Harmonic Emission Variability in Identical PV Inverters Installed in Different Solar Farms

Abstract

The increasing integration of photovoltaic (PV) power plants into electrical grids has raised concerns regarding the harmonic emissions caused by power electronic devices. Understanding the impact of large-scale PV plants on power systems is essential for ensuring grid stability and power quality. This study analyses and compares the current harmonic emissions of two inverters with identical topologies, installed in two different PV plants: a 35 MW plant in Spain and a 12 MW plant in Greece. The results reveal significant differences in the harmonic emission spectra between the two inverters, highlighting the influence of external conditions on the harmonic behaviour of PV inverters. These findings underscore the necessity of considering site-specific factors when assessing the harmonic impact of large-scale PV installations.

1. Introduction

In recent years, solar photovoltaic (PV) energy has experienced remarkable growth worldwide, driven by increasing environmental concerns, supportive regulatory frameworks, and the continuous reduction in system costs [1]. This trend is expected to continue in the coming decades, positioning PV as a key pillar in the transition toward sustainable energy systems [2].

As the penetration of PV plants in the power grid increases, so does the relevance of assessing their impact on power quality. One of the main concerns associated with PV integration is the generation of electrical harmonics, which originate primarily from the power electronic inverters used to convert direct current (DC) power from the panels into alternative current (AC) power for grid injection [3]. These harmonics can lead to a range of operational issues, including overheating of transformers and conductors, malfunctioning of protection devices, reduced efficiency of other power electronic equipment, and potential resonance phenomena in the network [4,5].

Given these risks, it is essential to evaluate the harmonic emission of inverters operating in real-world PV installations and to understand the extent to which external conditions—such as grid configuration, voltage distortion, cable layout, or local impedance—can influence their behaviour [6,7]. Such understanding is particularly important in large-scale PV plants, where the interaction between inverters and the grid can be complex and highly site-dependent.

The above-mentioned growing penetration of utility-scale PV plants has underscored the need to comprehensively characterize their harmonic emission patterns. For instance, refs. [8,9] analysed harmonics up to the 31st order in simulated 1.2 MW and 4 MW PV systems. Similarly, refs. [10,11] relied on synthetic data to assess harmonic distortion, while [12] combined simulations with a single-day measurement campaign on a 1.4 MW plant. Laboratory studies, such as those by [13,14], have provided valuable insights into harmonic behaviour under idealized conditions. However, these approaches often fail to capture the complex interactions between grid topology, environmental factors, and inverter operation that define real-world harmonic emissions.

The correlation between current and voltage harmonics in PV installations is well-documented [7,15], yet most field studies focus narrowly on compliance measurements at the Point of Common Coupling (PCC), as mandated by standards like IEEE 519, IEC 61000-3-6, and EN 50160 [16,17,18]. While these regulations ensure grid-compatible voltage quality, they offer limited insight into the generation-side harmonic dynamics of large PV plants—particularly how identical inverters behave across diverse grid environments. To date, comparative studies analysing harmonic emission disparities attributable to site-specific factors (e.g., impedance profiles, DC-side conditions) remain scarce. This work bridges that gap by examining real-world harmonic data from twin inverters operating in distinct electrical ecosystems, providing empirical evidence of how external variables reshape emission signatures.

Even when identical PV inverters are deployed, their harmonic behaviour may differ significantly depending on the electrical characteristics of the network where they are connected. This is mainly due to the influence of the external grid impedance, which modifies the voltage at the inverter terminals and consequently affects both the propagation and generation of harmonics.

To better understand this phenomenon, it is useful to distinguish between primary and secondary harmonic emissions. Primary emissions are those directly generated by the inverter’s internal switching and control mechanisms, whereas secondary emissions arise from the interaction between the Inverter and the external grid—particularly through the impedance and harmonic background of the network [19].

Figure 1 illustrates a commonly used conceptual model that helps to differentiate both contributions. In this scheme, the internal current source $E_1$ represents the inverter’s primary harmonic emission, while the background voltage $E_2$ and the network impedance $Z_2$ are responsible for the secondary emission due to voltage harmonic distortion at the PCC. As both the background distortion and the network impedance vary from one installation to another, the total harmonic current observed at the output of the inverter may differ even when using the same hardware and control parameters.

