0.023	0.0215	0.022	0.022	0.0229	0.0228	0.0105
Control execution time per sampling step (µs)	72	65	61	68	63	59	41
Control delay (ms)	1.85	1.62	1.48	1.56	1.44	1.38	0.92

demonstrate the effectiveness of various control techniques for UPQC in active power flow regulation based on key performance indices under similar operating conditions. The proposed NeuPQ-Net control technique outperforms the conventional and other advanced methods in most performance metrics. Specifically, it achieves the lowest peak time (0.255 s) and minimum peak value (10.01 pu), indicating faster response and improved stability. The overshoot value for NeuPQ-Net (0.182 pu) is significantly lower than other methods, highlighting better control precision and reduced transient disturbances. The settling time of 0.265 s is the shortest among all methods, reflecting quicker stabilization of the system after disturbances. Additionally, the settling maximum and minimum values are well-regulated, showing improved grid stability under dynamic conditions. Also, the NeuPQ-Net control exhibits a reduced control execution time of 41 µs and control delay of ms.

On analyzing the results tabulated in

Table 5. Evaluation of control techniques for UPQC in reactive power flow regulation.

Performance Indices	PI	FLC	MPC	SMC	GA	PSO	NeuPQ-Net
Peak Time (s)	0.5413	0.5332	0.5312	0.5327	0.5316	0.5326	0.41
Peak (pu)	0.9488	0.8135	0.7993	0.7999	0.793	0.7923	0.665
Undershoot (pu)	87.8895	88.5834	89.3283	88.3666	86.8853	87.3603	84.5
Overshoot (pu)	35.109	15.785	13.761	13.8478	12.8551	12.7617	10.125
Settling Max	0.9488	0.8135	0.7993	0.7999	0.793	0.7923	0.665
Settling Min	0.56	0.5833	0.5815	0.5846	0.581	0.5793	0.455
Setting time (s)	0.5725	0.567	0.5567	0.6036	0.542	0.5422	0.425
Rise Time (s)	0.0153	0.0123	0.01452	0.022	0.0131	0.0133	0.0105
Control execution time per sampling step (µs)	74	66	63	69	64	61	43
Control delay (ms)	1.92	1.71	1.55	1.63	1.49	1.44	0.95

, it is evident that the proposed NeuPQ-Net method shows the best performance for UPQC in reactive power flow regulation under similar operating conditions. It exhibits the lowest peak time (0.41 s) and peak value (0.665 pu), ensuring faster response and improved stability.

The comparison of source current THD in

Table 6. Comparison of THD.

References	Source Current THD (%)
In [12]	10.4
In [32]	2.55
In [33]	1.54
In [34]	2.54
Proposed	0.71

demonstrates the superior performance of proposed system in reducing harmonic distortion under similar load conditions. The THD value in the proposed system is 0.71%, which is significantly lower compared to in other works, such as 10.4% in [12], 2.55% in [32], 1.54% in [33], and 2.54% in [34]. The reduced THD value reflects the improved harmonic compensation capability of the proposed system, achieved through the integration of the QZSL converter and Neu-PQ-Net-based reference current generation. This confirms that the proposed system ensures better power quality and more stable operation under varying load and grid conditions.

The fitness function minimization performance is illustrated in Figure 36, which compares ZOA with other conventional ones. The ZOA converges more rapidly and attains the lowest final fitness value of 0.62, demonstrating superior optimization ability and faster convergence speed. The early sharp drop in ZOA denotes efficient balance, leading to quicker attainment of the global optimum.

5. Conclusions

Through UPQC implementation managed by the suggested NeuPQ-Net and aided by the QZSL converter with ZOA-PI control, this work has presented a novel framework for PQ enhancement and reactive power compensation in PV-integrated grid systems. The PV source voltage is effectively boosted by the QZSL converter, while the P&O MPPT algorithm ensures continuous maximum power extraction. The ZOA-PI controller provides precise and dynamic DC-link voltage regulation, overcoming the limitations of conventional PI tuning. The NeuPQ-Net, operating directly in the abc domain, successfully generates reference injected voltages for the series compensator and compensating currents for the shunt active filter, eliminating the reliance on SRF/DDSRF transformations and PLLs. Simulation and hardware prototype validations demonstrate that the proposed system delivers superior PQ performance, including significant reduction in THD to below IEEE-519 limits, robust mitigation of voltage sags and swells, improved load voltage regulation, and effective harmonic and reactive power compensation under nonlinear and unbalanced conditions. Comparative analysis further highlights the superiority of NeuPQ-Net over conventional PI, Fuzzy, and optimization-based control techniques in terms of faster settling time, reduced overshoot, and improved dynamic response. The integration of NeuPQ-Net with QZSL and ZOA-PI establishes an efficient, adaptive and practically viable solution for next-generation renewable grid applications, ensuring resilience against PV intermittency and grid disturbances while extending the operational lifetime of power electronic devices and connected loads.