Fuel cell system architecture for artificial intelligence model training

Abstract

To provide a fuel cell system architecture for artificial intelligence model training.SOLUTION: The present disclosure is directed to a system that employs fuel cell-based power generation for decentralized data centers that perform large, processing intensive tasks, such as training processes for large artificial intelligence models. The system utilizes various modules, such as energy storage systems, load banks, and other types of loads, to supplement power output by fuel cells, to store any excess power generated by fuel cell systems. As a result, swings in the power output by the fuel cell systems are minimized and the life of the fuel cell systems may be extended.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] background Technical Field FIELD OF THE DISCLOSURE

[0001] The present disclosure is directed to a system architecture and power control method for a fuel cell system. [Background technology]

[0002] 2. Description of Related Art

[0002] Fuel cells, such as solid oxide fuel cells (SOFCs), are energy conversion devices that can generate electricity and heat directly from hydrogen and hydrocarbon gases combined with an oxidant. Fuel cell systems can generate electrical power continuously without interruption. Fuel cell systems are highly efficient compared to traditional power generation devices. For example, in a diesel generator, diesel fuel and compressed air are ignited, converting the chemical energy of the fuel into thermal energy, which is then converted into mechanical energy (e.g., using the heat to drive a turbine), which is finally converted into electrical energy. In contrast, fuel cell systems avoid the conversion of mechanical energy to electrical energy. Instead, fuel cells generate electricity and heat through electrochemical reactions, contributing to clean baseload power and serving as a backup solution.

[0003]

[0003] Due to these advantages, fuel cell systems are desirable as a primary power source for a variety of applications that require reliable, sustainable, and clean energy. Fuel cell systems can be used to power distributed data centers that perform large-scale processing-intensive tasks, such as the training process of artificial intelligence (AI) models. Summary of the Invention [Means for solving the problem]

[0004] overview

[0004] The present disclosure is directed to a system using fuel cell-based power generation for distributed data centers performing large-scale intermittent processing-intensive tasks.

[0005] The system utilizes various support modules and techniques to accommodate power load demands on the fuel cell power module. If the fuel cell power module is unable to ramp up or increase as fast as the power load demand, additional energy is provided by one or more of an energy storage system, a load bank, and / or a secondary load (e.g., a utility grid or another connected microgrid). Conversely, if the fuel cell power module generates excess power, the excess power is provided to one or more of the energy storage system, the load bank, and the secondary load.

[0006]

[0006] As a result, the fluctuations in the power output by the fuel cell power module can be selectively minimized and the life of the fuel cell power module can be extended.

[0007] A brief description of some of the figures in the drawing

[0007] In the drawings, the same reference numbers identify similar features or elements. The dimensions and relative positions of features in the drawings are not necessarily drawn to scale. [Brief explanation of the drawings]

[0008] [Figure 1] 8 illustrates a power signal indicating power consumption during training and checkpointing of a model training process, according to one embodiment disclosed herein. [Figure 2]

[0009] 10 illustrates power signals indicating software crash and recovery of a model training process and power consumption during standby states, according to one embodiment disclosed herein. [Figure 3]

[0010] 1 illustrates a system according to one embodiment disclosed herein. [Figure 4]

[0011] 1 illustrates a fuel cell power system according to one embodiment disclosed herein. [Figure 5]

[0012] 1 illustrates power signals of a primary load, a fuel cell power system, and an energy storage system when the fuel cell power system follows a full load cycle of the primary load, according to one embodiment disclosed herein. [Figure 6]

[0013] 1 illustrates power signals of a primary load, a fuel cell power system, and an energy storage system as the fuel cell power system follows a portion of the primary load's duty cycle, according to one embodiment disclosed herein. [Figure 7]

[0014] 10 illustrates power signals of a primary load, a fuel cell power system, and an energy storage system when the fuel cell power system follows a portion of the primary load's duty cycle, according to another embodiment disclosed herein. [Figure 8]

[0015] 1 illustrates power signals of a primary load, a fuel cell power system, and an energy storage system when the fuel cell power system remains at a constant power level, according to one embodiment disclosed herein. [Figure 9]

[0016] 1 illustrates a method for managing power balance in a system according to one embodiment disclosed herein. [Figure 9A] 1 illustrates a method for managing power balance in a system according to one embodiment disclosed herein. [Figure 9B] 1 illustrates a method for managing power balance in a system according to one embodiment disclosed herein. DETAILED DESCRIPTION OF THE INVENTION

[0009] Detailed Description

[0017] In the following description, certain specific details are set forth to provide a thorough understanding of various aspects of the disclosed subject matter. However, the disclosed subject matter may be practiced without these specific details. In some instances, well-known structures, functions, and methods of manufacture of electronic devices, electronic components, and power systems are not described in detail to avoid obscuring the description of other aspects of the present disclosure.

[0010]

[0018] Unless the context requires otherwise, throughout the following specification and claims, the term "comprises" and variations thereof, such as "includes" and "comprising," are to be interpreted in an open and inclusive sense (i.e., including, but not limited to).

[0011]

[0019] References throughout this specification to "one embodiment" or "an embodiment" mean that a particular feature, structure, or characteristic described in connection with an embodiment is included in at least one embodiment. Thus, the appearances of the phrases "in one embodiment" or "in an embodiment" in various places throughout this specification are not necessarily all referring to the same aspect. Furthermore, particular features, structures, or characteristics may be combined in any suitable manner in one or more aspects of the disclosure.

