The transition towards renewable energy integration has placed significant demands on power conversion systems. In the context of photovoltaic (PV) generation, the grid-connected inverter is a critical component responsible for converting DC power from solar panels into high-quality AC power synchronized with the utility grid. Among various topologies, the three-level Neutral Point Clamped (NPC) inverter has emerged as a superior alternative to the conventional two-level inverter for medium to high-power applications. Its advantages are well-documented: increased voltage blocking capability, reduced voltage stress on power semiconductors, lower output voltage harmonic distortion, and consequently, smaller and more cost-effective filter components. This makes the three-level solar inverter an ideal candidate for efficient and reliable grid interconnection.
However, the enhanced performance of a three-level solar inverter comes with increased control complexity. A primary technical challenge inherent to the NPC structure is the potential imbalance of the DC-link midpoint voltage. This imbalance can lead to increased low-order harmonics in the output, elevated stress on components, and degraded waveform quality, thereby limiting the system’s reliability and performance. Therefore, developing effective and computationally efficient control strategies that ensure high-quality grid current injection while actively managing the midpoint potential is paramount for advanced solar inverter designs.
This article presents a comprehensive analysis and implementation of a control strategy for a three-phase, three-level NPC solar inverter. Our approach integrates several key techniques: a grid-voltage oriented vector control for power decoupling and regulation, a simplified Space Vector Pulse Width Modulation (SVPWM) algorithm to reduce computational burden, and a dedicated midpoint potential balancing control. We will detail the system architecture, derive the control laws, explain the modulation and balancing mechanisms, and finally present experimental results from a 20kW prototype to validate the proposed strategy’s effectiveness.
System Architecture and Overall Control Strategy
The power stage of the three-level NPC solar inverter under consideration is shown in the topological structure. It consists of a PV array, a DC-link split by two capacitors (C1 and C2), and the three-level NPC bridge. Each phase leg has four power switches (e.g., Sa1-Sa4) with two clamping diodes, creating three output voltage levels relative to the DC-link midpoint n: +Vdc/2, 0, and -Vdc/2. The output is connected to the grid through an L-type filter.
The core of our control strategy for the solar inverter is depicted in the following conceptual block diagram. The system employs a cascaded control structure. The outer loop performs Maximum Power Point Tracking (MPPT). We utilize a combined method of constant voltage and incremental conductance. This MPPT algorithm outputs a DC voltage reference Vdcref* corresponding to the optimal operating point. A PI regulator processes the error between this reference and the measured DC-link voltage Vdc. The output of this PI regulator is the active current reference id*, which dictates the real power flow from the PV array to the grid.
The inner control loop is a current regulator implemented in the synchronous rotating d-q reference frame. A Phase-Locked Loop (PLL) synchronizes the controller with the grid voltage by locking onto the grid voltage vector. By aligning the grid voltage vector along the d-axis, we achieve voltage orientation. In this oriented frame, the grid voltage components become Vd = Vm (the voltage magnitude) and Vq = 0. The instantaneous active power P and reactive power Q are then expressed as:
$$P = \frac{3}{2}(V_{d}i_{d} + V_{q}i_{q}) = \frac{3}{2}V_{m}i_{d}$$
$$Q = \frac{3}{2}(V_{q}i_{d} – V_{d}i_{q}) = -\frac{3}{2}V_{m}i_{q}$$
This elegant decoupling allows independent control of active power (via id) and reactive power (via iq). For unity power factor operation, the reactive current reference iq* is set to zero. The measured three-phase grid currents are transformed to the d-q frame and compared with their references. The errors are fed into PI controllers, whose outputs are the d- and q-axis voltage commands Vd* and Vq*. A grid voltage feedforward term is added to improve dynamic response and disturbance rejection. Finally, these voltage references are processed by the SVPWM module, which incorporates the midpoint balancing algorithm, to generate the gate signals for the three-level solar inverter.
