Medium Voltage Conversion Systems with Integrated Galvanic Isolation for Hybrid Photovoltaic Plants

Abstract

The demand for a more sustainable energy system is driving the development of renewable energy sources and green technologies within the electrical sector. However, integrating these technologies is challenging due to the increased complexity of the system components and grid architectures. This paper provides an overview of power electronic conversion systems that facilitate the connection of renewable energy sources (photovoltaic power plants) and direct-current energy storage systems to three-phase medium-voltage alternating-current grids. This paper presents a comprehensive study of the state-of-the-art converter architectures and proposes modifications and technological alternatives, providing insight into the future development of grid-interface power converters for hybrid energy systems.

1. Introduction

Climate change is one of today’s most widely discussed topics, particularly within the electricity sector, where production and consumption can significantly contribute to greenhouse-gas emissions [1]. Electricity generation has relied on fossil fuels such as coal, oil, and gas as primary energy sources for several decades. This practice significantly contributes to global warming by releasing large quantities of carbon dioxide, accounting for more than 40% of global emissions [1]. Therefore, the development and implementation of more sustainable energy systems represent a critical priority for modern power systems. In this context, the integration of renewable energy sources (RES) and other types of long-term energy storage technologies (e.g., hydrogen) is considered as promising solution, although it introduces significant technological and environmental challenges [2].

In recent years, governments have supported the rapid expansion of clean energy in electrical systems by introducing long-term strategies and supportive policies aimed at promoting the development and deployment of renewable energy sources. Solar power generation has maintained a high growth rate and has approximately doubled every three years since its introduction to the energy market, largely benefiting from policy-driven deployment targets. Photovoltaic (PV) installations have experienced significant growth, with an annual increase of around 25%, reaching a total installed capacity exceeding 1.8 TW by 2024 [3]. Approximately 90% of PV systems installed worldwide are grid-connected, typically through medium-voltage distribution networks that transfer the generated power to the electrical grid [4].

However, RES, such as PV and wind power, are intermittent and strongly dependent on weather conditions. Hydrogen is emerging as a promising solution for long-term energy storage and energy transport in response to this challenge. Excess energy can be stored in the form of hydrogen using water electrolysis (ELYZ). Conversely, the energy contained in hydrogen can be recovered using fuel cells (FCs) in a carbon-free manner during periods of increased electricity demand.

RES, FCs, and ELYZ systems have different electrical characteristics, particularly in terms of current levels, voltage ratings, and insulation constraints. Power converters, therefore, serve as interfaces between energy sources and electrical networks and play a key role in enabling efficient power transmission. For this reason, research into DC grids is crucial for improving energy conversion efficiency. As shown in Figure 1, DC sources, storage systems, and loads can be connected to a common DC grid to facilitate power exchange and reduce the number of conversion steps. This DC grid structure can have different grounding configurations, namely unipolar asymmetric, unipolar symmetric or bipolar [5,6,7]. Each grounding scheme has different design and implementation considerations depending on the DC network architecture. The details of these schemes can be found in [5,6,7] and are not further addressed in this paper.

This paper focuses on energy conversion topologies for connecting RES, FCs, and ELYZ systems operating at low-voltage/medium-voltage direct-current (LVDC/MVDC) levels to the medium-voltage alternating-current (MVAC) grid. The present article adopts an original approach by establishing a concise scope for feasible solutions based on recent technological advances in power electronic transformers, with reference to the contribution of SiC devices in galvanically isolated converters. In addition, it covers a wide range of research issues addressed in this specific field.

Furthermore, it introduces an original conversion method in the cascaded H-bridge architecture, which opens the way to the use of 4-quadrant switches. This approach offers significant opportunities for advanced power conversion that have not yet been fully explored in the existing literature or demonstrated in practical applications.

The paper is divided into six sections. After the introduction and the classification of possible configurations, the subsequent sections provide more detailed information on the power electronic structures and possible control schemes. These sections also analyze the advantages and limitations of each solution before presenting an overall comparison and conclusion.

2. Classification of Possible Configurations

Currently, the DC network used to integrate PV systems and other DC sources is limited to a voltage level of 1.5 kV⎓ in Europe and 2 kV⎓ in China. In Europe, this voltage level lies at the boundary between LVDC and MVDC. It is already standardized and compatible with several industrial applications. Furthermore, renewable energy technologies, such as DC sources or hydrogen-based loads (FCs and ELYZ), can be connected to a 1.5 kV⎓ DC bus without requiring an additional isolation stage. Recently, Silicon Carbide (SiC) power semiconductors with a voltage rating of 2.3 kV have become commercially available. These devices represent a technological pathway toward improving the efficiency and power density of 1.5 kV⎓ applications [8]. This development simplifies power electronic converter design by enabling two-level topologies with high performance [9]. For this reason, the present study focuses mainly on the converter interfacing the 1.5 kV⎓ DC collector with the 20 kV three-phase distribution grid. As a result, the DC/DC conversion stages between the RES, FCs, and ELYZ systems and the DC grid are not analyzed in detail. This includes the low-voltage, high-current DC/DC converters used in FC and ELYZ applications. Nevertheless, several recent publications [10,11,12] show that this converter stage is receiving increasing attention in the literature, especially regarding current-control strategies, low-impedance interfacing, and high-current magnetic design.

