The proliferation of distributed energy resources has led to a significant increase in the penetration of power electronic converters, particularly utility interactive inverters, into the grid. This integration poses substantial challenges to grid stability, especially under non-ideal conditions characterized by weak grids. A weak grid, typically defined by a low short-circuit ratio (SCR), exhibits non-negligible grid impedance, voltage fluctuations, and background harmonics. These factors severely degrade the performance of conventional control strategies for utility interactive inverters, leading to poor robustness, increased current harmonics, and potential instability. Therefore, developing advanced control algorithms that ensure reliable and high-quality operation of utility interactive inverters under weak grid conditions is of paramount importance.
Among various control paradigms, Finite Control Set Model Predictive Control (FCS-MPC) has garnered considerable attention for controlling utility interactive inverters. Its appeal lies in its intuitive concept, fast dynamic response, ability to handle nonlinearities, and direct inclusion of constraints. The core principle involves using a discrete model of the system to predict future behavior for all possible switching states of the inverter. An optimal switching state is then selected by minimizing a predefined cost function. However, the practical application of traditional FCS-MPC to utility interactive inverters, especially multi-level topologies like the Neutral-Point Clamped (NPC) inverter, faces several critical challenges in weak grids:
| Challenge | Description | Impact on Utility Interactive Inverter | |
|---|---|---|---|
| Parameter Sensitivity | FCS-MPC relies heavily on an accurate system model (filter inductance L and resistance R). Weak grids have variable and often unknown impedance parameters. | Model mismatch leads to steady-state error, increased harmonic distortion, and can even cause instability. | |
| Computational Burden | The number of candidate switching states grows exponentially with the inverter level. An NPC inverter has 27 states, requiring 27 predictions per cycle. | Limits the achievable switching frequency and control bandwidth, affecting the quality of the injected grid current. | |
| Limited Control Resolution | Only the intrinsic voltage vectors of the inverter are available, leading to a limited set of possible output voltages. | Causes significant current ripple and higher total harmonic distortion (THD) in the grid current. | |
| Neutral-Point Potential (NPP) Fluctuation | Inherent to NPC topology, the DC-link capacitor voltages can become unbalanced due to the neutral-point current. | Causes increased voltage stress on switches, degraded output waveform quality, and can trip the protection system. |
To address the critical issue of parameter sensitivity, data-driven or Model-Free Predictive Control (MFPC) has been proposed as a promising alternative for controlling utility interactive inverters. This approach eliminates the dependency on explicit system parameters. Instead, it uses measured current differences from previous control cycles to estimate the future current trajectory. The current prediction for the (k+2)th instant, accounting for computational delay, can be expressed based on the applied switching state S(m):
$$ \mathbf{i}(k+2) = \mathbf{i}(k) + \Delta \mathbf{i}_{p1}|_{S(m)} + \Delta \mathbf{i}_{p1}|_{S(m)} $$
Here, $$ \Delta \mathbf{i}_{p1}|_{S(m)} $$ represents the memorized current difference observed when the switching state S(m) was applied previously. The cost function, typically the absolute error between the reference and predicted current, is evaluated for all states:
$$ J_m = |\mathbf{i}^*(k+2) – \mathbf{i}(k+2)| $$
While this data-driven MFPC significantly enhances the robustness of the utility interactive inverter against parameter variations, its performance is still limited by the finite and coarse set of available voltage vectors, resulting in suboptimal current tracking and higher THD.
My proposed strategy, termed Virtual-Vector based Model-Free Predictive Control (V-MFPC), is designed to overcome the combined limitations of computational burden, limited resolution, and NPP fluctuation for NPC-type utility interactive inverters in weak grids. The core innovations are the construction of Virtual Voltage Vectors (VVVs) and a two-stage optimization process with vector pre-selection.
The first step involves constructing a richer set of output voltages. By applying different sequences of the basic switching states within one control period according to the volt-second balance principle, numerous VVVs can be synthesized. This effectively increases the granularity of the output voltage selection, allowing the controller to more accurately approximate the desired reference voltage. For instance, a virtual vector like v34 can be synthesized as an average of two adjacent medium vectors: $$ \mathbf{v}_{34} = 0.5\mathbf{v}_4 + 0.5\mathbf{v}_6 $$. Crucially, different switching sequences can produce the same average voltage vector but result in different neutral-point current profiles. This property is later exploited for NPP balancing.
However, constructing a large set of VVVs (e.g., 61 unique vectors) exacerbates the computational load. To solve this, I introduce a two-stage optimal vector selection process. In the first stage, the cost function is evaluated only for the six large vectors of the inverter. The large vector yielding the minimum cost identifies the sector where the optimal vector is most likely located. In the second stage, only the 12-15 pre-defined candidate vectors (including basic and virtual vectors) within and around this sector are evaluated. This pre-selection method dramatically reduces the number of cost function evaluations from 61 to about 18, while guaranteeing that the globally optimal vector is not missed, making the algorithm suitable for real-time implementation on a utility interactive inverter.
| Optimal Large Vector (Stage 1) | Candidate Vectors for Detailed Evaluation (Stage 2) |
|---|---|
| v16 | v2, v4, v17, v27, v28, v34, v39, v40, v41, v57, v58, v69 |
| v18 | v2, v6, v17, v19, v29, v34, v35, v42, v43, v44, v59, v60 |
| … | … |
Finally, after the optimal voltage vector is selected, the specific switching sequence used to synthesize it is chosen to minimize NPP fluctuation. The change in the DC-link capacitor voltage imbalance $$ \Delta v $$ is evaluated for the candidate switching sequences using the discrete model:
$$ \Delta v(k+1) = \Delta v(k) + \frac{T_s}{C} \sum_{x=a,b,c} i_{xo} (1 – |S_x|) $$
Where $$ i_{xo} $$ is the phase output current and $$ S_x $$ is the switching state (P=1, O=0, N=-1). The sequence that minimizes the future imbalance $$ |\Delta v(k+1)| $$ is applied. This integrated approach ensures that the utility interactive inverter not only tracks the current reference accurately but also maintains a balanced DC-link.
