Considerations for a Medium-Voltage DC Electrolysis Testbed Przemyslaw Koralewicz, Jae Hoon Choi, and Steve Hammond National Renewable Energy Laboratory NREL is a national laboratory of the U.S. Department of Energy Technical Report Office of Energy Efficiency & Renewable Energy NREL/TP-5000-94377 Operated under Contract No. DE-AC36-08GO28308 September 2025 This report is available at no cost from NREL at www.nrel.gov/publications. Considerations for a Medium-Voltage DC Electrolysis Testbed Przemyslaw Koralewicz, Jae Hoon Choi, and Steve Hammond National Renewable Energy Laboratory Suggested Citation Koralewicz, Przemyslaw, Jae Hoon Choi, and Steve Hammond. 2025. Considerations for a Medium-Voltage DC Electrolysis Testbed. Golden, CO: National Renewable Energy Laboratory. NREL/TP-5000-94377. https://www.nrel.gov/docs/fy25osti.94377.pdf. NREL is a national laboratory of the U.S. Department of Energy Technical Report Office of Energy Efficiency & Renewable Energy NREL/TP-5000-94377 Operated under Contract No. DE-AC36-08GO28308 September 2025 This report is available at no cost from 15013 Denver West Parkway NREL at www.nrel.gov/publications. Golden, CO 80401 303-275-3000 • www.nrel.gov This work was authored by NREL for the U.S. Department of Energy (DOE), operated under Contract No. DE- AC36-08GO28308. Funding provided by U.S. Department of Energy Office of Energy Efficiency and Renewable Energy Hydrogen and Fuel Cell Technologies Office. The views expressed herein do not necessarily represent the views of the DOE or the U.S. Government. U.S. Department of Energy (DOE) reports produced after 1991 and a growing number of pre-1991 documents are available free via www.OSTI.gov. Cover photos (clockwise from left): Josh Bauer, NREL 61725; Visualization from the NREL Insight Center; Getty-181828180; Agata Bogucka, NREL 91683; Dennis Schroeder, NREL 51331; Werner Slocum, NREL 67842. NREL prints on paper that contains recycled content. List of Acronyms ARIES Advanced Research on Integrated Energy Systems ARU active rectifier unit CDS collection and distribution system CGI controllable grid interface CGI2 second controllable grid interface CHIL controller hardware-in-the-loop HIL hardware-in-the-loop INU inverter unit MVDC medium-voltage DC NREL National Renewable Energy Laboratory PHIL power hardware-in-the-loop iii Executive Summary We present the results of a study focused on the feasibility of using medium-voltage direct current (MVDC) power distribution from wind power generation to electrolyzers for hydrogen production. This approach using hybrid energy generation in an MVDC microgrid offers many advantages, including possible improvements in efficiency, reliability, and installation cost compared to a more typical state-of-the-art AC distribution configuration. It also eliminates the need for transformers, which have been subject to price volatility and availability concerns. This study highlights the practical feasibility of MVDC distribution networks for integrating various energy sources, offering improved efficiency and reduced system complexity compared to conventional AC-based solutions. Future work will focus on enhancing fault protection strategies, scaling the system to larger renewable installations, and conducting hardware implementation at the National Renewable Energy Laboratory’s Flatirons Campus. This report shows that a DC collection and distribution system reduces the losses associated with the electrical conversion/distribution process relative to a state-of-the-art AC approach, improving overall efficiency by 5%. On the qualitative side, reducing the number of conversion stages is likely to improve reliability, reduce capital investment cost, enable the use of simpler control algorithms, and reduce the risk of instabilities and malfunctions. iv Table of Contents Executive Summary ................................................................................................................................... iv 1 Introduction ........................................................................................................................................... 1 1.1 Approach ....................................................................................................................................... 1 1.2 MVDC Power Distribution in Co-Located Wind and Electrolyzer Plant ..................................... 2 1.3 Why MVDC? ................................................................................................................................ 4 2 Flatirons Campus Demonstration ....................................................................................................... 7 2.1 Site Overview ................................................................................................................................ 7 2.2 Electrolyzer Characterization ........................................................................................................ 9 2.3 Planned Demonstration and Future Work ................................................................................... 10 3 MVDC Converter ................................................................................................................................. 12 3.1 Requirements Specification of the Converter.............................................................................. 12 3.2 Available Topologies and Selection Process Based on Requirements ........................................ 12 3.3 LLC Converter Selection and Design Process ............................................................................ 15 3.4 Simulink Modeling ...................................................................................................................... 18 3.5 Typhoon HIL Real-Time Model ................................................................................................. 20 3.6 System Simulation (CGI2 + DCDC + Electrolyzer) ................................................................... 23 4 Integration of Assets (Electrolyzer and Wind)................................................................................. 30 4.1 Wind Turbine Modification for DC Connection ......................................................................... 31 4.2 Co-Simulation of Wind and EZ................................................................................................... 31 4.3 Power Generation/Consumption Control Scheme....................................................................... 33 4.4 Scalability of the System (to 25 MW) ......................................................................................... 34 5 Protection ............................................................................................................................................ 37 5.1 Protection Overview .................................................................................................................... 37 5.2 Ground Fault Test........................................................................................................................ 37 5.3 Pole-to-Pole Fault Test ................................................................................................................ 44 5.4 Protection Summary .................................................................................................................... 49 6 Future Work......................................................................................................................................... 50 7 Conclusion .......................................................................................................................................... 51 References ................................................................................................................................................. 52 v List of Figures Figure 1. State-of-the-art wind to hydrogen conversion using medium-voltage AC distribution ................ 3 Figure 2. Proposed wind-to-hydrogen conversion using MVDC distribution .............................................. 3 Figure 3. Power conversion stages and typical efficiency in medium-voltage AC CDS (left) and proposed medium-voltage DC CDS (right) ............................................................................................. 4 Figure 4. An example 25-MW wind-electrolyzer DC CDS .......................................................................... 5 Figure 5. Depiction of NREL Flatirons Campus research assets. ................................................................. 7 Figure 6. Depiction of the CGI2 at Flatirons Campus .................................................................................. 8 Figure 7. CGI2 CHIL system ........................................................................................................................ 8 Figure 8. MVDC electrolyzer with DC-DC converter setup in CGI2 CHIL ................................................ 9 Figure 9. Simplified Simulink model of the electrolyzer .............................................................................. 9 Figure 10. Proposed MVDC electrolyzer to CGI2 interconnection at Flatirons Campus ........................... 10 Figure 11. Use of CGI2 for PHIL demonstration of partially simulated hybrid MVDC system ................ 11 Figure 12. Half-bridge LLC resonant converter ......................................................................................... 13 Figure 13. Series-parallel-connected LLC resonant converter ................................................................... 