Many technological solutions are currently under study, as shown in Figure 2. The first category includes solutions using transformers to connect energy sources to the MVAC grid while ensuring galvanic isolation. These topologies can be classified according to whether a low-frequency transformer (LFT) or a medium-frequency transformer (MFT) is connected to the MVAC grid. In the latter case, also known as Solid-State Transformer (SST) technology [13], the MFT is integrated into the power electronic conversion stages to ensure isolation between the AC and DC power sources. The second category corresponds to transformerless solutions, meaning that the generators must have reinforced insulation [14,15,16,17]. Typical examples reported in [18,19,20] have encouraged researchers to investigate this technology for high-power transformerless PV plants. The use of power electronics without isolation between the MVAC network and the sources is an interesting technological option, but it is hampered by major technological and regulatory obstacles [21], which place these solutions on a longer-term development horizon. It is conceivable that these issues could evolve in a manner similar to that observed for very high-power industrial electric motors [22].

The typical conversion chain based on an LFT, which operates at 50/60 Hz, has been used since the early days of solar power plants. As shown in Figure 3, the RES is connected to a DC/DC converter (typically a boost converter), which is then connected to a Voltage-Source Inverter (VSI) to convert DC power into AC power. The VSI is then connected to a step-up transformer to match the grid voltage level. This system is the most common method of connecting a megawatt-scale PV plant to the grid. It simplifies the conversion process, as it relies on a conventional power conversion chain and does not require a complex control scheme. However, its main disadvantage is that the LFT leads to a low-power-density system due to its significant weight and size. The core cross-sectional area of a transformer is inversely proportional to the frequency of the voltage applied to it; therefore, a higher frequency results in a smaller core [23]. Furthermore, a smaller core also means fewer turns of copper wire, resulting in reduced copper losses. A comparison between an LFT and an MFT is shown in

Table 1. Comparison between a LFT and a MFT [24].

	60 Hz Transformer	50 kHz Transformer
Power	25 kVA	31 kVA
Voltage ratio	5 kV/240 V	4 kV/400 V
Efficiency	98.3%	99.5%
Volume	230 dm3	2.8 dm3
Mass	160 kg	5.1 kg

[24]. As can be seen, for the same apparent power level, the volume and mass of the MFT are drastically reduced compared with those of the LFT, while the efficiency is improved. Although these values correspond to relatively early technologies, they already illustrate the superiority of the MFT over the LFT. With the advancement of magnetic materials and wide-bandgap (WBG) semiconductors, MFT technology has achieved significant performance improvements in recent years [25].

Furthermore, when more renewable energy sources are integrated on the DC side, the use of an LFT combined with a direct three-phase inverter, as shown in Figure 3, limits the transferable power. For example, to transfer several megawatts, the inverter would need to handle currents of several thousand amperes, which would result in increased conduction losses and would require a large amount of cabling [26] between the inverter and the LFT.

However, insulation of the MFT remains a critical practical challenge, especially under high-voltage Pulse Width Modulation (PWM) stress. Several studies have addressed this issue to optimize the design of the MFT while attempting to mitigate problems related to insulation degradation, partial discharges, and their effects on insulation lifetime [27,28,29,30]. Although an optimal solution requires a complex design trade-off among several parameters, significant progress has been made in the development of high-performance and reliable MFTs. Moreover, the optimal solution may vary depending on the specific application in which the MFT is implemented [28].

Due to recent advances in modern high-speed WBG power transistors, significant progress has been made in the development of Power Electronic Transformers (PETs) and Solid-State Transformers (SSTs) [31]. When combined with power converters, these transformers can interface RES with the MVAC grid. The use of WBG semiconductors such as SiC increases the power density of the entire conversion system. Consequently, additional converter configurations become feasible. The first topology reported in the literature is the MVAC collector architecture [32,33]. This architecture enables the use of Power Electronics Building Blocks (PEBBs), a modular approach that simplifies the design stage [34].

Moreover, the growing interest in connecting RES to MVDC grids has led to increased attention being paid to SST-based architectures, which implement an intermediate MVDC collector using a centralized inverter [35]. Other possible variations can be achieved by combining and adapting MVDC and MVAC collector architectures to take advantage of both topologies [36].

