The proliferation of renewable energy sources, particularly photovoltaic (PV) generation, has fundamentally altered the landscape of modern power systems. At the heart of interfacing a PV array with the utility grid lies the power electronic solar inverter. Its primary function is to convert the direct current (DC) output from the solar panels into grid-synchronized alternating current (AC). While enabling this vital energy conversion, the switching nature of solar inverters inherently introduces harmonic distortion into the grid. These harmonics, if not properly managed, can degrade power quality, interfere with protective devices, and cause additional losses. Therefore, a precise understanding and accurate modeling of the harmonic emission characteristics of grid-connected solar inverters are paramount for system planning, compliance assessment, and the design of mitigation strategies. This analysis aims to deconstruct the harmonic generation mechanisms within a three-phase, half-bridge, bipolar Sinusoidal Pulse Width Modulation (SPWM) inverter—a prevalent topology in medium to large-scale PV plants. We will develop a dynamic harmonic domain model that accounts for non-ideal factors like dead-time effects and incorporates the influence of filter parameters and control loops, ultimately presenting a Thevenin-equivalent harmonic source model for comprehensive system-level studies.
Photovoltaic System Characteristics and Grid Interface
The electrical pathway from sunlight to grid involves several stages. The PV array, under specific irradiance (S) and temperature (T) conditions, generates DC power. A Maximum Power Point Tracking (MPPT) algorithm is employed to extract the maximum available power. This DC power is then fed to the solar inverter, which performs the DC/AC conversion. For non-isolated systems common in many installations, the inverter’s output is filtered and directly connected to the grid, often through an LCL filter for effective harmonic attenuation. The aggregate output of a PV plant is a function of the configuration of individual modules. Assuming a module’s MPPT voltage and power are \(U_{PV,m}\) and \(P_{PV,m}\), respectively, and the array is configured with \(N_s\) series and \(N_p\) parallel strings, the total array voltage \(E\) and maximum power \(P_m\) are given by:
$$E = N_s U_{PV,m}$$
$$P_m = N_p N_s P_{PV,m}$$
The inverter must then process this power, and its output current \(I_g\) at the Point of Common Coupling (PCC) with grid voltage \(U_g\) (assuming unity power factor operation) is approximately:
$$I_g \approx \frac{P_m \eta}{\sqrt{3} U_g}$$
where \(\eta\) is the inverter efficiency. The core of the harmonic generation lies within the solar inverter stage and its modulation process.
Harmonic Composition in SPWM Solar Inverters
The output voltage of a three-phase SPWM solar inverter is not a perfect sinusoid but a train of pulses whose width is modulated by comparing a sinusoidal reference (modulating) wave with a high-frequency triangular (carrier) wave. This process generates two primary categories of harmonics: switching harmonics and dead-time distortion harmonics.
Switching (High-Frequency) Harmonics
The ideal modulated output voltage, ignoring non-idealities, contains harmonics grouped around multiples of the switching frequency (carrier frequency). Using double Fourier series analysis, the peak magnitude of the harmonic component at frequency \((mN + n)f_s\) can be derived, where \(f_s\) is the fundamental grid frequency, \(N\) is the carrier ratio (switching frequency divided by \(f_s\)), and \(m\) and \(n\) are integers. For a bipolar SPWM scheme with a DC link voltage of \(E\), the harmonic voltage magnitude \(U_{inv,h}\) is:
$$U_{inv,h} = \frac{2E}{m\pi} J_n\left(\frac{mM\pi}{2}\right), \quad \text{for } m > 0 \text{ and } n \neq 0$$
Here, \(M = U_s/U_c \leq 1\) is the modulation index (ratio of modulating wave amplitude \(U_s\) to carrier wave amplitude \(U_c\)), and \(J_n(\cdot)\) is the Bessel function of the first kind of order \(n\). These harmonics are primarily located at high frequencies (e.g., around \(N f_s\), \(2N f_s\), etc.) and are effectively attenuated by the output LCL filter of the solar inverter.