The complete control structure of the proposed V-MFPC for a utility interactive inverter is summarized as follows. In every control cycle:
- Sample grid currents $$ i(k) $$ and capacitor voltages.
- Apply the optimal switching sequence from the previous cycle.
- Update the table of current differences $$ \Delta i_{p1}|_{S(m)} $$ for all 61 switching states based on the latest measurements.
- Predict the future reference current $$ i^*(k+2) $$ using angle compensation.
- Stage 1 (Pre-selection): Calculate the cost $$ J $$ for the six large vectors using the data-driven prediction model. Select the sector.
- Stage 2 (Refinement): Calculate the cost $$ J $$ for the ~12 candidate vectors in that sector. Select the optimal voltage vector.
- Sequence Selection: For the chosen optimal vector, evaluate $$ \Delta v(k+1) $$ for all valid switching sequences that synthesize it. Apply the sequence that minimizes the NPP fluctuation.
To validate the effectiveness of the proposed V-MFPC strategy for utility interactive inverters, a comprehensive simulation study was conducted, comparing it against the conventional data-driven MFPC. The system parameters for the NPC utility interactive inverter are listed below.
| Parameter | Symbol | Value |
|---|---|---|
| DC-link Voltage | $$ U_{dc} $$ | 800 V |
| DC-link Capacitance | $$ C_1, C_2 $$ | 4700 µF |
| Grid Phase Voltage (RMS) | $$ E $$ | 220 V |
| Inverter-side Filter Inductance | $$ L $$ | 5 mH |
| Grid Equivalent Inductance (Weak Grid) | $$ L_g $$ | 0.1 mH |
| Equivalent Resistance | $$ R $$ | 0.2 Ω |
| Sampling & Control Period | $$ T_s $$ | 25 µs |
Steady-State Performance: Under a steady grid current reference of 20 A, the V-MFPC demonstrates superior current tracking compared to conventional MFPC. The expanded set of virtual vectors allows for a more precise approximation of the sinusoidal reference. Spectral analysis confirms this improvement. The Total Harmonic Distortion (THD) of the grid current with V-MFPC was 3.66%, compared to 4.39% with conventional MFPC, representing a significant 17% reduction. Furthermore, the NPP fluctuation was drastically suppressed. The maximum capacitor voltage deviation was reduced from 6.9 V with MFPC to only 2.2 V with V-MFPC, highlighting the effectiveness of the active switching sequence selection in the utility interactive inverter control loop.
Dynamic Performance: The dynamic response to a step change in the current reference from 15 A to 20 A is equally fast for both methods. More importantly, under a weak grid voltage sag condition (where the grid voltage amplitude dropped to 80% at t=0.1s), the V-MFPC controlled utility interactive inverter maintained stable operation with nearly unchanged current amplitude and a slightly improved THD from 3.78% to 3.42%. This demonstrates the inherent robustness and grid fault ride-through capability of the data-driven approach.
Parameter Robustness: To simulate the uncertain and variable impedance of a weak grid, the filter parameters of the utility interactive inverter were intentionally detuned. The following table summarizes the performance under parameter variations.
| Test Case | Parameter Change | Grid Current THD | Observation |
|---|---|---|---|
| Nominal | $$ L = 5 \text{mH}, R = 0.2 \Omega $$ | 3.66% | Baseline performance. |
| Inductance Variation | $$ L = 7.5 \text{mH} (+50\%) $$ | 3.12% | Stable operation, no distortion, THD even improved due to increased filtering. |
| Resistance Variation | $$ R = 0.3 \Omega (+50\%) $$ | 3.71% | Stable operation, negligible impact on THD and tracking accuracy. |
| Grid Voltage Sag | $$ E \rightarrow 0.8E $$ (Amplitude) | 3.42% | Stable synchronization, maintained current injection. |
These results conclusively show that the proposed V-MFPC strategy maintains excellent performance without any need for parameter re-tuning. The utility interactive inverter controlled by V-MFPC is inherently robust to the parameter uncertainties and disturbances prevalent in weak grids, a critical advantage over model-based predictive control methods.
In conclusion, the Virtual-Vector based Model-Free Predictive Control (V-MFPC) strategy developed in this work provides a comprehensive solution for enhancing the performance of NPC-type utility interactive inverters in challenging weak grid environments. By synergistically combining the robustness of a data-driven prediction model, the enhanced resolution of virtual voltage vectors, a computationally efficient two-stage vector pre-selection algorithm, and an active neutral-point potential balancing mechanism, the proposed method effectively addresses the key limitations of existing approaches. The utility interactive inverter achieves lower grid current harmonic distortion, significantly reduced DC-link capacitor voltage imbalance, and maintains excellent dynamic and steady-state performance even under significant grid impedance variations and voltage disturbances. This makes V-MFPC a highly suitable and reliable control strategy for the next generation of grid-supportive utility interactive inverters operating in weak and variable grid conditions.