3. Intermediate MVDC Collector Architecture with Three-Phase Injection

3.1. General System Overview

The MVAC collector architecture [26,31], one of the first proposed architectures, suffers from a major drawback related to power imbalance among the three phases. To address this issue, a new topology featuring an intermediate MVDC collector has been proposed [35,37,38,39]. As shown in Figure 4, each RES or energy storage unit is connected to a group of DC/DC converters and to a common DC bus. This bus can operate at either low-voltage (LVDC) or medium-voltage (MVDC) levels, with a nominal voltage of 1.5 kV⎓ or other values depending on the system. To avoid confusion with the intermediate MVDC collector, this bus is referred to as the LVDC bus. A group of isolated DC/DC converters arranged in an input-parallel output-series (IPOS) configuration is then connected from this bus to the MVDC collector. The MVDC collector is followed by a centralized VSI, which is necessarily a multilevel converter due to the voltage limitations of commercially available semiconductors. Each DC/DC converter contains an MFT. This relaxes the dielectric constraint on the source side [14] and enables the series connection of PEBB outputs to achieve a higher DC voltage. This configuration can also address current-balancing issues using the MVDC bus.

On the DC source side, all DC sources are connected to a common bus, which distributes power equally among the IPOS groups (Figure 4). As shown in this figure, each RES and FC/ELYZ system must pass through a DC/DC converter implementing a Maximum Power Point Tracking (MPPT) algorithm before being connected to the LVDC collector. This assumes that the sources are compatible with medium-voltage insulation constraints. This configuration enables equal power sharing, or current sharing, among the IPOS groups. Consequently, all the isolated DC/DC converters can be designed with identical voltage and current ratings.

However, additional issues related to the DC collector arise due to the higher dielectric stress that the MFTs [40], as well as the protective devices and cables, must withstand [21]. For instance, if the system is connected to a 20 kV three-phase grid, the MVDC voltage must be at least 40 kV⎓ when a conventional inverter is used [41]. This would not only exacerbate the issues but also raise concerns regarding the DC capacitors, since standards for MVDC systems are not yet fully developed [42]. To mitigate these drawbacks, a modified configuration is proposed (see Figure 5). As shown, the MVDC bus is now divided into two sections with middle point grounding (GND): +20 kV⎓/GND and GND/−20 kV⎓, which reduces the dielectric stress on the MFT.

3.2. DC/DC Isolated Converter

For unidirectional PV power transfer, the isolated DC/DC converter can be implemented using either a Dual Active Bridge (DAB) converter [43] or a Single Active Bridge (SAB) converter [37,38]. However, for bidirectional power exchange, SAB-based solutions cannot be considered. Since this converter requires galvanic isolation and has two voltage-source ports, the suitable topology involves two sub-converters connected through an inductive intermediate stage and an MFT. The two most promising topologies are shown in Figure 6 and correspond to either a DAB with a series-resonant (SR) high-frequency LC tank or a DAB without an SR high-frequency LC tank. Both topologies use a full-bridge structure. First, this structure can generate a zero-output-voltage state, thereby providing additional degrees of freedom for advanced modulation methods. Second, it delivers a voltage that is twice that of a half-bridge circuit. Therefore, for the same power transfer and voltage ratings, the current stress is lower in the former case than in the latter, which is an essential factor for the implementation of WBG devices [44].

Although the DAB-SR topology exhibits slightly higher average efficiency, the addition of extra passive components increases the system’s volume and thus reduces its power density [43]. In practice, DC-blocking capacitors are usually inserted into the AC link of the non-resonant DAB converter to prevent direct current from saturating the transformer [45]. The capacitance is selected such that the resonant frequency of the LC_DC circuit is at least ten times lower than the switching frequency (f_sw) [45]:

$${\mathrm{C}}_{\mathrm{DC} \mathrm{blocking}, \min } = {\displaystyle \frac{100}{4{\mathsf{\pi }}^2\times f_{\mathrm{sw}}^{ 2}\times \mathrm{L}}}$$

(1)

Meanwhile, for the DAB with a resonant tank operating in continuous-conduction mode, the resonant frequency is typically about 1.15 times lower than the switching frequency [46]. This makes the resonant capacitor C_RES much smaller than the DC-blocking capacitor in terms of capacitance. However, the voltage applied to the capacitor in the resonant case is much higher than in the non-resonant case and depends strongly on the operating point. Failure of this capacitor would severely affect the operation of the L_RES/C_RES tank, since higher voltage stress requires a capacitor with more demanding design constraints. Nevertheless, the resonant DAB topology is being actively investigated for high-power applications because of its excellent switching performance [47].

It should be noted that the interface converter is assumed to regulate and maintain the voltage of the DC grid (V_DC1), while the voltage V_DC2 is controlled by the connected inverter. Moreover, since other elements are also connected to the DC grid, this voltage control must be highly accurate, and large voltage variations are not acceptable under normal operating conditions. Therefore, Zero-Voltage Switching (ZVS) operation can be guaranteed even with the more limited ZVS operating range of the non-resonant DAB.

A simulation was carried out using PLECS software (version 4.9) to compare these two converters, as shown in Figure 7. For a given transferred power, under the same operating conditions and with identical parameters, the phase shift is smaller in the resonant case than in the conventional case. These results suggest that the DAB-SR may achieve higher power density. Additionally, the resonant feature helps reduce the turn-off switching losses in the converter. Nevertheless, when considering the primary voltage, an additional voltage stress can be observed across C_res, as opposed to the C_DC-blocking in the non-resonant case.