Simplified Three-Level SVPWM Algorithm
The SVPWM technique for a three-level solar inverter is conceptually similar to its two-level counterpart but involves a more complex set of space vectors. The 27 possible switching states map to 19 distinct voltage space vectors in the α-β plane, forming a hexagonal diagram with an inner hexagon. The traditional approach to determining the application times for the nearest three vectors involves sector identification and solving trigonometric equations, which is computationally intensive for a digital controller.
To simplify the implementation, we adopt a decomposition method. The key insight is that the three-level vector diagram can be subdivided into six overlapping two-level hexagons (sectors), each centered on a medium or large vector. We define a region variable S (1 to 6) to identify which of these two-level hexagons the reference voltage vector Vref lies in, and a sector variable N (1 to 6) within that hexagon.
| Condition on Vα and Vβ | Region S |
|---|---|
| Vβ > 0, Vβ < √3 Vα | 1 |
| Vβ > 0, Vβ > √3 Vα, Vβ < √3 |Vα| | 2 |
| Vβ > 0, Vβ > √3 |Vα| | 3 |
| Vβ < 0, |Vβ| < √3 |Vα| | 4 |
| Vβ < 0, |Vβ| > √3 |Vα|, |Vβ| < √3 Vα | 5 |
| Vβ < 0, |Vβ| > √3 Vα | 6 |
Once region S is determined, we find its center vector Vcentral. The original three-level reference Vref is then decomposed by subtracting this center vector:
$$V’_{ref} = V_{ref} – V_{central}$$
The new vector V’ref lies within a standard two-level hexagon centered at the origin. The dwell times T1, T2, T0 for the two active vectors and the zero vector of this two-level system can be calculated using the well-established two-level SVPWM formulas:
$$T_1 = T_s \cdot \left( \frac{\sqrt{3} |V’_{ref}|}{V_{dc}} \right) \sin(60^\circ – \theta’)$$
$$T_2 = T_s \cdot \left( \frac{\sqrt{3} |V’_{ref}|}{V_{dc}} \right) \sin(\theta’)$$
$$T_0 = T_s – T_1 – T_2$$
where Ts is the sampling period, |V’ref| is the magnitude of the decomposed vector, Vdc is the total DC-link voltage, and θ’ is the angle of V’ref within the two-level sector. Crucially, these calculated times (T1, T2, T0) are directly equal to the dwell times of the corresponding three-level vectors that synthesize Vref. This method bypasses the complex trigonometric calculations associated with direct three-level SVPWM, significantly reducing the computational load for the solar inverter’s digital signal processor.
The final step is to arrange the switching sequence. We follow principles to minimize switching losses and ensure smooth transitions: 1) Only one phase leg changes state at a time, and the change is limited to an adjacent voltage level (e.g., 1→0 or 1→2, but not 0→2). 2) The sequence starts and ends with the small vector closest to the reference to minimize narrow pulses. 3) The sequence is symmetric about the center of the switching period. For example, in a specific sector, a typical sequence might be [100] → [200] → [210] → [211] → [210] → [200] → [100].
Midpoint Potential Balancing Control
The midpoint potential imbalance in a three-level NPC solar inverter originates from the current flowing into or out of the neutral point n, which charges or discharges the DC-link capacitors C1 and C2 unevenly. The impact of different voltage vectors on the capacitor voltages is summarized below:
| Vector Type | Example States | Midpoint Current (inp) | Effect on Capacitors |
|---|---|---|---|
| Zero/Large Vector | [222], [111], [000], [2-2-2] | 0 | No effect |
| Medium Vector | [21-1], [12-1], [1-12] | ± ia, ± ib, ± ic | Uncontrollable |
| Positive Small Vector | [100], [010], [001] | +ix (x=a,b,c) | Charges C1, Discharges C2 |
| Negative Small Vector | [211], [121], [112] | -ix (x=a,b,c) | Discharges C1, Charges C2 |
As shown, redundant small vector pairs (e.g., [100] and [211]) have opposite effects on the midpoint current and, consequently, on the capacitor voltages. This redundancy provides the sole means for active control of the midpoint potential. Our balancing strategy dynamically adjusts the dwell time allocation between the two redundant small vectors within a switching period.
