The proliferation of renewable energy has fundamentally reshaped the global power landscape. Among these sources, photovoltaic (PV) generation, prized for its cleanliness and sustainability, stands as one of the fastest-growing segments. The grid-tied inverter serves as the critical interface connecting a PV array to the public utility grid, and its performance directly dictates the efficiency, safety, and power quality of the entire generation system. Within this context, leakage current emerges as a pervasive and critical challenge. This current, which flows through parasitic capacitances between the PV panels and the ground, not only poses potential safety hazards and electromagnetic interference (EMI) issues but can also degrade system efficiency and insulation. As PV installations scale, the importance of accurate leakage current monitoring and effective suppression becomes paramount for reliable grid integration.
The topology of a standard three-phase two-level voltage-source grid-tied inverter is foundational to this discussion. It typically consists of a DC-link capacitor, six power switches forming three phase legs, and an L-type output filter. The parasitic capacitance ($C_{PV}$) between the PV array and earth ground provides the path for common-mode leakage current ($i_{cm}$). The primary mechanism for leakage current generation is the high-frequency common-mode voltage ($v_{com}$) produced by the inverter’s switching actions, applied across this parasitic capacitance. The relationship can be simplified as:
$$ i_{cm}(t) \approx C_{PV} \frac{d v_{com}(t)}{dt} $$
Therefore, controlling the spectral content and magnitude of $v_{com}$ is essential for leakage current suppression.
Limitations of Conventional Leakage Current Monitoring
Accurate monitoring is the first step towards effective suppression. A common industrial practice involves routing all three phase conductors through the window of a single current transformer (CT). According to Ampère’s circuital law, the line integral of the magnetic field ($\mathbf{H}$) around a closed loop is equal to the total enclosed current:
$$ \oint \mathbf{H} \cdot d\mathbf{l} = i_{ga} + i_{gb} + i_{gc} = i_{cm} $$
In an ideal scenario with perfect symmetry, the magnetic field sensed by the CT core would be proportional only to the leakage current. However, in practice, the spatial arrangement of the three phase conductors within the sensor window is often asymmetric (e.g., triangular or non-equidistant). This asymmetry causes the magnetic fields generated by the large, balanced three-phase grid currents ($i_{ga}, i_{gb}, i_{gc}$) to not perfectly cancel in the external region where the CT core is located. Consequently, the sensor output becomes contaminated by a component proportional to the load current, leading to significant measurement error, especially under high-power conditions. This effect intensifies if the CT core approaches saturation due to the high instantaneous sum of phase currents.
Enhanced Monitoring Using a Coaxial Conductor Assembly
To overcome the spatial asymmetry problem, an improved monitoring method utilizes a specialized coaxial cable assembly. This assembly comprises three concentric conductors: an inner core, a middle layer, and an outer shield. Each of the three phase currents is carried by one of these concentric layers. Due to the perfect cylindrical symmetry of this structure, the magnetic fields produced by the three phase currents cancel each other completely in the space outside the outermost conductor. The magnetic field at any radial distance $r$ from the center is then solely due to the net current enclosed, which is the leakage current $i_{cm}$.
$$ H_{\phi}(r) = \frac{i_{cm}}{2\pi r} \quad \text{(for $r$ > outer conductor radius)} $$
A current sensor (e.g., a Rogowski coil or a CT) placed around this coaxial assembly will therefore produce an output voltage $v_{sens}$ that is a clean, direct measurement of the leakage current, independent of the magnitude of the three-phase load currents:
$$ v_{sens}(t) = k \cdot \frac{d i_{cm}(t)}{dt} \quad \text{or} \quad v_{sens}(t) = R_{burden} \cdot i_{cm}(t) $$
where $k$ is the sensor sensitivity. This method eliminates measurement inaccuracies caused by conductor placement and greatly reduces the risk of core saturation, providing a high-fidelity signal for both protection and control purposes in the grid-tied inverter system.
The Challenge of Leakage Current in Model Predictive Control
Finite Control Set Model Predictive Control (FCS-MPC) has gained traction in grid-tied inverter applications due to its intuitive design, fast dynamic response, and ability to handle multiple constraints. The standard approach for current control involves using a discrete-time model of the output filter to predict the future behavior of the grid current for each possible voltage vector ($\mathbf{v}_i$) generated by the inverter. For an L-filter, the continuous-time model is:
$$ L \frac{d\mathbf{i_g}}{dt} = \mathbf{v_i} – \mathbf{e_g} $$
where $\mathbf{i_g}$ is the grid current vector, $\mathbf{e_g}$ is the grid voltage vector, and $L$ is the filter inductance. Discretizing this model with a sampling period $T_s$ yields the prediction equation:
$$ \mathbf{i_g}^p(k+1) = \mathbf{i_g}(k) + \frac{T_s}{L} \left( \mathbf{v_i}(k) – \mathbf{e_g}(k) \right) $$
A cost function $g$ minimizes the error between the predicted current and its reference:
$$ g = \left[ i_{g,\alpha}^p(k+1) – i_{g,\alpha}^{ref}(k+1) \right]^2 + \left[ i_{g,\beta}^p(k+1) – i_{g,\beta}^{ref}(k+1) \right]^2 $$
The voltage vector minimizing $g$ is applied.