The operating point of the DAB converter significantly affects the currents flowing through the transformer (i_AC1(t) and i_AC2(t)), thereby requiring a complex modulation scheme to generate the control signals for the switches [43,48]. The most common modulation method is phase-shift modulation, in which both full-bridge circuits operate with a duty cycle of 0.5, and the transferred power and its direction are determined by the phase shift between V_AC1(t) and V_AC2(t). More advanced methods have also been proposed to minimize the transformer RMS current, losses, and required magnetizing inductance, such as triangular and trapezoidal current-mode modulation [43]. However, these schemes are more complex in terms of implementation and computational effort [49]. The DAB-SR topology offers more degrees of freedom for parameter control than the non-resonant DAB [50]. However, due to the former’s non-symmetric structure, reverse power flow (V_DC2 → V_DC1) limits the voltage gain.

3.3. Possible Solutions for the MVDC Connected DC-AC Converter

This part of the study focuses on the DC/AC conversion stage. Due to the aforementioned reasons, such as the limited voltage ratings of the semiconductor components compared with the DC bus voltage, the converter must be a multilevel topology. Several options can be considered, including the following multilevel converter topologies: the Neutral-Point-Clamped (NPC) converter shown in Figure 8a [51,52], the Flying Capacitor Converter (FCC) shown in Figure 8b [53], and their variants, namely the Stacked Multicell Converter (SMC) shown in Figure 8c [54], the Multiplexed Converter (xPlexed) shown in Figure 8d [55,56], and the Modular Multilevel Converter (MMC) shown in Figure 8e [57], which is one of the most widely used topologies for high-voltage applications.

Although the NPC converter has been extensively developed and is widely used in industry, it does not fully meet the requirements of this medium-voltage grid-connected application. In fact, the number of clamping diodes (or transistors in the case of an active neutral-point-clamped topology) becomes significant due to the high DC bus voltage, which increases inverter losses. Furthermore, the use of series-connected diodes may cause voltage imbalances during switching transients, resulting in increased electrical stress on the diodes and reduced system reliability [58,59]. The xPlexed converter may encounter a similar issue, for which a dedicated control strategy would be required [55]. This makes it one of the least attractive solutions, alongside the NPC topology [60].

A comparative study must, therefore, be conducted on the other three solutions to evaluate their applicability. A comprehensive comparison for a medium-voltage application, with an AC-side voltage of 20 kV (three-phase) and a DC bus voltage of 42 kV⎓, is presented in Figure 9 [35]. This figure compares the considered topologies, with the MMC implemented using half-bridge submodules as elementary cells [35].

The results show that all three topologies exhibit similar thermal performance in terms of power semiconductor operation. However, the MMC emerges as the most promising candidate, since it provides higher nominal power capability, lower stored energy, and the highest power density. The MMC also offers the highest degree of modularity, as it is composed of several converter cells. This characteristic further facilitates system reconfiguration in the event of a fault or when the DC bus voltage needs to be increased. The other two solutions rely on floating capacitors, which increase the design complexity and require careful management of capacitor pre-charging and voltage balancing.

While the isolated DC/DC converter requires SiC power semiconductors to enable higher switching frequencies with low switching losses, the centralized converter can use insulated-gate bipolar transistors (IGBTs) to optimize conduction losses. Since the MMC comprises several submodules (and the other considered topologies are also multilevel converters), the apparent switching frequency of the modulated signal increases even if the switching frequency applied to the transistors is low. Moreover, given the technological maturity of modern IGBTs, conduction-optimized IGBTs with blocking voltages of 3.3 kV, 4.5 kV, and 6.5 kV are widely available [61]. As a result, the total number of submodules can be reduced.

Several submodule configurations can be considered for the MMC, including the Half-Bridge (MMC-HB), the Asymmetrical H-Bridge, the Full-Bridge, and hybrid topologies. The MMC-HB, although it does not provide fault-current limiting capability, offers higher efficiency than the other solutions, since it employs fewer semiconductor devices [62]. For the sake of simplicity in the present analysis, only the MMC-HB configuration is considered in the remainder of this paper.

4. Technological Variations and Building Blocks for Power Electronics

Combining MVDC and MVAC collector architecture enables multiple converter configurations to be developed. As shown in Figure 5, RES and energy storage systems can be connected to an MVDC bus through a DC/DC converter, typically a boost converter used to increase the voltage level or perform MPPT in the case of PV systems. In this case, the MVDC bus can be set at 1.5 kV⎓ or 3 kV⎓ [14] and can be considered a DC grid. However, when integrating FC/ELYZ systems, it should be noted that the maximum output voltage of the converter is 1.5 kV⎓. If the leakage current is excessive [63], an isolated converter must be implemented for protection purposes. A DC/AC conversion stage is then used to interface the DC grid with the AC grid. This paper proposes two different approaches for such a system.