14 Figure 14. Voltage gain vs. normalized frequency plots for various Qe values .......................................... 16 Figure 15. Voltage gain vs. normalized frequency plots with fn_max and fn_min indicated by solid red vertical lines ........................................................................................................................................ 17 Figure 16. Simulink model of a three-phase series-parallel-connected LLC converter and the electrolyzer connected to CGI2 DC switching model via cable model ..................................................... 18 Figure 17. Input voltage and output current after filter of each converter in Simulink .............................. 19 Figure 18. Output voltage, output current and output power of the converter in Simulink ........................ 20 Figure 19. Three-phase LLC resonant converter and the electrolyzer model in Typhoon HIL .................. 20 Figure 20. CGI2 CHIL with LLC resonant converter and electrolyzer as load .......................................... 21 Figure 21. Input voltage and output current of the LLC resonant converters ............................................. 22 Figure 22. Total output power, current, and voltage of the three-phase LLC resonant converter .............. 22 Figure 23. Resonant converter input voltage and output current with input voltage balancing controller . 23 Figure 24. Electrolyzer power dropping from 100% to 20% ...................................................................... 24 Figure 25. electrolyzer input current and voltage during power ramp-down.............................................. 24 Figure 26. Current and voltage on cable during power ramp-down ........................................................... 25 Figure 27. Pole-to-ground voltage on filter during power ramp-down ....................................................... 25 Figure 28. DC output filter current and neutral-to-ground current during power ramp-down ................... 26 Figure 29. LLC resonant converter input voltage and output current during power ramp-down ............... 26 Figure 30. Electrolyzer power ramp-up from 20% to 100% ....................................................................... 27 Figure 31. Electrolyzer input current and voltage during power ramp-up .................................................. 27 Figure 32. Current and voltage measured at MVDC cable during power ramp-up .................................... 28 Figure 33. DC filter voltage during power ramp-up ................................................................................... 28 Figure 34. DC filter current and neutral-to-ground current during power ramp-up ................................... 29 Figure 35. LLC resonant converter input voltage and output current during power ramp-up .................... 29 Figure 36. Type 4 wind turbine topology ................................................................................................... 30 Figure 37. MVDC modified Type 4 wind turbine topology ....................................................................... 30 Figure 38. Co-simulation Scenario 1 setup ................................................................................................. 32 Figure 39. Co-simulation Scenario 2 setup ................................................................................................. 32 Figure 40. Grid-side converter control and wind vs. electrolyzer voltage droop control scheme .............. 33 Figure 41. Effect of wind speed change on the electrolyzer power and the grid power ............................. 34 Figure 42. Two different wind profiles and MVDC voltage level .............................................................. 35 Figure 43. Full 30-minute operational test of 25-MW wind-hydrogen system .......................................... 35 Figure 44. Fault test bed and fault location ................................................................................................. 38 Figure 45. Pole-to-ground fault current measured on DC filter in Simulink (top two charts) and in Typhoon HIL (bottom three charts) ....................................................................................... 39 vi Figure 46. Pole-to-ground DC voltage measured on DC filter in Simulink (top two charts) and in Typhoon HIL (bottom two charts) ........................................................................................................ 40 Figure 47. Electrolyzer response to pole-to-ground fault ........................................................................... 42 Figure 48. Modulation signals during the pole-to-ground fault .................................................................. 43 Figure 49. Fault messages on CGI2 CHIL control panel for pole-to-ground fault ..................................... 44 Figure 50. DC current on DC filter during pole-to-pole fault in Simulink (top) and in Typhoon HIL (bottom).................................................................................................................................. 44 Figure 51. DC voltage on DC filter during pole-to-pole fault in Simulink (top) and in Typhoon HIL (bottom).................................................................................................................................. 45 Figure 52. Electrolyzer response to pole-to-pole fault ................................................................................ 47 Figure 53. Modulation signals during the pole-to-pole fault ...................................................................... 48 Figure 54. Fault messages on CGI2 CHIL control panel for pole-to-pole fault ......................................... 49 List of Tables Table 1. Requirements for DC-to-DC Step-Down Converter ..................................................................... 12 Table 2. List of Parameters Used for LLC Resonant Converter Design ..................................................... 18 vii 1 Introduction 1.1 Approach For this effort, we first performed a detailed study to determine challenges and requirements for devices to be procured to allow for demonstration of such a distribution system at the National Renewable Energy Laboratory’s (NREL’s) Flatirons Campus. An essential decision was the selection of topology and the design of a DC/DC step-down converter, which is presented in Section 4. LLC resonant converter topology allows high- frequency isolation between input and output stages, which in turn allows for the connection of multiple stages in series at an input where voltage is above 4 kV and in parallel at an output where voltage is below 500V. This allows for the use of widely available low-voltage components in a modular design. Additionally, LLC topology is perfect for this application because the input-to-output voltage ratio is relatively stable in all practical operating points. The unidirectionality of this application also allows further simplifications to the topology. The LLC topology is commonly used in industry for various DC/DC converters requiring isolation. Once the converter was designed, various models allowing for simulation of future system with multiple levels of fidelity were developed. The MATLAB Simulink model is an offline representation of the system, including switching level models of NREL’s second controllable grid interface (CGI2) with its filters and DC/DC conversion with an electrolyzer model. It allows for offline simulation of phenomena at sub-millisecond scale necessary for protection design, as shown in Section 6. Additionally, it was used to cross-validate the quality of the Typhoon hardware-in-the-loop (HIL) model. This model executes in real time, which allows us to use the controller hardware-in-the-loop (CHIL) CGI2 replica to further increase model fidelity. The Typhoon model consists of an exact replica of the CGI2 power hardware with the proposed DC filter and DC/DC converter supplying the electrolyzer model. It is a very detailed representation of the future Flatirons Campus medium-voltage DC (MVDC) system proposed in the next stage of this project. The model allows us to demonstrate steady and dynamic operation, as shown in Section 4, and to validate the behavior of the system during fault conditions, as shown in Section 6. By performing numerous simulations using the models described above, we were able to determine the requirements for power converters and the CGI2 DC filter and distribution network that need to be procured. For example, we characterized CGI2 DC filters and cables’ extreme short-term current withstand threshold level to determine the proper sizing of these components for safe shutdown in the event of fault conditions. Moreover, the system is designed such that complex and less mature MVDC breaker technology is not necessary, as redundant fast-acting protections of the CGI2 and electrolyzer were verified to safely stop the system in case of extreme fault conditions. This document will serve as a reference for possible future design. As a last step, we propose a method of control for wind generation and electrolyzer operation that ensures power balancing during high-variability wind scenarios. The method does not require a central controller to communicate to all the assets but depends on each device’s 1 measurements of input DC voltage, which translate into power dispatch decisions. In real-world deployments, however, systems will be interconnected and central controllers will be available. Having a low-level fail-safe backup control scheme in place will allow more reliable operation of wind turbine connected electrolyzer system. The control scheme is explained in Section 5 along with a description of wind turbine system simplification to allow for MVDC interface. Additionally, the results of operating a 25-MW system consisting of two wind turbines and two electrolyzers with weak grid interface are presented to demonstrate that this control scheme is feasible for multiple generation and load devices with extreme variability. This model will be used in the future together with the CGI2 and electrolyzer to demonstrate full system performance even in the absence of an MVDC wind converter. By using power hardware-in-the- loop (PHIL), one electrolyzer in the model will be replaced with the real system at NREL’s Flatirons Campus using CGI2 as an interface between simulated and real power domain at the MVDC terminal. 