Recent studies on Active Distribution Networks (ADNs) have focused on the joint planning of Distributed Generation (DG) and Energy Storage Systems (ESSs) using advanced bi-level programming to handle operational uncertainties [20]. While these frameworks optimize location and capacity for economic efficiency, they often rely on simplified harmonic models. Our work addresses this gap by quantifying how site-specific grid conditions—fundamental for ADN planning—significantly alter the harmonic emission of identical inverter hardware.

While the overall harmonic impact and regulatory compliance of large-scale PV plants at the PCC have been previously characterized by the authors—specifically regarding voltage harmonic effects on transmission networks [21]—the present work shifts the focus to the inverter’s AC terminals. This methodological choice is fundamental for isolating the direct interaction between the power electronics’ control loops and the local grid impedance, avoiding the masking effects caused by the aggregation of multiple units and internal plant infrastructure. In this context, the study aims to compare the current harmonic behaviour of two identical inverters installed in different utility-scale PV plants to assess how external factors impact emission characteristics when internal hardware and control strategies remain unchanged. Ultimately, understanding these source-level variations is a prerequisite for developing the high-fidelity models required to accurately predict and mitigate the aggregate harmonic effects observed at the PCC.

2. Database

Two measurement campaigns were carried out in two different large-scale grid-connected PV power plants located in Europe. The first campaign was conducted in a 35 MW PV plant located in Spain (hereafter referred to as Plant A) over a period of two months, while the second campaign was performed in a 12 MW PV plant located in Greece (hereafter referred to as Plant B) over a period of four weeks. In both cases, current and voltage measurements were obtained at the output of a single string inverter (referred to as Inverter A and Inverter B, respectively), using identical equipment and configuration. The two inverters under analysis share the same model, manufacturer, and topology, which enables a direct comparison of their harmonic emission under different grid and installation conditions.

The measurements were acquired employing a Fluke TM 1760 Power Quality Analyser (Fluke Corporation, Everett, WA, USA), a Class-A compliant device meeting the requirements of IEC 61000-4-30:2015+A1:2021 [22]. Current sensing was performed with Fluke TPS FLEX 24 Flexible Current Transducers (Rogowski-type), featuring manufacturer-specified accuracy of 1% for magnitude and 0.5° phase displacement at 23 °C ± 2 K, across the 48–65 Hz fundamental band. These transducers provide a linear frequency response up to 3.0 kHz, covering the entire harmonic spectrum analysed in this study without magnetic saturation issues. The current sensing range was fixed between 2 A and 200 A to match the inverter’s nominal current (126 A) while maintaining sensitivity at lower loads. To ensure the reliability of low-magnitude harmonic components derived via Fast Fourier Transform, potential stochastic errors were mitigated through a robust statistical averaging process of at least 10,000 data points per power-loading bin.

While field temperatures varied during the campaigns in Spain and Greece, the high-precision transducers used are designed for outdoor stability, and their Class-A certification ensures that total measurement uncertainty remains within the rigorous limits required for comparative field assessments.

The recorded data comprises time-series with 3-s granularity, containing RMS measurements of voltage and current along with derived quantities: active, reactive, and apparent power, plus power factor. Voltage and current harmonic/interharmonic analysis was conducted by the instrument’s embedded processing following IEC 61000-4-7:2002+A1:2009 (section 5.6) [23], utilizing 200 ms (10/12-cycle) observation windows with gapless subgroup aggregation, without any post-processing smoothing algorithms.

The electrical distribution layout of Plant A is illustrated in Figure 2. Similarly, Figure 3 displays the topology of Plant B. In both cases, red markers indicate the the locations of power quality monitoring devices.

3. Methodology

This study aims to evaluate how external conditions influence the harmonic emission of power electronic inverters in utility-scale PV plants. The analysis focused on the comparison of current harmonic distortion as a function of inverter loading. From the available dataset, current harmonic amplitudes up to the 50th harmonic (H50) were extracted at 3-s intervals for each inverter, in accordance with the IEC 61000-4-7 standard.

As illustrated in Figure 1, the total harmonic current measured at the inverter terminals is the vector sum of primary emissions (inherent to the inverter’s switching and control) and secondary emissions (resulting from the interaction with background voltage and grid impedance). Given that the inverter models at both sites are identical, their primary emission is assumed to be equivalent. Therefore, the observed differences provide a quantitative measure of the impact of secondary emissions. While a full decoupling of these components would require internal control parameters and extensive network modelling beyond the scope of this study, the empirical evidence presented here serves to quantify the degree of site dependency in large-scale PV installations.