4.1. Solution with Isolated DC/DC Converters and Single-Phase Voltage Source Inverters (LVDC Collector + Cascaded H-Bridge)

Figure 10 shows the Cascaded H-Bridge (CHB) topology, in which current and power imbalances are inherently mitigated, since the power is shared equally among the phases through a common DC bus. This configuration has already been investigated in the literature, mainly for LVDC/MVAC conversion, and has recently become a popular topology for MVDC/MVAC conversion in renewable energy applications [36,64,65,66]. This topology uses the PEBB concept to create a modular system. More precisely, the number and composition of the converter blocks in each phase remain identical. In this case, a PEBB consists of an isolated DC/DC converter associated with a single-phase DC/AC converter, as shown in Figure 11. Multiple PEBBs in an IPOS configuration can then be used for the DC/AC conversion stage. Several converter solutions can be considered for this structure. Since there is an indirect connection between the two converters, each topology can be considered separately. In a simplified control scheme, the DC/DC converter controls the DC voltage of the DC grid, while the inverter controls the active and reactive power transferred to the AC grid, thus the voltage of the DC-link connected to the DC/DC converter and the current of the AC grid. The entire system must be capable of bidirectional power transmission to enable energy storage on the DC side (e.g., in batteries or ELYZ systems).

From the discussion in Section 3.2, the non-resonant DAB converter is selected because it offers several advantages and provides more stable operation. However, in this case, the output capacitor must handle low-frequency harmonic components, since it is connected to a single-phase inverter. The low-frequency current flowing through the capacitor can be calculated as a function of the modulation index m_a, grid current (with RMS value I_outrms) and frequency ω:

$${\mathrm{i}}_{\mathrm{CLF}} = {\displaystyle \frac{{\mathrm{m}}_{\mathrm{a}}{\mathrm{I}}_{\mathrm{outrms}}\sqrt{2}}{2}}\cos \left(2\mathsf{\omega }\mathrm{t}-\mathrm{\phi }\right)$$

(2)

The relation between the current and the voltage of the capacitor:

$${\mathrm{C}}_{\mathrm{LF}}{\displaystyle \frac{\mathrm{d}{\mathrm{v}}_{\mathrm{DC}2}}{\mathrm{dt}}} = {\mathrm{i}}_{\mathrm{CLF}}$$

(3)

With a given voltage ripple, the capacitor can be determined:

$${\mathrm{C}}_{\mathrm{LF}} = {\displaystyle \frac{{\mathrm{m}}_{\mathrm{a}}{\mathrm{I}}_{\mathrm{outrms}}\sqrt{2}}{2\mathsf{\omega }∆{\mathrm{V}}_{\mathrm{DC}2, \max }}}$$

(4)

A single-phase inverter can be used. However, the choice of the appropriate converter topology and modulation method can affect the system’s overall performance. As previously mentioned, a full-bridge converter is preferable, as it reduces the number of PEBBs required to provide a 20 kV three-phase output. To fully exploit the full-bridge circuit, unipolar PWM can be used to improve the output voltage harmonic performance, and two strategies can be considered. In the first approach, two switching legs operate at the switching frequency, and phase-shift, phase-disposition, or phase-opposition-disposition modulation can be employed [67]. While these methods can improve the quality of the output voltage and current, they also increase switching losses, since two switching cells operate at high frequency. Alternatively, a modified unipolar scheme can be used, in which one switching leg operates at the grid frequency and the other at the switching frequency [68]. In this case, the efficiency is higher than that of the former approach, but the harmonic performance is degraded.

As shown in Figure 11, an inverter connected to the AC grid can be implemented using IGBTs to minimize conduction losses. Since the inverter is based on a full-bridge configuration, cascading several inverters helps reduce harmonics even at low switching frequencies. However, the voltage ratings of SiC devices are still limited for high-power applications; therefore, the ratio V_DC1:V_DC2 should be 1:1 to enable full ZVS operation with an input voltage of 1.5 kV⎓. Consequently, since each inverter is connected to a single isolated DC/DC converter, the voltage ratings of the IGBTs are comparable to those of the SiC devices.

4.2. Solution with a Direct Isolated DC/AC Converter

As shown in Figure 12, the isolated DC/AC conversion system consists of an inverter connected to a matrix converter (MC), i.e., a direct AC/AC converter, through an AC link and an MFT. Two possible configurations can be considered for the design of the AC link: a resonant LCC link or a direct connection, as illustrated in Figure 13a and Figure 13b, respectively. However, the first option, which includes additional reactive components, is not favorable in this case, as it would reduce the power density. This option is also more suitable for lower-power applications and for DC/DC conversion [43]. In either case, the single-phase inverter generates a high-frequency AC output, which is then converted by the MC to the grid frequency.