1.2 MVDC Power Distribution in Co-Located Wind and Electrolyzer Plant There is growing worldwide interest in hydrogen production and use. Hydrogen is receiving much attention as a vital resource for industries such as steel, ammonia, long-haul transportation, shipping, and aviation (Gulli 2023). However, the production of hydrogen has been challenging due to high costs and electrical demand (Saur 2008; European Commission 2023). Multiple ways to reduce the cost of hydrogen generation are being studied (Eichman 2020). One approach focuses on co-locating a large wind power plant with hydrogen production and consumption or storage at remote locations where a strong grid connection may not be feasible or available (Harrison 2009). Typical large wind turbine farms would be connected to a high- voltage transmission system through medium-voltage AC collection and distribution systems (CDS), as shown in Figure 1. Typical wind turbines are designed to operate at low-voltage levels and consist of AC/DC and DC/AC power converters and transformers to match output voltage to a medium-voltage CDS. In a co-located system, the electrolyzer would be interconnected through AC CDS, thus requiring an additional matching transformer at the AC/DC conversion stage. If the location is electrically remote, installation of a large capacity of power converters may cause serious stability issues and may require additional specialized solutions, such as flexible AC transmission systems and grid- forming wind technology, which could additionally increase cost and reduce overall reliability. 2 AC DC G DC AC G AC DC DC AC ... MV AC AC DC H2 Distribution Figure 1. State-of-the-art wind to hydrogen conversion using medium-voltage AC distribution The approach we explore here is to use an MVDC CDS (also referred to more simply as DC CDS) to transfer power from the wind turbines, with the fewest power conversion stages, directly to the electrolyzer, as shown on Figure 2. A system like this will be partially decoupled from the high-voltage transmission system by using an AC/DC converter that can be rated at a power level multiple times smaller than the capacity of the wind converters. Section 5 discusses a scenario in which a 1-MW converter was deemed sufficient for a 25-MW wind system. Such a small converter is easier to integrate with a transmission grid in remote locations. As an alternative, the same system can be isolated from the grid completely by replacing the grid- facing inverter with a relatively small battery energy storage system with the same rating of 1 MW and modest storage capacity sufficient to cover dynamic deficiencies and periods of standby time with no wind generation. AC G DC AC G DC AC DC MV DC DC DC H2 Distribution Figure 2. Proposed wind-to-hydrogen conversion using MVDC distribution A crucial argument for DC CDS is the potential to reduce the losses associated with the electrical conversion/distribution process. Figure 3 shows all conversion stages of a state-of-the-art AC CDS and proposed DC CDS with typical conversion efficiencies marked. AC CDS efficiency 3 totals 88% while the proposed DC system offers 93% efficiency, an increase of 5 percentage points in overall system efficiency. While this efficiency estimation is very rough estimation by assuming typical conversion efficiencies, the efficiency improvement should allow for significant reduction in the levelized cost of hydrogen production. Further work at Flatirons Campus will focus on demonstrating this efficiency gain on real systems at the megawatt scale. Additional benefits to a DC system are real but harder to quantify. Reducing the number of conversion stages is likely to improve reliability and reduce capital investment cost. Potential reliability is higher in a system with fewer components that can break down. Also, by reducing the number of conversion stages, simpler control algorithms can be used that reduce the risk of instabilities and malfunctions. Also, a reduced number of components can lead to lower capital investment, especially in light of increased demand for and short supply of distribution transformers that depend on rare earth materials (Moseman 2024). Figure 3. Power conversion stages and typical efficiency in medium-voltage AC CDS (left) and proposed medium-voltage DC CDS (right) 1.3 Why MVDC? Typical state-of-the-art electrolyzers operate at DC voltage below 500 V, and land-based wind turbines are typically low-voltage devices with AC voltages ranging between 480 V and 690 V, so it is natural to question, “Why not use low voltage for CDS?” Figure 4 helps illustrate this problem better. It shows a system consisting of ten 2.5-MW wind turbines collecting all the power to a centrally located DC switchgear installed near the electrolyzer and AC/DC grid interconnection. Wind farms need clearance between the turbines; thus, the distance between each wind turbine and the switchgear can easily exceed thousands of feet. Transferring 2.5 MW of power at 1,000 volts direct current (VDC) would require conductors capable of carrying 2,500 4 A. Using 750-kcmil 1 aluminum photovoltaic cables, 12 parallel runs would be needed (five for each pole). Assuming current commercial cable prices, a single turbine interconnection would cost $60,000 for each 1,000 feet, which poses significant capital investment cost. An advantage of a low-voltage solution is that it would use components that are readily available, e.g., DC/DC converters, DC breakers, disconnects, fuses. This is unfortunately not the case with MVDC, which is not currently a common power distribution solution. The DC/DC converters and DC breakers for medium voltage are not readily available or in early development stages. Electrical standards for MVDC distribution are also not readily available. Figure 4. An example 25-MW wind-electrolyzer DC CDS On the other hand, by using higher DC voltage, we can reduce cable costs. For the same reason, AC CDS transformers are used for each turbine, and a medium voltage is typically selected for CDS. Selection of appropriate DC voltage level is also an interesting topic. With higher voltage, the costs of cables should be lower; on the other hand, the availability of equipment decreases exponentially. An aluminum equivalent of a typical AC CDS at 2.5 MW at 13.8 kV would use 16 kilovolts direct current (kVDC): • 3 cables rated at 105 A at 13.8 kV AC ($6,000/1,000 ft) • 2 cables rated at 156 A at 16 kVDC ($4,500/1,000 ft). 1 1 kcmil denotes 1,000 circular mils, a unit of measurement for the area of an electrical conductor with a circular cross section 5 This means that at 16 kVDC, the cost of cables would be similar to the cost of cables in an AC distribution system. However, at 16 kVDC, the cost of DC/DC converters and protection system can increase the capital investment cost to the point that the system is not affordable. The system proposed at Flatirons Campus is rated at 5 kVDC, as it is believed that at this voltage level there are commercial solutions available or prototypes nearing commercialization for DC/DC converters, DC protection, and wind turbines. NREL has worked with multiple vendors and is aware of others currently offering or developing medium-voltage DC/DC converters. Multiple DC breaker solutions, e.g., used in the traction business, are available at 3.5 kVDC. Also, there are Type 4 medium-voltage wind turbine power conversion systems readily available from some vendors. Additionally, 5 kVDC is a preferred solution for demonstration at Flatirons Campus because of the abundance of experience the Flatirons Campus team has with this voltage level from working with multiple medium-voltage drives and power converters and because of the availability of the CGI2 DC interface, which is rated at 5 kVDC. A demonstration at 5 kVDC would enable research into the fundamental concepts of MVDC distribution and the challenges with step-up and step-down conversion and protection concepts. The experience gained during that demonstration can later translate into a better understanding of MVDC distribution systems at higher voltage levels above 10 kVDC or even up to 50 kVDC. Above 50 kVDC, it may not be practical to implement the DC CDS, as these voltages are more appropriate for long-distance transmission rather than distribution systems, and associated costs rise exponentially. 