To quantify the relationship between harmonic emission and inverter power output, the data were grouped into power intervals expressed as a percentage of the inverter’s rated capacity. For each interval, the mean values of each harmonic were calculated. This process was performed independently for Inverter A and Inverter B, yielding two correlation matrices.

Subsequently, the relative variation in correlation values between the two inverters was computed for each harmonic and power range as per Equation (1). The resulting variations were represented in a set of comparative heatmaps, separately for each phase and distinguishing between odd and even harmonics. Positive values indicate that Inverter A exhibits higher correlation levels than Inverter B, while negative values reflect the opposite trend.

$$Va{r}_{H_n}={\displaystyle \frac{I_{H_n}^A-I_{H_n}^B}{I_{H_n}^B}}$$

(1)

This approach enables the identification of harmonic orders and operating conditions under which the influence of external factors—such as grid impedance, voltage distortion, cable length, and overall network characteristics—becomes more significant. By isolating the harmonic behaviour of two identical inverters, the methodology supports a deeper understanding of how installation and grid context can affect power quality in PV systems.

While the monitoring periods differed in duration and seasonality (May 2022 for Site A and Oct–Dec 2021 for Site B), the impact on the comparative analysis is mitigated by the power-normalization approach. By binning the harmonic data according to fundamental current levels, the analysis focuses on the inverter’s operational state rather than seasonal energy trends. This ensures that the comparison reflects how identical hardware responds to the same loading conditions under different grid environments, regardless of the time of year the data was captured.

To ensure that the observed differences are rooted in grid-converter interaction rather than environmental or operational coincidences, several safeguards were implemented. First, the use of identical Class A instrumentation across both sites ensures that measurement uncertainty remains consistent and negligible. Second, the data aggregation strategy acts as a statistical filter; by averaging measurements within each power-loading bin over an extended monitoring period, the influence of transient environmental outliers or rapid irradiance fluctuations is effectively mitigated, revealing the underlying steady-state harmonic signature. Finally, the spectral selectivity of the variations—where specific frequency bands exhibit divergent behaviours despite similar power levels—further confirms an electrical coupling mechanism (impedance-based) rather than a broad-spectrum response to environmental factors like temperature or soiling.

4. Results

Figure 4 presents the relative variation (%) of the correlation between current harmonic components and generated power. The analysis is performed separately for odd (top row) and even harmonics (bottom row) across the three phases, where red tones indicate Inverter A shows greater values than Inverter B and blue tones represent the opposite.

The results reveal significant differences in harmonic behaviour between the two sites, despite the inverters having the same model and control topology. These differences are consistent across the three phases.

The absolute magnitudes summarized in

Table 1. Minimum and maximum absolute mean values of current harmonics (H2–H50) per power-loading bin for Inverters A and B (in mA).

Power Range (%)	Inv A—Min	Inv B—Min	Inv A—Max	Inv B—Max
0–1	14.39	8.48	58.01	82.20
1–5	28.76	22.78	519.73	847.61
5–10	30.87	19.04	827.08	1055.83
10–15	29.10	21.04	875.10	1033.06
15–20	30.50	20.53	752.72	838.71
20–25	30.13	20.95	759.13	917.37
25–30	29.59	22.03	757.49	818.87
30–35	31.83	22.61	696.75	630.05
35–40	33.03	23.04	607.31	507.79
40–45	32.81	23.11	564.54	648.15
45–50	31.16	23.14	609.31	908.27
50–55	32.01	23.77	727.51	1130.31
55–60	32.21	22.56	739.15	1179.01
60–65	35.82	23.34	768.93	1159.99
65–70	37.40	24.69	836.98	1139.33
70–75	38.90	26.94	940.53	1132.88
75–80	37.62	29.47	1063.02	1120.64
80–85	39.09	32.71	1062.48	1087.54
85–90	37.12	33.26	1023.40	1091.03
90–100	40.96	39.62	1170.42	1160.05

confirm the significance of the recorded emissions and the reliability of the instrumentation. For power levels above 1%, average harmonic currents consistently remain above 20 mA, ensuring they are well within the measurement sensitivity of the Class-A equipment. In the lowest power range (<1%), current magnitudes decrease to approximately 8 mA. While these are lower values, the statistical consistency of this interval is supported by a remarkably large dataset of up to 300,000 points, whereas all other power bins are composed of at least 10,000 samples. This high data density explains why the correlation patterns in the initial power range may exhibit slight deviations from the trends observed at higher loading levels while still remaining statistically representative of the inverter’s behaviour under near-no-load conditions.