The solution based on direct AC-link connection offers several advantages, since it eliminates the need for intermediate energy-storage capacitors. These capacitors can be considered environmentally burdensome, since the application shown in Figure 11 typically uses metallized polypropylene film capacitors [69,70]. Furthermore, a lower number of passive components results in lower potential losses, increased reliability, and higher power density. This topology also provides inherent protection against short circuits. Indeed, the direct AC/AC converter comprises four-quadrant switches which, in this case, are composed of two anti-series metal–oxide–semiconductor field-effect transistors (MOSFETs); with an appropriate modulation scheme, short-circuit states can be avoided [43]. One of the earliest examples of a high-power matrix converter used thyristors [71,72]. A similar topology was proposed in [73], but it uses a common medium-frequency current link instead of an MVDC collector. More recently, the conversion process has become more efficient and total harmonic distortion has been reduced thanks to the availability of fast-switching devices with improved performance. The proposed structure can be explored using WBG devices such as SiC transistors.

With a simplified control scheme, the two converters operate in an interdependent manner: the inverter controls the AC-link voltage V_AC, while the MC controls the AC-link current i_AC. Therefore, the transferred power can be controlled by adjusting the phase shift between the AC-link voltage and current by an angle Ψ, as shown in Figure 14.

The average power can be expressed by the following equation:

$$\mathrm{P} = {\mathrm{V}}_{\mathrm{DC}}{\mathrm{i}}_{\mathrm{out}}\left(1-{\displaystyle \frac{2\mathsf{\Psi }}{\mathsf{\pi }}}\right)$$

(5)

This topology enables soft switching of the semiconductor devices. Specifically, the transistors in the VSI can achieve zero turn-on loss through Zero-Voltage Switching (ZVS), while the transistors in the direct AC/AC converter can achieve zero turn-off loss through Zero-Current Switching (ZCS). However, these characteristics require auxiliary circuits (zero-voltage and zero-current detectors), as shown in Figure 15.

However, this direct topology uses an inductive voltage source at the input of the 4Q current-source inverter. This requires protection of the diodes against overvoltage during reverse recovery [74]. One possible solution is to add a voltage suppression circuit (i.e., a voltage-clamping circuit) to protect the converter [75]. Nevertheless, this introduces additional power losses, since the energy stored in the leakage inductance must be dissipated. The losses in the voltage suppression circuit can be calculated using [76]:

$$\mathrm{P} = 2f_{\mathrm{sw}}{\displaystyle \frac{2{\mathrm{I}}_{\mathrm{RRM}}^2}{1-\frac{{\mathrm{V}}_{\mathrm{DC}}}{{\mathrm{V}}_{\mathrm{z}}}}}{\mathrm{L}}_{\mathrm{lk}}$$

(6)

With I_RRM is the peak reverse-recovery current, V_z is the clamping voltage and L_lk is the transformer leakage inductance. As can be seen, this calculation depends on the leakage inductance, which is itself determined by the design parameters of the MFT. Therefore, the resulting loss calculation is not fully representative of the overall evaluation and is neglected here, since an accurate assessment would require a detailed transformer design methodology. In addition, the choice of V_z also affects these losses.

Another aspect that must be considered is the commutation of the four-quadrant switch [77], in order to avoid an open circuit at the output and a short circuit at the intermediate link. A four-step commutation strategy makes it possible to control the direction of current flow through the switch [77].

5. General Comparison

The three most promising topologies were selected for a comparative evaluation:

• 1st solution: Intermediate MVDC collector architecture with three-phase injection using an MMC-HB
• 2nd solution: Isolated DC/DC converters combined with cascaded single-phase voltage-source inverters
• 3rd solution: A direct isolated DC/AC converter

The following criteria were selected to evaluate these three solutions:

• Number of semiconductor devices;
• Semiconductor utilization factor;
• Number of reactive components and stored energy;
• Power losses.

The comparison is carried out under the same operating conditions, as shown in

Table 2. System parameters.

Nominal transferred power	4 MW
DC bus voltage	1.5 kV
AC bus voltage	20 kV
Grid frequency	50 Hz
Semiconductors’ blocking voltage	2.3 kV
Semiconductors’ switching frequency	20 kHz

.

5.1. Number of Semiconductor Devices

Table 3. Recommended DC voltage for grid-connected inverter [78].

AC Voltage Levels (kV 3~)	Recommended DC Voltage Levels (kV⎓)
6	±6
10	±10
20	±20
35	±35
110	±110

presents the recommended DC voltage levels for various AC voltage levels. In this study, for a 20 kV three-phase AC system, the recommended DC bus voltage is ±20 kV⎓ (i.e., 40 kV⎓).

For the first solution, assuming that the MVDC bus voltage is 45 kV⎓, 30 DAB converter modules are required. Each DAB converter consists of eight SiC MOSFETs. For the HB-MMC, if the capacitor voltage of each submodule is 1.5 kV⎓, then the total number of submodules in each arm is also 30.

For the second solution, 15 converter blocks are required for each phase of the PEBB-based architecture. Each block consists of eight SiC MOSFETs in the DAB stage and four IGBTs in the inverter stage. The same number of blocks and semiconductor devices can be used for the third solution.

Table 4. Number of SiC MOSFETs and IGBTs used for three cases.