6 2 Flatirons Campus Demonstration 2.1 Site Overview NREL Flatirons Campus is home to the Advanced Research on Integrated Energy Systems (ARIES) research platform and a wide variety of assets (Figure 5) that allow critical research in integrated energy systems, future energy systems, and grid integration space at a multimegawatt scale with an underground medium-voltage distribution system that mimics distribution systems at similar power scale. Among Flatirons Campus assets, the most noticeable are a 1-MW hydrogen system consisting of electrolyzer, fuel cell, storage, and corresponding power converters; a solar photovoltaic array; wind turbines; a battery energy storage system; and two CGIs. The first CGI, which was commissioned in 2013, has a 7-MW capacity at 13.2-kV output voltage level. The second CGI, commissioned in 2024, has a 20-MW capacity (Figure 6). A unique capability of CGI2 is that it supports 10-MW DC output at up to 5,000 VDC. CGI2 was commissioned to expand AC research capability and to add DC research capability at Flatirons Campus. Figure 5. Depiction of NREL Flatirons Campus research assets. Illustration by Joshua Bauer, NREL 7 Figure 6. Depiction of the CGI2 at Flatirons Campus Along with CGI2, a controller hardware-in-the-loop (CHIL) version of CGI2 (CGI2 CHIL) was also commissioned (Figure 7). CGI2 CHIL has an identical control system to CGI2. The only difference is that the power hardware, such as passive components, integrated gate-commutated thyristors, and transformers, are being simulated in a digital real-time simulator (Figure 8). CGI2 CHIL is built with three Typhoon HIL digital real-time simulators. It provides fast real-time HIL capabilities. CGI2 CHIL runs at 2-µs time steps, which can simulate all the power hardware and some controls with switching-level fidelity. Also, because the CGI2 CHIL system has an identical control system as CGI2, it replicates important details such as control interactions, protection functions, various delays within the system, and modulation schemes, further increasing the fidelity of our system. Figure 7. CGI2 CHIL system 8 Figure 8. MVDC electrolyzer with DC-DC converter setup in CGI2 CHIL 2.2 Electrolyzer Characterization NREL’s ARIES research platform currently includes a 1.25-MW hydrogen electrolyzer from Nel. This electrolyzer is research-ready and could be used as part of an MVDC distribution demonstration. It is instrumented and data are available for its characterization. The most important information from the point of view of electrical design is its voltage-current (V-I) characteristics in various operating ranges. According to the V-I data gathered from operation during a long-term variable power operation experiment, the electrolyzer is mostly linear, so it can be modeled using a voltage source and resistor shown in Figure 9. This model was used through the course of all analysis and simulation presented in this report. The few points that do not follow the linear patterns are due to the electrolyzer not operating at nominal thermal conditions during startup. These operating conditions should also be considered during design of an electrolyzer converter. Figure 9. Simplified Simulink model of the electrolyzer 9 2.3 Planned Demonstration and Future Work Figure 10 shows a satellite view of NREL Flatirons Campus research pads with the existing hydrogen system and CGI2. Proposed locations of new components are indicated, including CGI2 DC filter, MVDC distribution line, DC/DC step-down converter, and low-voltage interconnection to electrolyzer. New components need to be designed, procured, and installed. The analysis in this document serves as a reference for these designs and is crucial for determining component specifications and protection strategies. Figure 10. Proposed MVDC electrolyzer to CGI2 interconnection at Flatirons Campus Once the system is completed, a demonstration will be possible of the practical implementation of an MVDC distribution system and necessary controls to achieve reliable operation using CGI2 DC-to-DC output. We will also use the CGI2 DC output to achieve a PHIL interface to demonstrate assets that are not available at Flatirons Campus such as multiple wind turbines with greater capacity and a large-scale hydrogen system, as shown in Figure 11. 10 NREL Flations Campus HW Real-time simulation demonstartion AC DC DC Grid DC AC AC DC DC DC DC Filter DC Breaker DC H2 AC G DC AC AC G DC G DC AC G DC AC DC MV DC Distribution DC DC DC H2 DC DC DC PV Figure 11. Use of CGI2 for PHIL demonstration of partially simulated hybrid MVDC system In preparation for the abovementioned demonstration, we need to perform detailed modeling, simulations, and analysis specific to Flatirons Campus. In the following sections, we share details of the modeling and simulation process and analyze the data gathered to specify requirements and design and prepare procurements for necessary components. For this report, the abovementioned studies are done in a simulation environment using both Simulink and Typhoon HIL based on the CGI2 CHIL. 11 3 MVDC Converter 3.1 Requirements Specification of the Converter On the Flatirons Campus, the nominal DC bus voltage for an MVDC distribution network is planned at 4,300 V, and the rated input voltage for the electrolyzer varies from 320 V to 450 V, depending on the current level. As the electrolyzer degrades over its lifespan, its maximum voltage is expected to increase to 500 V. To connect the electrolyzer to an MVDC distribution network, a DC-to-DC step-down converter is required. To ensure optimal performance of the electrolyzer, the converter must meet certain requirements, as listed in Table 1. Table 1. Requirements for DC-to-DC Step-Down Converter Item Values Electrolyzer maximum power (continuous) 1.25 MW Electrolyzer maximum power (10 s) 1.5 MW Nominal input voltage 4,300 V Operating input voltage range 4,100-4,500 V Output voltage range 300-500 V Output current range 0-3,000 A Power flow Unidirectional Efficiency >95% Additionally, a wider range of DC voltage operation can be considered to optimize the converter design, e.g., 4,100 V–4,500 V, so that the output current is proportional to the input voltage, as demonstrated in Section 5. That may simplify design for certain topologies, as the input-to- output voltage ratio range can be significantly reduced. 3.2 Available Topologies and Selection Process Based on Requirements There are several DC-to-DC step-down converter topologies available, but not all of them are suitable for the medium-voltage level due to a lack of available components meeting medium- voltage specifications—such as the buck converter. The most common concept for MVDC conversion is to use multiple series-connected inverters, which at the module level are low- voltage-rated with multiple components readily available. For low-voltage outputs, these modules will have to be connected in parallel at the output. That topology requires galvanic isolation between module inputs and outputs, which excludes multiple simple DC/DC topologies. A typical approach in the industry to achieve a step-down in the DC voltage level with isolation is to use DC-to-AC followed by AC-to-DC conversion. Step-down is achieved in the AC stage by using a high-frequency or low-frequency transformer. We have decided that the resonant LLC converter topology may be the most suitable for our application because of its: 1. Narrow range of input-to-output voltage ratio 2. Unidirectional power flow 12 3. High-frequency isolation that allows series parallel operation 4. Low losses of LLC resonant converter. An LLC topology was used for the sake of the study; however, it is understood that once the procurement process starts, another topology may be found to be optimal. The results shown here can be used to specify converter requirements but do not dictate the exact topology to be used by a vendor. The resonant LLC converter topology can provide reliable and efficient voltage step-down conversion from 4.3 kV to 540~320 V using low-voltage components by using three series- connected modules at the input and paralleling them at the output. An example schematic of a half-bridge LLC resonant converter and three-phase series-parallel-connected LLC resonant converters is shown in Figure 12 and Figure 13, respectively. Figure 12. Half-bridge LLC resonant converter Figure 13 shows only three modules, but in reality, multiple parallel converters may need to be used due to the current limitation of high-frequency switching devices. For simplification of modeling, no paralleling was used. 13 Figure 13. Series-parallel-connected LLC resonant converter The main components of a resonant LLC converter are series inductors, capacitors, and shunt inductors. As shown in Figure 13, transformers provide galvanic isolation between the primary and secondary sides. This isolation allows us to flexibly configure grounding on the output side and to use a series-parallel connection for the converter. Also, the transformer’s magnetizing inductors act as a shunt inductor in the LLC circuit. There are several advantages to the resonant LLC converter. It can provide a wide range of output voltage with only minor modification to the switching frequency. Also, it can achieve zero-voltage switching during its entire operation when properly designed. As the electrolyzer’s input voltage level varies between 540 V and about 320 V, having a wide range of output voltage levels with only small modification to the switching frequency will ensure the converter’s interoperability with the electrolyzer. Additionally, since power flows in only one direction in the electrolyzer, we can further simplify the LLC resonant converter topology. 14 3.3 LLC Converter Selection and Design Process There is a significant amount of literature available to support the design of the converter. We identified the design criteria and followed the process described in the literature (Texas Instruments 2015). To start the design process, we first need to identify the maximum input minimum output voltage (𝑉𝑉𝑜𝑜𝑜𝑜𝑜𝑜_𝑚𝑚𝑚𝑚𝑚𝑚 ) for each converter. We use three converters connected in series to power the electrolyzer. Specifications shown in Table 1 are used to define the parameters. 