For the odd harmonics, distinct patterns can be identified across different harmonic order ranges. For harmonics below H13, Inverter B consistently exhibits higher correlation values with power than Inverter A across nearly the entire power range. These low-order harmonics are known to be intrinsically linked to the energy conversion process in PV inverters, and their behaviour reflects fundamental operational characteristics of the devices. In the intermediate range—from H13 to H33—a crossover trend is observed: Inverter A presents higher correlation values for power levels below approximately 50% of rated capacity, whereas Inverter B shows higher correlations beyond this threshold. This behaviour suggests a power-dependent shift in the emission characteristics for these harmonics. In contrast, for higher-order odd harmonics above H35, Inverter A appears to consistently show greater correlation values across the full power spectrum. This indicates that the divergent behaviour between sites persists even at higher frequencies, suggesting that the identical control hardware responds differently to each electrical environment.

Regarding even harmonics, higher correlation values in Inverter A are observed, as reflected by the predominance of red tones, whose intensity increases for higher-order harmonics. This suggests that the even harmonic content is more sensitive to the specific installation conditions. The possible physical mechanisms behind this attenuation at Plant B, such as the passive filtering effect of the AC infrastructure, are further analysed in the Section 5.

The fact that such disparities arise between two functionally identical inverters operating under different electrical environments is consistent with the hypothesis that external factors—such as grid impedance, cable length, and local network conditions—influence harmonic emission patterns. These observations emphasize the relevance of site-specific harmonic assessments, even when standardized inverter technologies are employed, and highlight the challenges of generalizing harmonic performance from a single installation.

5. Discussion of the Results

The results reveal significant differences in harmonic behaviour between the two sites, despite the inverters having the same model and control topology. These differences are consistent across the three phases and can be attributed to the following factors.

5.1. Grid Impedance and Network Topology

The observed disparities in low-order odd harmonics (below H13)—where Inverter B exhibits higher correlation with power—align with the expected influence of grid impedance on harmonic emission. The higher sensitivity observed in Inverter B could be explained by a higher grid impedance at its PCC, which would result in greater voltage distortion for a given current harmonic injection. This follows from the relationship

$$V_h=Z_h·I_h$$

(2)

where $V_h$ is the harmonic voltage, $Z_h$ is the impedance at harmonic order h, and $I_h$ the harmonic current. A higher $Z_h$ at Plant B would amplify the sensitivity of harmonic currents to power variations, explaining the stronger correlation. These findings are consistent with controlled laboratory experiments, such as those by Xu et al. [6], which demonstrate that variations in source impedance and supply voltage distortion directly alter the harmonic footprint of PV inverters across their entire operating range.

For intermediate odd harmonics (H13–H33), the crossover trend (Inverter A dominant below 50% power, Inverter B above) is consistent with potential resonance phenomena. This behaviour suggests that shifts in the inverter’s operational state (e.g., modulation index adjustments or control loop dynamics) may interact differently with the site-specific passive network. As noted in [6], the interaction between the inverter’s output filter and the external grid conditions is highly dependent on the fundamental power level, which supports our observation of frequency-selective variations as power loading increases. For instance, the distinct topologies and scales of the plants (35 MW vs. 12 MW) imply different equivalent impedances at the inverter terminals. This includes not only the AC cabling but also the leakage impedance of the step-up transformers, which acts as a series Inductance (L) that significantly influences the frequency response and the coupling of low-order harmonics. In larger plants, the aggregation of these magnetic and capacitive components creates a unique impedance profile that can amplify or attenuate specific frequency bands compared to smaller installations; however, a detailed characterization of these resonant frequencies and their precise root causes falls outside the scope of the present study and remains a subject for future research.