	1st Solution	2nd Solution	3rd Solution
Number of SiC MOSFETs	240	360	540
Number of IGBT	360	180	–
Total Switches	600	540	540

summarizes the number of semiconductor devices used in each case.

5.2. Semiconductor Utilization Factor

This factor, together with its inverse counterpart (the semiconductor oversizing factor), indicates how effectively the semiconductor devices in the system are utilized [79]. The utilization factor is defined as follows:

$$\mathrm{U} = {\displaystyle \frac{\mathrm{P}}{\sum _{\mathrm{i} =1}^{\mathrm{N}}{\mathrm{VA}}_{\mathrm{RMS}, {\mathrm{switch}}_{\mathrm{i}}}}}$$

(7)

where VA_RMS is based on the blocking voltage and the RMS current of the switch. The calculation results show that the second solution has the highest utilization factor (approximately 0.062), while the third solution has the lowest value (approximately 0.043); the first solution has a value of approximately 0.059. This is because the third solution uses a direct AC-link configuration, which implies that the full AC output current flows through all the switches. In other cases, an isolated DC/DC stage is present, allowing a lower RMS current to flow through the switches in this stage.

5.3. Number of Reactive Components and Stored Energy

Only two types of reactive components are considered: inductors and capacitors. The grid-connected output filter is neglected. The AC-link inductor is associated with the isolated DC/DC stage, in which an additional inductance is usually connected to the transformer, since the leakage inductance may not meet the design requirements. Therefore, the number of DC/DC stages corresponds to the number of external inductors.

Regarding the capacitance, the non-resonant DAB topology is considered; therefore, only the output capacitor is taken into account. This capacitor has a smaller capacitance in the first case, since the output is connected to a three-phase inverter. In the second case, however, the output capacitor must be able to filter low-order harmonics (100 Hz), which requires a much higher capacitance value. In the first solution, which is based on the MMC, additional capacitors are present in each MMC submodule. These capacitors must have a relatively large capacitance, since they are used for energy storage. Ideally, the third solution does not require any additional inductors or capacitors in the converter.

Table 5. Number of inductors and capacitors used for three cases.

	1st Solution	2nd Solution	3rd Solution
Number of inductors	30	45	0
Number of high capacitance capacitors	180	45	0
Number of low capacitance capacitors	30	0	0

summarizes the number of reactive components required for each of the three solutions mentioned above. The third solution requires no additional components, meaning that it offers the highest power density and the best reliability from the passive-component perspective. In contrast, the first solution requires a large number of capacitors, which increases the cost and size of the system.

These reactive components also reflect the energy-storage capability of the system. For capacitors, the electric energy factor E is defined as the ratio of the total energy stored in the capacitor to the nominal power of the system [79]:

$$\mathrm{E} = {\displaystyle \frac{\sum _{\mathrm{i} =1}^{{\mathrm{N}}_{\mathrm{capa}}}\frac{1}{2}{\mathrm{C}}_{\mathrm{i}}\times {\mathrm{V}}_{\mathrm{i}}^2}{\mathrm{P}}}$$

(8)

The smaller the E factor, the better the capacitor footprint. Qualitatively, the first solution includes a much larger number of high-capacitance capacitors, making it the least favorable in terms of E. Similarly, the magnetic energy factor (k) represents the cost of the inductor. Since the size of an inductor is related to the apparent power, the factor k can be derived as follows [79]:

$$\mathrm{k} = {\displaystyle \frac{\sum _{\mathrm{i} =1}^{{\mathrm{N}}_{\mathrm{ind}}}{\mathrm{S}}_{\mathrm{ind}}}{\mathrm{P}}}$$

(9)

5.4. Semiconductor Losses

Power losses in the semiconductor devices directly reflect the performance of the power conversion system. Power losses are also present in the reactive components due to parasitic resistances, as well as in the transformers. However, the third solution introduces additional losses associated with leakage inductance, which can significantly affect the efficiency of the overall system. For the sake of simplifying the evaluation, this study considers only semiconductor losses. Future work will take into account the losses of the entire system.

To compare the semiconductor losses, a PLECS simulation was performed. Since the third solution uses only SiC MOSFETs, a direct comparison with solutions using different semiconductor technologies would not be appropriate. Therefore, all solutions were implemented using SiC MOSFETs under identical conditions. In this case, the CAB5R0A23GM4 module from Wolfspeed (Durham, NC, USA) [80] was selected. Figure 16 shows the evolution of the system’s efficiency, considering semiconductor losses only, as a function of the transferred power at unity power factor. As can be seen, the MMC-based solution exhibits higher losses, since it uses more switches than the other two solutions.

It should be noted that, at this stage, the comparison neglects all losses in the MFT and in the passive components. The core and copper losses of the MFT may become comparable to semiconductor losses. Furthermore, the parasitic resistance of DC-link capacitors can significantly affect the performance of the conversion system under high-voltage-ripple conditions.