4,500 𝑉𝑉𝑖𝑖𝑖𝑖_𝑚𝑚𝑚𝑚𝑚𝑚 = = 1,500 V 3 4,100 𝑉𝑉𝑖𝑖𝑖𝑖_𝑚𝑚𝑚𝑚𝑚𝑚 = = 1,366.67 V 3 Using these minimum and maximum voltages, we can also calculate the nominal input and 𝑉𝑉𝑖𝑖𝑖𝑖_𝑛𝑛𝑛𝑛𝑛𝑛 = = 1,433.33 V 2 𝑉𝑉𝑜𝑜𝑜𝑜𝑜𝑜_𝑛𝑛𝑛𝑛𝑛𝑛 = = 430 V 2 With the calculated nominal input and output voltages, the transformer turns ratio (𝑛𝑛) can be obtained using the equation below: 𝑉𝑉𝑖𝑖𝑖𝑖_𝑛𝑛𝑛𝑛𝑛𝑛 1,433.33 𝑛𝑛 = = = 3.3333 Once the transformer turns ratio is determined, the minimum voltage gain (𝑀𝑀𝑔𝑔_𝑚𝑚𝑚𝑚𝑚𝑚 ) and maximum voltage gain (𝑀𝑀𝑔𝑔_𝑚𝑚𝑚𝑚𝑚𝑚 ) values can be calculated using the following equation. These voltage gains are used for parameter derivation. 𝑀𝑀𝑔𝑔_𝑚𝑚𝑚𝑚𝑚𝑚 = = 1.3171 𝑉𝑉𝑖𝑖𝑖𝑖_𝑚𝑚𝑚𝑚𝑚𝑚 2 𝑀𝑀𝑔𝑔_𝑚𝑚𝑚𝑚𝑚𝑚 = = 0.7111 𝑉𝑉𝑖𝑖𝑖𝑖_𝑚𝑚𝑚𝑚𝑚𝑚 2 15 The next step in this process is the selection of the inductance ratio (𝐿𝐿𝑛𝑛 ) and the quality factor of the series resonant circuit (𝑄𝑄𝑒𝑒 ). Selecting the correct 𝐿𝐿𝑛𝑛 and 𝑄𝑄𝑒𝑒 is crucial, as they will determine the operational region of the converter. Different 𝐿𝐿𝑛𝑛 and 𝑄𝑄𝑒𝑒 values will result in narrower or wider frequency control bands. For a resonant LLC converter, it is necessary to stay within an inductive operation region, as it requires enough inductive power to achieve zero-voltage switching. A more detailed selection process of 𝐿𝐿𝑛𝑛 and 𝑄𝑄𝑒𝑒 is described by Texas Instruments (2015). For our resonant LLC converter design, 𝑄𝑄𝑒𝑒 = 0.5 and 𝐿𝐿𝑛𝑛 = 2.5 have been selected. The selected 𝑄𝑄𝑒𝑒 and 𝐿𝐿𝑛𝑛 can be verified by plotting the equation below for different values of 𝑄𝑄𝑒𝑒 (Figure 14). In this equation, 𝑓𝑓𝑛𝑛 represents the normalized switching frequency, which is normalized based on our nominal frequency (𝑓𝑓𝑜𝑜 ) of 10 kHz. In a real-life application, the nominal switching frequency would be higher than 10 kHz to allow for smaller passive components. For this report, a frequency of 10 kHz is used to ensure accurate real-time simulation. This parameter can be easily modified when we are designing the physical system. 𝐿𝐿𝑛𝑛 × 𝑓𝑓𝑛𝑛2 𝑀𝑀𝑔𝑔 = 𝑎𝑎𝑎𝑎𝑎𝑎( ) [(𝐿𝐿𝑛𝑛 + 1) × 𝑓𝑓𝑛𝑛2 − 1] + 𝑗𝑗[(𝑓𝑓𝑛𝑛2 − 1) × 𝑓𝑓𝑛𝑛 × 𝑄𝑄𝑒𝑒 × 𝐿𝐿𝑛𝑛 ] Figure 14. Voltage gain vs. normalized frequency plots for various Qe values To use the plot, we also need to determine the maximum and minimum quality factors (𝑄𝑄𝑒𝑒_𝑚𝑚𝑚𝑚𝑚𝑚, 𝑄𝑄𝑒𝑒_min ). These quality factors are calculated based on the overload factor. For our case, the overload factor is 120%, allowing for 120% overcurrent for a short period of time. Thus, 𝑄𝑄𝑒𝑒_𝑚𝑚𝑚𝑚𝑚𝑚 = 𝑄𝑄𝑒𝑒 = 0.5 Once the plot is obtained and 𝑄𝑄𝑒𝑒_𝑚𝑚𝑚𝑚𝑚𝑚 and 𝑄𝑄𝑒𝑒_𝑚𝑚𝑚𝑚𝑚𝑚 are determined, we can find the minimum and maximum operating normalized frequencies of the converter. The maximum operating normalized frequency (𝑓𝑓𝑛𝑛_𝑚𝑚𝑚𝑚𝑚𝑚 ) can be obtained by finding the point where 𝑄𝑄𝑒𝑒_𝑚𝑚𝑚𝑚𝑚𝑚 and 𝑀𝑀𝑔𝑔_𝑚𝑚𝑚𝑚𝑚𝑚 16 cross. The minimum operating normalized frequency (𝑓𝑓𝑛𝑛_𝑚𝑚𝑚𝑚𝑚𝑚 ) can be obtained by finding the Figure 15. Voltage gain vs. normalized frequency plots with fn_max and fn_min indicated by solid red vertical lines We can determine from the plot in Figure 15 that 𝑓𝑓𝑛𝑛_𝑚𝑚𝑚𝑚𝑚𝑚 = 0.735 and 𝑓𝑓𝑛𝑛_𝑚𝑚𝑚𝑚𝑚𝑚 is 1.766. These design against the plot. The next step is to calculate the effective resistance (𝑅𝑅𝑒𝑒 ) : 𝑅𝑅𝑒𝑒 = 𝜋𝜋2 ×𝐼𝐼𝐼𝐼 In this equation, 𝐼𝐼𝑜𝑜 represents the output current, which can be calculated using the nominal power output, 𝑃𝑃, and the nominal output voltage, 𝑉𝑉𝑜𝑜𝑜𝑜𝑜𝑜_𝑛𝑛𝑛𝑛𝑛𝑛 . 𝑃𝑃 1.25 MW 𝐼𝐼𝑜𝑜 = = = 969.00 A 𝑉𝑉𝑜𝑜𝑜𝑜𝑜𝑜_𝑛𝑛𝑛𝑛𝑛𝑛 430 V 𝑅𝑅𝑒𝑒 = 1.3322 ohm With 𝑅𝑅𝑒𝑒 , we can also derive the resonant capacitance (𝐶𝐶𝑟𝑟 ), resonant inductance (𝐿𝐿𝑟𝑟 ), and magnetizing inductance (𝐿𝐿𝑚𝑚 ) using the following equations: 1 𝐶𝐶𝑟𝑟 = = 2.3893 × 10−5 F 1 𝐿𝐿𝑟𝑟 = (2𝜋𝜋𝑓𝑓 )2 = 1.0601 × 10−5 H 𝑜𝑜 ∗𝐶𝐶𝑟𝑟 17 𝐿𝐿𝑚𝑚 = 𝐿𝐿𝑟𝑟 × 𝐿𝐿𝑛𝑛 = 2.6504 × 10−5 H Table 2 shows a summary of the derived parameters. Table 2. List of Parameters Used for LLC Resonant Converter Design Item Symbol Values Maximum input voltage 𝑉𝑉𝑖𝑖𝑖𝑖_𝑚𝑚𝑚𝑚𝑚𝑚 1,500 V Minimum input voltage 𝑉𝑉𝑖𝑖𝑖𝑖_𝑚𝑚𝑚𝑚𝑚𝑚 1,366.67 V Nominal input voltage 𝑉𝑉𝑖𝑖𝑖𝑖_𝑛𝑛𝑛𝑛𝑛𝑛 1,433.33V Maximum output voltage 𝑉𝑉𝑜𝑜𝑜𝑜𝑜𝑜_𝑚𝑚𝑚𝑚𝑚𝑚 540 V Minimum output voltage 𝑉𝑉𝑜𝑜𝑜𝑜𝑜𝑜_𝑚𝑚𝑚𝑚𝑚𝑚 320 V Nominal output voltage 𝑉𝑉𝑜𝑜𝑜𝑜𝑜𝑜_𝑛𝑛𝑛𝑛𝑛𝑛 430 V Turns ratio 𝑛𝑛 3.33 Maximum voltage gain 𝑀𝑀𝑔𝑔_𝑚𝑚𝑚𝑚𝑚𝑚 1.3171 Minimum voltage gain 𝑀𝑀𝑔𝑔_𝑚𝑚𝑚𝑚𝑚𝑚 0.7111 Inductance ratio 𝐿𝐿𝑛𝑛 2.5 Quality factor 𝑄𝑄𝑒𝑒 0.5 Nominal frequency 𝑓𝑓𝑜𝑜 10 kHz Effective resistance 𝑅𝑅𝑒𝑒 1.3322 ohm Resonant capacitance 𝐶𝐶𝑟𝑟 2.3893 × 10−5 F Resonant inductance 𝐿𝐿𝑟𝑟 1.0601 × 10−5 H Magnetizing inductance 𝐿𝐿𝑚𝑚 2.6504 × 10−5 H 3.4 Simulink Modeling Using these parameters, we built a three-phase series-parallel connected LLC converter in the Simulink simulation environment to confirm its operation (Figure 16). Along with the converter, using the voltage vs. current characteristics of the electrolyzer, we also built a linearized simulation model of the electrolyzer. Figure 16. Simulink model of a three-phase series-parallel-connected LLC converter and the electrolyzer connected to CGI2 DC switching model via cable model 18 For the initial verification process of the converter operation and its interaction with the electrolyzer, the CGI2 Simulink DC switching model was used to provide 4,300 VDC to the system. Additionally, different frequency inputs were provided to each converter. This was done intentionally to put the converters out of balance to see their reaction. Then, these results were used to cross-validate the performance of the Simulink model and the Typhoon HIL real-time model. The simulation results shown in Figure 17 illustrate the input voltage and output current after the filters of each converter. Each converter is provided with different switching frequencies: 10,000 Hz, 10,200 Hz and 9,500 Hz. As each converter has different frequency inputs, the output voltage and current levels are different. The second converter with 10,200 Hz frequency input has the highest input voltage and output current while the third converter with 9,500 Hz frequency input has the lowest input voltage and output current. Additionally, examining the output voltage and output current ripple (Figure 18), the amplitude of voltage ripple is about 1 V (0.2%), and the amplitude of current ripple is about 2.5 A (0.007%), which are very small. Figure 17. Input voltage and output current after filter of each converter in Simulink 19 Figure 18. Output voltage, output current and output power of the converter in Simulink 3.5 Typhoon HIL Real-Time Model After confirming the correct operation of the converter in the Simulink environment, a real-time simulation model of the resonant LLC converter was built (Figure 19). For the real-time simulation platform, Typhoon HIL was used. Then, the converter and the electrolyzer were connected to the CGI2 CHIL DC output to evaluate its interaction with CGI2 CHIL (Figure 20). Figure 19. Three-phase LLC resonant converter and the electrolyzer model in Typhoon HIL 20 Figure 20. CGI2 CHIL with LLC resonant converter and electrolyzer as load Similar to the Simulink-based model, the electrolyzer is modeled using the voltage vs. current characteristic, and the converter is configured in a series-parallel connection. As mentioned above, the converters were intentionally put out of balance. The converters were provided with the same frequency set point as the Simulink model. The results were then cross-validated against those from the Simulink simulation. Comparing the results, it is shown in Figure 21 that the input voltage and output current distributions are similar to those of the Simulink simulation results. Figure 22 also shows that the voltage ripple was about 2 V, and the current ripple was about 20 A. The current is relatively higher than the current from the Simulink simulation, but it was still small enough. The difference between the Simulink and Typhoon HIL model results could be caused by the complexity of the Typhoon HIL model. In the Simulink model, a simplified DC switching model was used. However, the Typhoon HIL model uses CGI2 CHIL, which contains more passive elements, and the converters switch at 2-µs time steps. Even though the current ripple was slightly higher, the converter was able to provide sufficient power while stepping the voltage down to within the operational range of the electrolyzer. In real-life applications, CGI2 CHIL represents a DC grid bus, meaning this combination of CGI2 CHIL and the electrolyzer effectively simulates an electrolyzer connected to a DC grid. 21 Figure 21. Input voltage and output current of the LLC resonant converters Figure 22. Total output power, current, and voltage of the three-phase LLC resonant converter After cross-validating both models’ performance, an input voltage balancing controller was added. The voltage balancing controller will take input voltage level at each converter as an input and adjust the switching frequency of each converter to evenly distribute the input voltage among three converters. A plot of input voltage level at each converter after adding the controller is shown in Figure 23. 