However, this classical FCS-MPC has two major shortcomings concerning leakage current in a grid-tied inverter:
- Lack of Leakage Current Consideration: The cost function only tracks grid current performance. The switching states selected to optimize current tracking may generate large common-mode voltages, excelling the parasitic capacitance and leading to high leakage current.
- Parameter Sensitivity: The prediction accuracy is critically dependent on the precise knowledge of the filter inductance $L$. In reality, $L$ varies with temperature, current level, and manufacturing tolerances. A mismatch between the model parameter $L_m$ and the actual value $L_a$ causes a steady-state current error and, more critically for this discussion, distorts the prediction of the common-mode state. The erroneous prediction can lead the controller to select suboptimal switching vectors that inadvertently increase the actual leakage current. The prediction error $\Delta \mathbf{i_g}^{err}$ due to parameter mismatch $\Delta L = L_a – L_m$ is approximately:
$$ \Delta \mathbf{i_g}^{err}(k+1) \approx \frac{T_s \cdot \Delta L}{L_m^2} \left( \mathbf{v_i}(k) – \mathbf{e_g}(k) \right) $$
This error directly feeds into the cumulative error for the leakage current prediction.
| Control Strategy | Primary Control Objective | Parameter Dependence | Explicit Leakage Current Suppression |
|---|---|---|---|
| Classical PI with PWM | Grid Current Tracking | High (PI Tuning) | No (Relies on Topology) |
| Classical FCS-MPC | Grid Current Tracking | Very High (Model Accuracy) | No |
| FCS-MPC with CMV Penalty | Grid Current + CMV Reduction | Very High | Indirect (via CMV) |
| Data-Driven FCS-MPC (Proposed) | Grid Current + Direct Leakage Current Suppression | Very Low | Yes |
Data-Driven Predictive Control for Simultaneous Current Tracking and Leakage Suppression
The proposed enhanced control method for the grid-tied inverter addresses both shortcomings by integrating a data-driven model identification step within the FCS-MPC framework and formulating a multi-objective cost function.
1. Data-Driven Model Identification:
Instead of relying on a fixed physical model $L$, a data model is built online. Rearranging the discretized system equation yields:
$$ \Delta \mathbf{i_g}(k) = \mathbf{i_g}(k) – \mathbf{i_g}(k-1) = \mathbf{\Theta}(k) \cdot \left( \mathbf{v_i}(k-1) – \mathbf{e_g}(k-1) \right) $$
where $\mathbf{\Theta}(k) = T_s / L_a$ is an unknown time-varying parameter that encapsulates the true system dynamics and unmodeled disturbances. A Recursive Least Squares (RLS) estimator with forgetting factor $\lambda$ (typically 0.95-0.99) is used to identify $\mathbf{\Theta}$ in real-time:
$$
\begin{aligned}
\mathbf{\phi}(k) &= \mathbf{v_i}(k-1) – \mathbf{e_g}(k-1) \\
\mathbf{y}(k) &= \Delta \mathbf{i_g}(k) \\
e(k) &= \mathbf{y}(k) – \hat{\mathbf{\Theta}}(k-1) \mathbf{\phi}(k) \\
\mathbf{K}(k) &= \frac{\mathbf{P}(k-1) \mathbf{\phi}(k)}{\lambda + \mathbf{\phi}^T(k) \mathbf{P}(k-1) \mathbf{\phi}(k)} \\
\mathbf{P}(k) &= \lambda^{-1} \left( \mathbf{I} – \mathbf{K}(k) \mathbf{\phi}^T(k) \right) \mathbf{P}(k-1) \\
\hat{\mathbf{\Theta}}(k) &= \hat{\mathbf{\Theta}}(k-1) + \mathbf{K}(k) e(k)
\end{aligned}
$$
This estimator continuously adapts $\hat{\mathbf{\Theta}}$ to match the actual plant, making the control loop robust to parameter variations, aging, and nonlinearities.
2. Integrated Leakage Current Prediction:
The leakage current is defined as the sum of the three-phase currents: $i_{cm} = i_{ga} + i_{gb} + i_{gc}$. In the $\alpha\beta$ frame, this translates to a specific projection. Therefore, predicting the leakage current necessitates predicting the sum of the three individual phase currents. Using the data-driven model, the future current for phase $x$ under a candidate voltage vector $\mathbf{v}_i^j$ is predicted as:
$$ i_{g,x}^p(k+1 | \mathbf{v}_i^j) = i_{g,x}(k) + \hat{\Theta}_x(k) \cdot \left( v_{i,x}^j(k) – e_{g,x}(k) \right) $$
The corresponding leakage current prediction is then:
$$ i_{cm}^p(k+1 | \mathbf{v}_i^j) = \sum_{x=a,b,c} i_{g,x}^p(k+1 | \mathbf{v}_i^j) $$
The reference for the leakage current is ideally zero: $i_{cm}^{ref} = 0$.