A more detailed study will be carried out in future work to fully evaluate the efficiency and feasibility of each solution and to provide a more realistic assessment. Qualitatively, the losses in the passive components (inductors and capacitors) can be estimated from the fact that the number of passive components differs significantly among the three solutions (Table 5). As a result, the first solution would exhibit the highest losses, whereas the third solution would exhibit the lowest losses in terms of parasitic-resistance-related dissipation.

5.5. Overview

The performance summary is presented in the form of a radar chart in Figure 17. Each axis represents a specific parameter or quantity. A weight is assigned to each parameter, and by connecting the corresponding values, an area is obtained. The largest or most uniform area provides a reasonable basis for selection.

Among the solutions, the first solution, shown in green, which is based on an intermediate MVDC collector and MMC submodules on the network side, shows a fairly uniform area, except for the number of reactive elements (capacitors) and, consequently, the associated stored energy. As a result, its overall performance is lower than that of the other architectures.

The second structure, based on CHB (blue line), achieves the highest scores for four criteria. However, it includes more reactive components than the third solution, which is based on a direct DC/AC converter.

The third solution, which is based on a direct DC/AC converter (red line), achieves overall scores comparable to those of the second solution (with four criteria reaching the maximum score), while exhibiting the lowest stored energy due to its simplicity (direct DC/AC conversion without an intermediate stage). However, technological barriers still exist, and only limited research is available for high-power applications. This structure is therefore noteworthy, especially since recent advances in SiC devices, including high-blocking-voltage Junction Barrier Schottky (JBS) and Schottky diodes, may make it possible to overcome these technological difficulties.

One criterion that is particularly important but does not appear in the radar chart, because it is related to the current technological maturity of the proposed solutions, is time to market. In this respect, the isolated DC/DC + MMC approach (solution 1) offers the advantage of a shorter time to market, since the MMC submodule technology already exists and has been experimentally validated [81,82]. The second solution is an emerging approach that involves combining converters whose operation is already well documented in the literature. The third solution requires significant investment in research and development. While it offers substantial advantages in terms of component count, in-depth studies are still required to assess its behavior under blocked-switch conditions and to evaluate the contribution of SiC semiconductor devices, which are becoming increasingly widespread in the market.

Another aspect that should be considered is technological maturity. The MMC-based solution has attracted significant interest from both academia and industry for MVDC applications, largely because MMC technology has historically been used in HVDC systems [83,84]. The cascaded H-bridge architecture has also received increasing attention in this field [85,86], since it offers a modular approach while using fewer components than the MMC-based solution. Meanwhile, the direct DC/AC solution, although it has been investigated for traction applications [87], has received limited attention in grid-connected applications and therefore remains the least developed among the three proposed architectures.

6. Conclusions

The paper presents two methods for the direct injection of photovoltaic power into 20 kV three-phase medium-voltage systems, thereby eliminating the need for a 50 Hz transformer. These methods also ensure galvanic isolation between the medium-voltage grid and the energy sources.

The following electrical architectures are therefore possible:

• The protection of energy sources can be achieved using isolated DC/DC converters, which are currently the subject of active research, followed by commercially available non-isolated DC/AC converters. The advantage of this architecture is that it incorporates an intermediate MVDC link, which can be used to transport electrical energy to an MVDC energy hub. The corresponding MVDC voltage level is approximately 42 kV for power injection into a 20 kV three-phase grid, which naturally poses significant insulation challenges.
• It is possible for an LVDC bus to be shared among energy sources. The low-voltage bus is then electrically isolated from the MVAC grid via isolated DC/AC converters. In this case, since power is injected between the grid lines and the neutral point, the required dielectric strength is lower than in the former architecture. This is because phase voltages, rather than line-to-line voltages, are considered (i.e., the line-to-line voltage is divided by the square root of 3). This approach could facilitate the faster deployment of such solutions.

From a technological standpoint, several components of the architecture are difficult to assess and compare in terms of their overall value and the hidden challenges associated with their implementation.

The first issue that appears to have been resolved is the contribution of the semiconductor industry through the widespread availability of affordable power modules with voltage ratings of 1.2 kV, 1.7 kV, 2.3 kV, and 3.3 kV. The use of these high-speed switching semiconductor devices makes it possible to reduce the size of the transformers integrated into the isolated conversion stages while achieving very high conversion efficiency.

The use of these ultra-fast semiconductor devices helps address the challenge of integrating the isolation transformer at high frequency, but it also gives rise to new phenomena, such as the aging of electrical insulation under high dV/dt stress. This issue is not limited to the aforementioned elements; it also applies to other insulating materials (both solid and liquid) subjected to steep voltage waveforms. Another consequence of these high dV/dt values is the generation of disturbance currents, which can lead to measurement and signal-acquisition issues. Although this aspect is not discussed in the present article, it deserves further investigation.

Once these issues have been addressed, it is anticipated that SST technology will undergo significant growth due to the increasing power demand of applications such as data centers and electric truck charging stations, which are unlikely to be supplied at low voltage. Consequently, it is the medium-voltage power interface that will drive the need for galvanic isolation in converters, thereby ensuring compatibility with low-voltage energy sources and their insulation constraints.