22 Figure 23. Resonant converter input voltage and output current with input voltage balancing controller 3.6 System Simulation (CGI2 + DCDC + Electrolyzer) The wind-to-hydrogen system will not run at steady state only. It will also face some transient events during its operation. To investigate transient effects on an MVDC distribution network and the electrolyzer, some dynamic operations were performed by changing the resonant converter’s current reference manually while keeping the CGI2 CHIL DC output voltage steady at 4,300 V. The first dynamic case was the electrolyzer power ramp-down. The power was ramped down from 100% (1.25 MW) to 20% (250 kW) within 50 ms. Figures 24–29 show various waveforms collected during this transient event at various points of the circuit such as the MVDC cable, the electrolyzer, the converter, and the DC filter. During this transient event, the power consumption of the electrolyzer decreased from 1.25 MW to 250 kW in 20 ms (Figure 24). As a result, the voltage at the input of the resonant converter increased by 20 V while the current at the input decreased to nearly 10 A (Figure 25). As shown in Figure 26 and Figure 27 the voltage overshoot on the cable and the filter during this transient was very small. The currents measured at DC output filter and neutral-to-ground did not increase as there was no ground fault associated with this transient event (Figure 28). Lastly, the input voltage at each converter had some overshoot as well, but the overshoot was only 20 V, which is small enough for the converter to handle safely (Figure 29). 23 Figure 24. Electrolyzer power dropping from 100% to 20% Figure 25. electrolyzer input current and voltage during power ramp-down 24 Figure 26. Current and voltage on cable during power ramp-down Figure 27. Pole-to-ground voltage on filter during power ramp-down 25 Figure 28. DC output filter current and neutral-to-ground current during power ramp-down Figure 29. LLC resonant converter input voltage and output current during power ramp-down The second transient operation was power ramp-up from 20% to 100% within 50 ms, increasing the active power, voltage and current of the electrolyzer rapidly (Figure 30 and Figure 31). The responses of the system during power ramp-up transient operation was very similar to those during power ramp-down. During the transient event, the MVDC cable saw some voltage undershoot with an amplitude of approximately 30 V, and the filter voltage was similarly affected (Figure 32 and Figure 33). There was no visible filter current and neutral-to-ground current change (Figure 34). Lastly, the LLC resonant converter input voltage had voltage undershoot of 15 V, which is very small (Figure 35). 26 Figure 30. Electrolyzer power ramp-up from 20% to 100% Figure 31. Electrolyzer input current and voltage during power ramp-up 27 Figure 32. Current and voltage measured at MVDC cable during power ramp-up Figure 33. DC filter voltage during power ramp-up 28 Figure 34. DC filter current and neutral-to-ground current during power ramp-up Figure 35. LLC resonant converter input voltage and output current during power ramp-up These two dynamic operation cases confirm that the electrolyzer coupled with CGI2 CHIL is safe to operate even during transient events. However, only basic transient events were studied. Therefore, more complex and extreme transient operations should be investigated in the future. 29 4 Integration of Assets (Electrolyzer and Wind) The goal of this research is to simulate electrolytic hydrogen production with a wind turbine. To achieve this, it is necessary to model a wind turbine as well. For this purpose, we based our model on Type 4 wind turbines. A typical Type4 wind turbine system is illustrated in Figure 36. Figure 36. Type 4 wind turbine topology Currently available Type 4 wind turbine models undergo AC-to-DC and DC-to-AC conversion stages to connect to an AC distribution system. In addition to these conversion stages, a transformer is also required. Our objective is to eliminate the DC-to-AC conversion stage and the transformer, allowing the wind turbine and electrolyzer to connect directly via a DC distribution system. As discussed in the previous section, the wind turbine and the electrolyzer will be installed in close proximity, and the grid will be positioned at a greater distance. Although grid connection will still require DC-to-AC conversion, the direct connection between the turbine and the electrolyzer can be achieved using an MVDC distribution system. By removing the conversion stage and the transformer, we can improve both power and economic efficiency. Reducing the number of components not only decreases capital investment for similar systems but also minimizes power losses associated with conversion and transformation processes. The proposed modification for a Type 4 wind turbine is shown in Figure 37. Figure 37. MVDC modified Type 4 wind turbine topology Due to computational limitations in the Typhoon HIL platform, the wind turbine is simulated in the RTDS real-time digital simulation environment. A working 2.5-MW Type 4 wind turbine model is available in RTDS, which includes key mechanical characteristics such as pitch angle control and torque control. This model also accepts wind speed profiles to accurately represent real-world turbine behavior. We modified the 2.5-MW wind turbine model to create a version compatible with a DC grid connection. 30 4.1 Wind Turbine Modification for DC Connection To achieve MVDC connection of the wind turbine, there are some modifications that must be made to the wind turbine model. First, the DC-to-AC conversion stage was removed. The wind turbine will only have a DC link capacitor before being connected to an MVDC grid. Additionally, a scaling of DC link voltage was necessary. The wind turbine model’s rated DC link voltage was lower than the intended DC voltage level of an MVDC distribution system. It was rated at 1.2 kV while the MVDC distribution system was rated at 4.3 kV. However, if the DC voltage level changes, other components such as inductors and capacitors should also be adjusted. For the purpose of demonstration and to reduce the complexity, instead of changing all the parameters, the scaling of the DC link voltage was done within CHIL. 4.2 Co-Simulation of Wind and EZ Since the wind turbine and the electrolyzer, along with LLC resonant converters, are simulated in different real-time environments, a co-simulation approach is required to seamlessly integrate both systems. To achieve this, analog input and output interfaces available on RTDS and Typhoon-HIL were utilized. When performing a co-simulation, selecting a common node is crucial for accurate interaction between two systems. Given that the turbine and electrolyzer are interconnected via an MVDC bus, the DC bus serves as the common node in our simulation setup. In this section, we discuss two scenarios. The first scenario assumes a weak grid without any grid events (Figure 38). For the first scenario, the grid, wind turbines and a virtual electrolyzer are modeled in RTDS while another electrolyzer is modeled in Typhoon HIL. This scenario represents possible PHIL interface with real hardware. The electrolyzer simulated in Typhoon- HIL can later be replaced with the electrolyzer at NREL Flatirons Campus to study electrolyzer’s interaction with simulated grid systems. The second scenario is specific to the NREL Flatirons Campus, where the wind turbine and the electrolyzer are connected to CGI2 (Figure 39). In the second scenario, CGI2 represents a strong grid, as CGI2’s capacity of 20MW is higher than the other assets’ capacity. This scenario is to study electrolyzer’s interaction with CGI2. CGI2 and the electrolyzer modeled in Typhoon-HIL will be replaced by actual CGI2 and the electrolyzer at Flatirons Campus, once they are fully commissioned. For Scenario 2, the grid (CGI2) and the electrolyzer are modeled in Typhoon-HIL while the wind turbines and a virtual electrolyzer are modeled in RTDS. 31 Figure 38. Co-simulation Scenario 1 setup Figure 39. Co-simulation Scenario 2 setup 32 4.3 Power Generation/Consumption Control Scheme There is growing interest in hydrogen production in areas that are either underserved by the utility grid or in remote areas where there is no grid. To produce hydrogen from hybrid energy systems rather than the grid, a control scheme must be implemented to regulate the electrolyzer’s power consumption, ensuring that it operates in harmony with the hybrid energy system power generation. To achieve this, we developed a communication-less power consumption control scheme using an MVDC voltage level. In this control scheme, the MVDC voltage level is monitored by each device in the system to determine the power generation level of the wind turbine, and the output current is adjusted accordingly. The electrolyzer will decide its power consumption level based on DC voltage at its terminal. Once the electrolyzer’s DC voltage level reaches its maximum, signifying that it is consuming its maximum power, a wind turbine will start to curtail its power production by controlling its pitch angle. For this case, the grid converter is only responsible for DC voltage regulation and auxiliary power to turbines and the electrolyzer. DC voltage regulation and auxiliary power will only require a small amount of power. A small capacity converter that is less than 10% of the total electrolyzer and the wind capacity will be enough to handle the operation, which also decreases the cost of installation. This grid converter will act as a virtual capacitor to allow drift on DC voltage. This entire process is accomplished by only looking at the DC voltage level at each device in the system. Figure 40 illustrates such control. Figure 40. Grid-side converter control and wind vs. electrolyzer voltage droop control scheme When the wind power generation is at 1.25 MW or above—the maximum capacity the electrolyzer can consume—the current controller commands the electrolyzer to draw its full capacity, supplying 3,000 A at 415 V to consume 1.25 MW. For the wind turbine, the control will curtail the power generation to be at 1.25 MW, even if there is more wind available to generate more power, minimizing the grid involvement. However, when power generation falls below 1.25 MW, the electrolyzer’s power consumption is curtailed by limiting the output current from the converter. The output current level is determined from the pre-characterized curves based on DC voltage level, which is shown in Figure 40. Using this control scheme, electrolytic hydrogen production directly from hybrid energy systems can be achieved without requiring a communication protocol between systems. Initial simulation results after implementing this control approach are shown in Figure 41. 