3. Multi-Objective Cost Function:
The core innovation is the design of a cost function that simultaneously penalizes grid current tracking error and leakage current magnitude:
$$ g^j = \underbrace{\left[ i_{g,\alpha}^p(k+1 | \mathbf{v}_i^j) – i_{g,\alpha}^{ref}(k+1) \right]^2 + \left[ i_{g,\beta}^p(k+1 | \mathbf{v}_i^j) – i_{g,\beta}^{ref}(k+1) \right]^2}_{\text{Grid Current Tracking}} + \lambda_{leak} \cdot \underbrace{\left[ i_{cm}^p(k+1 | \mathbf{v}_i^j) \right]^2}_{\text{Leakage Current Suppression}} $$
The weighting factor $\lambda_{leak}$ provides a tunable trade-off. A higher $\lambda_{leak}$ prioritizes leakage current suppression, potentially at a slight cost to current tracking bandwidth or harmonic distortion. The optimal voltage vector $\mathbf{v}_i^{opt}$ is selected as:
$$ \mathbf{v}_i^{opt} = \arg \min_{\mathbf{v}_i^j \in \mathcal{V}} \, g^j $$
where $\mathcal{V}$ is the set of all admissible switching vectors for the grid-tied inverter.
Overall Control Architecture and Performance Summary
The complete control system for the advanced grid-tied inverter synthesizes the monitoring and suppression techniques. The coaxial sensor provides a high-accuracy measurement of $i_{cm}$, which is used both for protective relay functions and as a feedback signal within the data-driven predictive controller. The controller executes the following sequence every sampling period:
- Measurement: Sample grid currents $\mathbf{i_g}(k)$, grid voltages $\mathbf{e_g}(k)$, and leakage current $i_{cm}(k)$.
- Identification: Update the data-driven model parameter $\hat{\mathbf{\Theta}}(k)$ using the RLS algorithm based on the previous applied voltage vector.
- Prediction: For each candidate voltage vector $\mathbf{v}_i^j$, calculate the future grid currents $\mathbf{i_g}^p(k+1)$ and the future leakage current $i_{cm}^p(k+1)$ using the identified $\hat{\mathbf{\Theta}}(k)$.
- Evaluation & Selection: Compute the cost $g^j$ for each vector. Select and apply the vector $\mathbf{v}_i^{opt}$ that minimizes $g$.
| Aspect | Conventional Approach (CT + Classical MPC) | Proposed Approach (Coaxial Sensor + Data-Driven MPC) |
|---|---|---|
| Monitoring Accuracy | Low. Highly sensitive to conductor placement and load current magnitude. Prone to saturation. | High. Immune to phase current magnitude and spatial asymmetry due to concentric design. |
| Control Robustness | Low. Performance degrades significantly with $L$ filter parameter mismatch. | High. Online parameter identification maintains accurate prediction despite parameter drift. |
| Leakage Suppression | Passive/Indirect. Relies on topology or CMV reduction, not explicitly controlled. | Active/Direct. Explicitly minimized via a dedicated term in the cost function. |
| Grid Current THD | Can be low with accurate parameters, but may increase under mismatch or when targeting low CMV. | Consistently low. The controller finds an optimal trade-off between tracking and suppression. |
| Computational Burden | Moderate (standard predictions). | Higher (adds RLS identification and leakage prediction). |
The mathematical foundation of the suppression mechanism lies in the controller’s ability to evaluate the common-mode voltage impact of each switching state implicitly through the leakage current prediction. Vectors that generate low $i_{cm}^p$ are favored by the cost function. The data-driven model ensures this evaluation remains accurate even as the physical system changes, allowing the grid-tied inverter to maintain optimal performance across its operating range. The synergy between the robust monitoring hardware and the adaptive, multi-objective control software creates a comprehensive solution for managing leakage current, enhancing the safety, efficiency, and power quality of photovoltaic generation systems connected to the grid.
In conclusion, the integration of a coaxial conductor-based monitoring system with a data-driven, multi-objective model predictive control strategy presents a significant advancement for grid-tied inverter technology. This approach directly tackles the core limitations of existing methods: measurement inaccuracy from spatial asymmetry and control performance degradation from parameter uncertainty. By providing a precise measurement of leakage current and incorporating its direct suppression as a fundamental control objective within a robust predictive framework, this method ensures that modern grid-tied inverters can achieve superior performance, meeting stringent safety and grid code requirements while maximizing energy yield from photovoltaic installations.