Dead-Time Effect and Low-Forder Harmonics
A crucial non-ideal factor in practical solar inverters is the mandatory dead-time (\(\Delta t\)). This is a short delay inserted between turning off one switching device and turning on the complementary device in the same leg to prevent shoot-through faults. During this dead-time interval, the output current freewheels through the anti-parallel diodes, causing a voltage error pulse. The net effect is an average voltage distortion that is dependent on the polarity of the output current. This distortion introduces low-order harmonic voltages (e.g., 5th, 7th, 11th, 13th…) at the inverter terminals. The magnitude of the dead-time-induced harmonic voltage \(U_{d,h}\) for harmonic order \(n\) (odd and non-triplen) is approximately:
$$U_{d,h} = \frac{4E N f_s \Delta t}{n\pi}, \quad n=5,7,11,\dots$$
Importantly, the fundamental component of this dead-time voltage (\(n=1\)) is in opposition to the fundamental output current, effectively reducing the fundamental output voltage of the solar inverter. This has a direct impact on the required modulation index \(M\). From the steady-state phasor diagram of the inverter-filter-grid system, the relationship can be derived as:
$$M \approx \sqrt{ \left( \frac{8P_o^2}{3U_g^2 E^2} \omega_s^2 (L_1+L_2)^2 \right) + \left( \frac{\sqrt{2}U_g}{E} – \frac{2\sqrt{2} N f_s \Delta t}{\pi} \right)^2 }$$
where \(L_1\) and \(L_2\) are the inverter-side and grid-side inductances of the LCL filter, and \(\omega_s = 2\pi f_s\). This equation clearly shows that the required modulation index \(M\) decreases as the dead-time \(\Delta t\) increases, affecting the operating point of the solar inverter. The following table summarizes the sources and characteristics of these two harmonic types.
| Harmonic Type | Primary Cause | Dominant Frequency Range | Dependence | Nature in Model |
|---|---|---|---|---|
| Switching Harmonics | Ideal PWM Modulation | High (near switching freq. multiples) | Modulation Index (M), DC Voltage (E) | Voltage Source |
| Dead-Time Harmonics | Non-ideal Switching (Dead-Time) | Low (odd non-triplen orders) | Dead-Time (Δt), DC Voltage (E), Switching Freq. (Nf_s) | Current Source |
Mathematical Modeling of the LCL-Filtered Inverter with Control
To build a comprehensive harmonic model, we must integrate the power stage with its control system. A common and effective control strategy for LCL-type solar inverters is capacitor current feedback active damping. This method mitigates the resonance peak of the LCL filter without needing passive resistors. The single-phase equivalent control block diagram can be constructed, where the controller \(G_i(s)\) (often a PI or PR regulator) processes the error between the reference current \(i_g^*\) and the measured grid current \(i_g\). The capacitor current \(i_c\) is fed back with gain \(H_c(s)\) to provide damping. The plant includes the LCL filter impedances and the PWM gain \(G_{inv}\).
The closed-loop transfer function from the reference to the grid current, and importantly, the output impedance of the inverter seen from the PCC, can be derived. This output impedance \(Z_{eq}(s)\) is key to understanding how the inverter interacts with grid background harmonics. For a system with inverter-side inductance \(L_1\), filter capacitance \(C\), grid-side inductance \(L_2\), and controller \(G_i(s) = K_p + K_i/s\), the output impedance under capacitor current feedback takes a complex form influenced by all these parameters.
Development of a Thevenin-Equivalent Harmonic Source Model
The core insight for accurate modeling is recognizing the distinct propagation mechanisms of high-frequency switching harmonics and low-frequency dead-time harmonics. The former are best represented as a harmonic voltage source behind the inverter’s output impedance, as they originate from the modulation process. The latter, being inherently dependent on the load current, behave more like a current source injecting harmonic currents.
Therefore, we decouple these effects to form a composite Thevenin-Norton equivalent model. The overall model for harmonic studies consists of two parts:
- Fundamental Frequency Power Flow Model: This model calculates the steady-state operating point: DC voltage \(E\), output power \(P_o\), modulation index \(M\), and fundamental current \(I_g\). It uses the equations governing the PV array and the fundamental network equations of the AC side.
- Harmonic Domain Equivalent Model: This model superimposes harmonic sources on the fundamental operating point.
- The switching harmonics are modeled as a Thevenin voltage source \(U_{oc}(h)\), where \(h\) denotes the harmonic order. \(U_{oc}(h)\) is calculated using the double Fourier series formula for the ideal PWM output, and it is placed behind the closed-loop output impedance \(Z_{eq}(j\omega_h)\).
- The dead-time low-frequency harmonics are modeled as a Norton current source \(I_D(h)\). The injection current is calculated as \(I_D(h) = U_{d,h} / Z_{in}(j\omega_h)\), where \(Z_{in}(j\omega_h)\) is the equivalent impedance seen by the dead-time voltage source within the control loop. This impedance is a function of the controller and filter parameters:
$$Z_{in}(s) = \frac{s^3 L_1 C L_2 + s^2 C H_c G_{inv} L_2 + s(L_1+L_2) + H_g G_i G_{inv}}{G_i G_{inv}}$$
where \(H_g\) is the grid current feedback gain. The final harmonic injection model at the PCC for a single solar inverter is thus a combination of \(U_{oc}(h)\) in series with \(Z_{eq}(h)\) and a parallel current source \(I_D(h)\). For a plant with multiple inverters, the individual models are aggregated at the respective connection points.