5.2. Environmental and DC-Side Factors

The consistent dominance of Inverter A in high-order odd harmonics (above H35) and even harmonics is consistent with potential non-ideal conditions in its DC subsystem or ambient environment:

• Temperature and Irradiance Transients: Transient environmental factors, such as rapid irradiance changes or thermal gradients, are known to induce DC voltage ripples (e.g., at 100/120 Hz) which could modulate the inverter’s switching frequency, elevating sideband harmonics (e.g., $f_{sw}\pm 2f_{grid}$). This is particularly relevant for even harmonics, which often arise from asymmetry in the DC-link voltage or half-wave rectification effects.
• PV Module Degradation or Soiling: Uneven soiling (e.g., dust accumulation) or mismatch in Plant A’s strings could introduce DC-side current imbalance, exciting triple-n harmonics (third, ninth) that propagate to the AC side through the inverter’s modulation process.

5.3. Collector System Impedance and High-Frequency Attenuation

The predominance of even harmonics in Inverter A (red tones) is compatible with the hypothesis of higher impedance in the AC collector system —comprising transformer leakage and cable parameters— of Plant B acting as a passive filter. Since collector system impedance increases with frequency ($Z_c=R+jwL$), the infrastructure at Plant B could potentially attenuate harmonic currents above 1 kHz (approximately H17–H25 in 50 Hz systems). This mechanism aligns with the observed rise in inverter A’s high-order harmonic correlation relative to the other site.

6. Conclusions and Future Work

This paper has presented a comparative study on the harmonic emission behaviour of two identical string inverters operating in two utility-scale PV plants located in different countries (Spain and Greece). By analysing the correlation between harmonic current magnitudes and the generated active power over various power intervals, significant differences have been observed despite the inverters sharing the same topology and control strategy.

This study demonstrates that the harmonic emission profiles of identical PV inverters can vary significantly across different installations, even when operating under similar power output conditions. The observed differences—particularly in low-order harmonics (H5–H13), the power-dependent crossover in intermediate orders (H13–H33), and the dominance of high-frequency content at one site—highlight how external factors outweigh internal inverter design in shaping harmonic behaviour. The interplay between grid impedance, cable infrastructure, and environmental conditions emerges as the primary driver of these variations. Higher grid impedance at one plant amplified low-order harmonic sensitivity, while resonant interactions with local network components selectively exacerbated emissions in specific frequency bands. Meanwhile, the prevalence of high-order harmonics at the other site suggests DC-side asymmetries, possibly due to irradiance transients, thermal effects, or PV module mismatch, with cable impedance acting as a passive filter that attenuated high-frequency content differently at each location.

These findings underscore that harmonic emissions cannot be assessed in isolation from the electrical and environmental context. From a planning perspective, the site-dependent signatures identified in this research suggest that power quality constraints should be more rigorously integrated into ADN frameworks. Specifically, the integration of real-world harmonic variability is essential for advanced bi-level optimization models [20] to prevent unforeseen resonance issues during the joint allocation of DG and ESS. The study challenges the assumption that standardized inverter designs guarantee consistent harmonic performance, emphasizing instead the need for localized assessments tailored to each installation’s unique characteristics.

This research moves beyond site-specific observations by establishing that the harmonic identity of a PV inverter is not a fixed characteristic but a dynamic interaction. These results provide a cautionary tale for power system planners, suggesting that standard harmonic models must incorporate a variability factor derived from the specific grid strength and plant topology.

To gain deeper insight into the origin of the observed differences, future research should focus on analysing the same inverter models under varying power levels and grid conditions, in order to isolate the influence of the external environment on harmonic behaviour. In parallel, the proposed methodology should be extended to different inverter models or topologies to evaluate whether similar patterns are observed and to assess the general applicability of the findings. It is also essential to perform multi-point harmonic measurements, including at the PCC, to better understand how harmonic currents propagate through the system and to distinguish the specific contributions of the inverters from those of the grid or other network elements. Furthermore, characterizing or estimating the grid impedance seen by the inverter could help clarify the role of local network characteristics in shaping the harmonic response. Finally, future studies should consider transient operating conditions—such as start-up events or rapid changes in power output—to evaluate how dynamic interactions affect harmonic emissions beyond steady-state behaviour. These lines of work will support a more comprehensive understanding of harmonic phenomena in real PV environments and inform improved assessment and mitigation practices.