Let Tred be the total time allocated to a pair of redundant small vectors in a given switching sequence. We define a control factor ρ (-1 ≤ ρ ≤ 1). The dwell times for the positive small vector (Tup) and the negative small vector (Tdown) are then given by:
$$T_{up} = T_{red} \cdot \frac{(1 + \rho)}{2}$$
$$T_{down} = T_{red} \cdot \frac{(1 – \rho)}{2}$$
The control factor ρ is generated by a controller, typically a simple proportional or PI controller, based on the voltage imbalance ΔVc = VC1 – VC2.
$$ \rho = K_p \cdot (V_{C1} – V_{C2}) $$
If VC1 > VC2 (ΔVc > 0), the controller produces a positive ρ. According to the equations, this increases Tup (e.g., more time for [100]) and decreases Tdown (e.g., less time for [211]). Since state [100] draws current from the midpoint, it discharges C1 and charges C2, thereby reducing the imbalance. Conversely, if VC1 < VC2, a negative ρ is produced, favoring the negative small vector to correct the imbalance. This method is seamlessly integrated into the SVPWM algorithm, ensuring stable operation of the three-level solar inverter.
Experimental Verification
To validate the proposed integrated control strategy for the three-level solar inverter, a 20kW three-phase laboratory prototype was developed. The key parameters of the experimental setup are listed below.
| Component | Parameter | Value / Part Number |
|---|---|---|
| Power Module | IGBT | Infineon 3-level NPC Module |
| DC-Link Capacitor | Capacitance / Voltage | 2200 µF / 900V (each) |
| Grid Filter | Inductance / Current | 3 mH / 35A |
| Controller | Processor | TI DSP TMS320F28335 |
| Control Parameters | d/q-axis PI (Kp, Ki) | 0.05, 8 |
| Control Parameters | DC-link PI (Kp, Ki) | 10, 8 |
| Grid Voltage | Line-to-Line RMS | 380 V |
| Switching Frequency | fsw | 5 kHz |
The system was tested under steady-state grid-connected operation at approximately 15kW of output power. The measured three-phase grid voltages and currents are sinusoidal and in phase, demonstrating successful unity power factor operation as dictated by the control strategy. The current waveform quality is a critical metric for a grid-connected solar inverter. A detailed harmonic analysis was performed on the phase-A current. The Total Harmonic Distortion (THD) for all three phases was measured below 3% (e.g., 2.57%, 2.59%, 2.56%), which complies with stringent grid interconnection standards such as IEEE 1547.
The effectiveness of the midpoint potential balancing control was also verified. The plot shows the AC component of the midpoint voltage difference (VC1 – VC2). The oscillation amplitude is contained within a 5V peak-to-peak range under steady-state load conditions. This demonstrates that the proposed redundant vector time allocation method effectively suppresses the low-frequency drift and minimizes the ripple in the midpoint potential, ensuring stable and reliable operation of the three-level NPC solar inverter.
Conclusion
This article has presented a robust and practical control strategy for a three-level NPC solar inverter. The strategy synergistically combines grid-voltage oriented vector control for decoupled active and reactive power regulation, a simplified SVPWM algorithm that reduces computational complexity by decomposing three-level vectors into two-level equivalents, and an active midpoint potential balancing control based on the dynamic adjustment of redundant small vector dwell times. The integration of voltage feedforward further enhances the system’s dynamic response to grid disturbances.
The theoretical analysis was substantiated by experimental results from a 20kW prototype. The solar inverter achieved high-performance grid interconnection, characterized by sinusoidal currents with low THD (<3%), unity power factor operation, and well-regulated midpoint voltage. The proposed methods offer a balanced solution between performance and implementation complexity, making them highly suitable for the development of efficient and reliable medium- to high-power three-level solar inverters. Future work may explore the extension of these control principles to other multilevel topologies or the integration of advanced grid-support functions like fault ride-through and harmonic compensation.