Simulation Case Study and Model Validation
To validate the proposed modeling approach, a simulation model of a 500kW grid-connected PV system was established based on typical parameters. The system uses a three-phase half-bridge SPWM solar inverter with an LCL filter. Key parameters are listed below.
| Parameter | Symbol | Value |
|---|---|---|
| Rated Power | \(P_o\) | 500 kW |
| Grid Voltage (Line-to-Line) | \(U_g\) | 380 V |
| Grid Frequency | \(f_s\) | 50 Hz |
| DC Link Voltage Range | \(E\) | 450 – 800 V |
| Switching Frequency | \(f_{sw}\) | 1350 Hz (N=27) |
| Inverter-side Inductor | \(L_1\) | 160 μH |
| Grid-side Inductor | \(L_2\) | 50 μH |
| Filter Capacitor | \(C\) | 1800 μF |
| Dead Time | \(\Delta t\) | 3 μs |
| Current Controller | \(G_i(s)\) | \(K_p + K_i/s\) with \(K_p=0.45\), \(K_i=2150\) |
The simulation was run under varying DC link voltage conditions emulating changes in solar irradiance. The harmonic current distortion predicted by the proposed analytical model was compared against results from a detailed time-domain simulation (considered as “measured” data).
High-Frequency Harmonic Validation: The model showed excellent accuracy in predicting switching harmonics. For instance, the distortion trend for the 25th (1250 Hz) and 29th (1450 Hz) harmonics closely matched the simulation data as the DC voltage varied. The average error was around 2.2% for the 25th and 1.6% for the 29th harmonic. The inverse relationship between harmonic distortion and DC link voltage was correctly captured by the model.
Low-Frequency Harmonic Analysis: At a DC voltage snapshot of 625V, the low-order harmonic spectrum was analyzed. The model accurately predicted the presence and relative magnitudes of harmonics like the 7th, 13th, 19th, etc. However, discrepancies were observed for the 5th and 11th harmonics, where the simulation yielded higher distortion. This is attributed to factors not fully captured in the single-inverter model, such as the aggregation effect of multiple inverters in a plant, the influence of the step-up transformer’s impedance, and potential interaction with very low-order switching sidebands. This finding underscores that while the proposed model provides a highly accurate foundation, system-level studies must also consider aggregation effects. The comparison is summarized conceptually below.
| Harmonic Order | Model Prediction | Simulation Result | Agreement Level | Potential Reason for Discrepancy |
|---|---|---|---|---|
| 5th, 11th | Low | Higher | Moderate | Aggregation effects, transformer interaction. |
| 7th, 13th, 19th… | Medium | Medium | High | Core dead-time mechanism well modeled. |
| 25th (near f_sw) | Trend Captured | Trend Captured | Very High | Switching harmonic model is accurate. |
| 29th (near f_sw) | Trend Captured | Trend Captured | Very High | Switching harmonic model is accurate. |
Conclusion
This analysis presents a rigorous methodology for modeling the harmonic output of grid-connected solar inverters. By decomposing the inverter’s output into components arising from ideal SPWM modulation and the non-ideal dead-time effect, and by integrating these with the dynamics of the LCL filter and the capacitor-current-feedback control loop, a dynamic harmonic domain model was developed. The key outcome is a Thevenin-Norton equivalent harmonic source model that distinguishes between voltage-source-type high-frequency switching harmonics and current-source-type low-frequency dead-time harmonics. This distinction is crucial for accurate studies of harmonic propagation in networks with high penetration of PV generation.
The simulation-based validation confirms that the model can predict the harmonic emission trends of a solar inverter with high accuracy, especially for switching harmonics and the core low-order dead-time harmonics. The model provides a powerful analytical tool for several critical applications: estimating the harmonic distortion at the PCC of a PV plant under varying operational conditions (irradiance, temperature), assessing the hosting capacity of distribution networks for PV, and providing a validated basis for the specification and design of power quality mitigation equipment, such as active harmonic filters. Future work may focus on extending the model to account for multi-inverter aggregation effects, unbalanced grid conditions, and the impact of advanced modulation techniques used in modern solar inverters.