33 Figure 41. Effect of wind speed change on the electrolyzer power and the grid power The wind turbine produces maximum power at a wind speed of 12 m/s. As shown in Figure 41, as the wind speed reaches 12 m/s, the wind turbine starts to generate maximum power, and the electrolyzer’s power consumption level closely follows the power generation of the wind turbine. As soon as the wind speed exceeds 12 m/s, the turbine started to adjust the pitch angle of the blades to curtail the power production. Therefore, the grid’s involvement during this operation is minimized. All these interactions are based on the MVDC voltage level, which is shown in blue in the first plot. It is visible from the plot that the MVDC voltage level closely follows the wind speed variability. 4.4 Scalability of the System (to 25 MW) We believe that a single 1.25-MW capacity electrolyzer and a single 2.5-MW wind turbine are insufficient to fully demonstrate real-world operation. Typically, wind turbines are deployed in a cluster with total capacities in the tens of megawatts. A single electrolyzer would not be enough to utilize all the power generated from such a setup. To fully harness wind-generated power, multiple electrolyzer stacks are needed to scale up hydrogen production. Although the electrolyzer at Flatirons Campus is only 1.25 MW, and the resonant LLC converter we designed is also rated for 1.25 MW, both can be easily scaled by connecting multiple units in series. To demonstrate this scalability, we expanded our testbed by incorporating multiple wind turbines and electrolyzers, increasing the system’s overall capacity. First, we increased the capacity of a single wind turbine to 12.5 MW and added a second identical turbine in RTDS. Similarly, we scaled the electrolyzer capacity to 12.5 MW and added a second identical electrolyzer. To enhance the realism of our simulation, we also introduced a 30-minute wind profile captured at Flatirons Campus. 34 Since wind turbines in real-world conditions do not experience identical wind speeds, we applied two different wind profiles. To achieve this, we took the original wind profile and introduced time delays to account for the spatial separation between turbines. Additionally, we incorporated a 1-Hz sinusoidal wave with an amplitude of 1 meter per second to represent small variations in wind speed. The resulting two wind profiles used in the simulation are shown below. Additionally, an MVDC voltage level is plotted in the figure. This plot illustrates the change of DC voltage level depending on the wind profile due to voltage droop control that was implemented (Figure 42). Figure 42. Two different wind profiles and MVDC voltage level Using these wind profiles, we conducted a full 30-minute operational test of the expanded testbed. The results are presented in Figure 43. Figure 43. Full 30-minute operational test of 25-MW wind-hydrogen system 35 Analyzing the total wind power generation and the total electrolyzer consumption, we observe that, due to the implemented control scheme, the electrolyzers consistently consume less power than what is generated by the wind turbines. Consequently, there is minimal reliance on grid power, successfully demonstrating the production of electrolytic hydrogen from hybrid energy systems. This validation confirms that our proposed approach—leveraging an MVDC distribution network for wind-to-hydrogen conversion and implementing a communication-less control scheme—is a viable alternative to traditional AC distribution networks for large-scale hydrogen production. 36 5 Protection 5.1 Protection Overview Although we demonstrated that the proposed concept is a probable solution to produce electrolytic hydrogen from hybrid energy systems, protection is still a concern for the DC system. There are many readily available protection equipment options for AC systems, and it is relatively easy to implement protection due to the nature of AC systems—AC crosses zero every 1/60 s. AC protection equipment, such as a breaker, can latch at zero current when there is a fault, minimizing the energy dissipated. Unlike AC, DC will not reach zero during operation, which means that when a fault happens, a lot of energy will be dissipated. Also, our test articles, such as CGI2 and the resonant LLC converter, contain large capacitors that store a lot of energy. Thus, it is important to investigate the possible impact of faults on the DC system. For the fault cases, we mainly focused on two major fault types: pole-to-ground fault and pole- to-pole fault. This study is specific to NREL’s Flatirons Campus and investigates the impact of the fault on CGI2, the MVDC distribution system, and the electrolyzer, including the resonant converter. Because faults are extremely dynamic events, we first had to confirm the integrity of our real- time simulation model at dynamic levels. To do so, we ran a Simulink-based model for the aforementioned fault cases and cross-validated the results with real-time simulations. The initial responses from CGI2 CHIL were nearly identical to those of the Simulink model. This indicates the reliability of our CGI2 CHIL model. Next, we compared and analyzed the results of the two fault cases from both Simulink and real-time simulations, as discussed below. For all results presented here, the initial responses from real-time simulations are similar but do not exactly match the responses from the Simulink simulation. This is because the CGI2 CHIL model includes the CGI2 protection system, whereas the Simulink model does not. When a fault occurs, the CGI2 controller detects it and activates the protection system, such as integrated gate- commutated thyristor pulse removal. This protection system disconnects CGI2 from the fault, protecting the internal circuit from any damage. Using CGI2 CHIL allows us to see exactly how the fault detection and protection actions will be executed by the CGI2 controller. 5.2 Ground Fault Test For the pole-to-ground fault test, we injected a fault on the load side before the resonant converter (Figure 44). The fault was applied between the positive pole and the ground for 0.005 seconds. We applied the fault after the converter reached steady-state operation at full power. After applying the fault, we examined measurements from various locations on the circuit. We also conducted an identical test on the real-time simulation model using CGI2 CHIL. The fault location and fault duration were as described above, and measurements were taken from identical locations in the circuit. 37 Figure 44. Fault test bed and fault location As shown in Figure 45, as soon as the fault occurred, we observed approximately 7,000 A of fault current flowing through the grounding cable, which then distributed into 4,000 A flowing from the positive pole to ground at the DC filter and around 3,000 A flowing from the negative pole to ground at the DC filter. This is the anticipated current flow path, as the cable and filters are grounded. 38 Figure 45. Pole-to-ground fault current measured on DC filter in Simulink (top two charts) and in Typhoon HIL (bottom three charts) The voltage level on each pole also began to change after the fault. Normally, since the nominal voltage of our MVDC distribution network is 4,300 V, the voltage level on capacitors at each pole would be 2,150 V. However, after the fault, the voltage on the positive pole capacitor rapidly discharged to zero, while the voltage on the negative pole capacitor rose to 3,500 V for a short period before gradually discharging (Figure 46) These results indicate that we need to select capacitors that can withstand short-term overvoltage, undervoltage, and overcurrent when designing the DC filter. 39 Figure 46. Pole-to-ground DC voltage measured on DC filter in Simulink (top two charts) and in Typhoon HIL (bottom two charts) We also examined the resonant converter and the electrolyzer during the fault. The output voltage level decreased as the input voltage level failed to maintain 4,300 V. Initially, while there was still some voltage at the input, each converter attempted to maintain 1,000 A to provide a total of 3,000 A. However, as the input voltage dropped to an insufficient level to sustain converter operation, the converters stopped operating. During the simulation, the converters ceased operation approximately 0.02 s after the fault (Figure 47). Additionally, the fault current observed on the high-voltage side did not affect the converter or the electrolyzer, as the primary and secondary sides of the converter are galvanically isolated by the transformer. The only necessary protection to implement is a control scheme that stops the modulation of the converter when a fault is detected. 40 (a) Input voltage of the resonant converter during pole-to-ground fault (b) Output current of the converter during pole-to-ground fault 41 (c) Output power, output voltage and output current of the converter during the pole-to-ground fault Figure 47. Electrolyzer response to pole-to-ground fault Lastly, we examined CGI2’s reaction to the fault. Thanks to the 2-µs time step, we were able to observe the modulation signals of CGI2. CGI2 has active rectifier units (ARUs) and inverter units (INUs) that allow us to generate the desired voltage level. By analyzing the modulation signals of both ARUs and INUs, we observed that modulation stopped approximately 0.0035 s after the fault (Figure 48). This was due to fault detection by the CGI2 controller. As a result of the fault, the pole-to-ground voltage level on the filter exceeded the protection system’s threshold, triggering the ARUs and INUs to trip and isolate CGI2 from the fault. Figure 49 shows the fault messages on CGI2 CHIL control panel during this fault. 42 (a) Active rectifier unit modulation signals during the pole-to-ground fault (b) Inverter unit modulation signals during the pole-to-ground fault Figure 48. Modulation signals during the pole-to-ground fault 43 Figure 49. Fault messages on CGI2 CHIL control panel for pole-to-ground fault 5.3 Pole-to-Pole Fault Test The location of the pole-to-pole fault was the same as in the pole-to-ground fault case. The pole- to-pole fault was applied between the positive and negative poles for 0.005 s. As in the previous case, the fault was applied after the converter reached steady-state operation at full power. Analyzing the simulation results, we observed that the fault caused approximately 5,000 A of current on both the negative and positive poles of the cable. The fault current then flowed into the filter. Since this was a pole-to-pole fault, the fault current was symmetrically distributed between both poles (Figure 50). Figure 50. DC current on DC filter during pole-to-pole fault in Simulink (top) and in Typhoon HIL (bottom) At the same time, the DC filter output voltage on both poles dropped to zero approximately 3.4 ms after the fault. Since the voltage level dropped to zero, as in the previous case, the resonant converters were unable to continue operation and shut down automatically. Figure 51 shows that as soon as the fault occurred, the resonant converter input voltage exhibited a rapid fluctuation. As a result, the output current of each converter reached zero approximately 3 ms after the fault, delivering no active power to the electrolyzer (Figure 52). 44 Figure 51. DC voltage on DC filter during pole-to-pole fault in Simulink (top) and in Typhoon HIL (bottom) 45 (a) Input voltage measured on each converter during pole-to-pole fault (b) Output current measured on each converter during pole-to-pole fault 46 (c) Output power, output voltage and output current of the converter measured during pole-to- pole fault Figure 52. Electrolyzer response to pole-to-pole fault Finally, examining the modulation signals of CGI2’s ARUs and INUs, we found that the ARUs stopped modulating approximately 0.06 s after the fault (Figure 53), while the INUs stopped 0.0028 s after the fault (Figure 54). Looking at the current measurements on the INUs, we detected a large fault current of about 1,600 A—sufficient to trigger the overcurrent protection on the CGI2 controller. According to the fault message from the CGI2 controller, the ARUs faulted due to an issue in the INUs, and the INUs faulted due to overcurrent detection, which aligns with the measurements taken at the INUs. 47 (a) ARU modulation signals (b) INU modulation signals Figure 53. Modulation signals during the pole-to-pole fault 48 Figure 54. Fault messages on CGI2 CHIL control panel for pole-to-pole fault 5.4 Protection Summary Analyzing the results from both offline and real-time simulations, we did not observe any abnormal behavior in terms of current, voltage, or converter and inverter performance that could cause severe damage to connected assets. The results demonstrated that our assets, CGI2 and the resonant converters, can shut themselves down within 2–3 ms without exceeding realistic design criteria, ensuring safe operation for both personnel and equipment. Although the operation is sufficiently safe, the resonant converter was exposed to short-term overvoltage during the pole-to-ground fault. Therefore, we must ensure that the converter components have adequate overvoltage ratings. Additionally, it would be beneficial to implement undervoltage and overvoltage detection mechanisms on the converter’s input voltage side. During a fault, the converter does not need to remain operational, so it is ideal to shut it down immediately upon detecting undervoltage or overvoltage to isolate both the converter and the electrolyzer from the fault. Beyond the converter, the DC filter is also exposed to stresses during faults. We observed that for both fault cases, high fault currents between 4,000 A and 6,000 A flowed through the filter, particularly through its capacitors. Using this information, we can select appropriate DC filter capacitors that can withstand the observed fault currents as well as capacitor fuses to protect the capacitors in the event of a more severe fault with even higher current levels. Assuming the above considerations and recommended protections are implemented, we do not identify a critical need for DC breakers or high-current DC fuses in this MVDC distribution network configuration connecting CGI2 and the electrolyzer at the NREL Flatirons Campus. However, other use cases involving additional assets, such as a battery storage system, may require different protection devices. 49 6 Future Work While the analysis in this report demonstrated that the proposed MVDC distribution system is a probable solution for integrating the wind turbine and the electrolyzer, there are still areas that require further research. Particularly, optimization of protection strategies and evaluation of their application to different assets and configuration will be necessary for broader implementation. For instance, this report only investigated basic fault cases. Although they are the most common fault cases, it is necessary to investigate more complex fault cases such as high impedance fault to properly design the protection system. Additionally, our test bed is designed around Type 4 wind turbines, which are primarily used for offshore applications. To expand the applicability of our proposed solution, it will be important to incorporate other types of wind turbines. For example, Type 3 wind turbines, also known as doubly fed induction generator wind turbines, are more commonly used for land-based installations. Integrating Type 3 wind turbines will require further development to account for their different topology. Finally, validating the proposed solution with physical systems at NREL’s Flatirons Campus will be essential. Once all necessary assets are commissioned, confirming that the observed simulation behaviors align with those seen in the physical system’s operation will be critical for ensuring the system’s reliability and effectiveness. 50 7 Conclusion This report demonstrated the feasibility of an MVDC distribution network to connect hybrid energy sources, such as wind, directly to electrolyzers for electrolytic hydrogen production. By implementing a resonant LLC converter, we successfully achieved efficient DC voltage step- down while ensuring reliable power distribution to the electrolyzer. The proposed control scheme that does not depend on communication effectively regulated the power consumption of the electrolyzer based on available power from the hybrid energy systems, minimizing grid dependency; thus, it will ensure reliable hydrogen production. By utilizing both Simulink models and the CGI2 CHIL real-time model, we verified the system’s performance under various circumstances, including two critical fault scenarios. The results confirm that the converter and CGI2 can operate safely without external protection equipment, if correct components that withstand the stresses are used. While the proposed solution shows significant promise, further research is needed to develop better protection strategies that would cover more complex fault scenarios and expand compatibility with different wind turbine topologies. Future work will also include validation of the performance in a physical test bed at NREL’s Flatirons Campus. By using the proposed solution combined with effective control strategies, this report presented a scalable, more efficient, and more economical solution for hybrid energy systems providing power for electrolytic hydrogen production. 51 References Eichman, Josh, Mariya Koleva, Omar J. Guerra, and Brady McLaughlin. 2020. Optimizing an Integrated Renewable-Electrolysis System. Golden, CO: National Renewable Energy Laboratory. NREL/TP-5400-75635. https://www.nrel.gov/docs/fy20osti/75635.pdf. European Commission: Directorate-General for Energy, ETIP SNET, Iliceto, A., Samovich, N., Souza e Silva, N., & Ilo, A. 2023. Hydrogen’s impact on grids : impact of hydrogen integration on power grids and energy systems, Publications Office of the European Union. https://data.europa.eu/doi/10.2833/556144. Gulli, C., Heid, B., Waardenburg, M., & Noffsinger, J. 2023. Global energy perspective 2023: Hydrogen outlook. McKinsey & Company. https://www.mckinsey.com/industries/oil-and- gas/our-insights/global-energy-perspective-2023-hydrogen-outlook. Harrison, K. 2009. Wind-To-Hydrogen Project: Operational Experience, Performance Testing, and Systems Integration. https://doi.org/10.2172/951802. Moseman, Andrew. 2024. “This Essential Element of the Power Grid Is in Critically Short Supply.” IEEE Spectrum. https://spectrum.ieee.org/transformer-shortage. Saur, G. 2008. Wind-To-Hydrogen Project: Electrolyzer Capital Cost Study. https://doi.org/10.2172/944892. Texas Instruments. 2015. “Power Tips: Designing an LLC Resonant Half-Bridge Power Converter.” https://www.ti.com/lit/ta/ssztch0/ssztch0.pdf?ts=1745544446495. 52