[0012]

[0020] As discussed above, fuel cell systems can provide efficient, clean, and continuous power and are therefore a desirable option for a variety of applications. However, the power consumption profile of many applications can present challenges to fuel cell systems. While fuel cell systems, particularly high-temperature technology-based fuel cell systems (e.g., SOFC systems), easily handle an application's base load, they may face rapid changes in load (frequent, fast-changing loads and long, multi-day standby times). For example, SOFC systems may not be able to adjust their power levels to accommodate load changes within the desired time period. Furthermore, frequent, fast-changing loads can degrade fuel cell systems and shorten their lifespan.

[0013]

[0021] An AI model includes both a selected set of algorithms and the data used to train those algorithms to enable them to make the most accurate predictions. These algorithms can be thought of as equations with undefined coefficients. An AI model comes into being when an algorithm understands a data set to determine the best-fit coefficient values, thereby generating a model for prediction. AI model training, therefore, refers to the process of feeding data to an algorithm, examining the results, and modifying the model output to improve accuracy and efficiency. Usefully, AI model algorithms typically process vast amounts of data.

[0014]

[0022] The power consumption profile of the AI ​​model training process is particularly challenging for fuel cell systems due to their highly fluctuating loads: the model training process varies between multiple different loads due to training processes, checkpointing, software crashes and recovery, and standby states.

[0015]

[0023] Examples of power consumption profiles during training, checkpointing, software crash and recovery, and standby states are described below with respect to FIGS.

[0016]

[0024] 1 illustrates a power signal 10 indicating power consumption during training and checkpointing of a model training process, according to one embodiment disclosed herein. The model training process changes between training and checkpointing.

[0017]

[0025] The vertical axis is an amplitude axis and the horizontal axis is a time axis. The amplitude axis indicates the power level of the power signal 10. The amplitude axis may have any amplitude unit (e.g., watts, volts, amperes, etc.) that indicates power. The time axis may have any time unit that indicates time.

[0018]

[0026] The power signal 10 is the load power high value LP H and low load power value LP L The load power high value LP H is the power consumption during the training process, and t T The training period is: L is the power consumption during checkpoint processing, and t CP The checkpoint period is

[0019]

[0027] High load power LP H and low load power value LP L The power level of depends on the type and size of the model being trained. L High load power LP H The ratio may be, for example, 15% to 25%. H and low load power value LP L The power level of may also vary over time.

[0020]

[0028] Training period t T and checkpoint period t CP depends on the type and size of the model being trained and the amount of data stored during checkpoints. T The checkpoint period t can be from several tens of seconds to several tens of minutes. CP The training period t can be from a few seconds to a few minutes. T and checkpoint period t CP may also vary over time.

[0021]

[0029] As can be seen in FIG. 1, the power consumption during training and checkpointing varies significantly, with varying periods of high load power. H and low load power value LP L The power signal 10 switches between two consecutive low load power values ​​LP L During this time, the load power is high for training. H increases to.

[0022]

[0030] FIG. 2 illustrates a power signal 12 that indicates software crash and recovery of the model training process and power consumption during standby states, according to one embodiment disclosed herein.

[0023]

[0031] 1, the vertical axis is the amplitude axis and the horizontal axis is the time axis. The amplitude axis indicates the power level of the power signal 12.

[0024]

[0032] In FIG. 2, two training processes are performed: training job 1 and training job 2. In each of training job 1 and training job 2, the power signal 12 is trained according to the t T Load power high value LP with training period H and,t CP Load power low value LP with checkpoint period L It varies between.

[0025]

[0033] Specifically, in training job 1, the power signal 12 is T1 Load power high value LP with training period H1 and,t CP1 Load power low value LP with checkpoint period L1 In training job 2, the power signal 12 varies between t T2 Load power high value LP with training period H2 and,t CP2 Load power low value LP with checkpoint period L2 It varies between.

[0026]

[0034] The power level value and duration of the power level may vary between training processes. For example, in FIG. 2, the low load power value LP L2 is the low load power value LP of training job 1 L1 Higher, t of training job 1 T1 The training period of training job 2 is tT2 training period of training job 1 is shorter than t CP1 The checkpoint period of training job 2 is t CP2 shorter than the checkpoint period.

[0027]

[0035] During training job 1, the model training process experiences a software crash (e.g., the training process fails or is not performed correctly) and takes steps to recover from the software crash (e.g., restart training job 1). When the software crash occurs at time T1, the power level of power signal 12 drops to 0 for a software crash recovery period t SCR Load power software crash recovery value LP SCR (Load power low value LP L1 (lower). Software crash recovery period t SCR is usually t CP1 When the software recovers at time T2, the power level of the power signal 12 is set to the load power high value LP H1 Return to.

[0028]

[0036] Training job 1 is completed at time T3, and the model training process enters a standby state, which is a low-power state in which the training process is stopped and minimal or no processing is performed. As a result, at time T3, the power level of power signal 12 is set to 0 for the standby period t SB Load power standby value LP SB (Load power software crash recovery value LP SCR The standby state ends at time T4, and then training job 2 begins.

[0029]

[0037] 2, the power consumption during software crash, recovery and standby states shows further fluctuations in power consumption in addition to the fluctuations due to training and checkpointing. The power signal 12 varies the load power in response to a software crash, as well as the software crash recovery value LP SCR When it recovers, the load power drops to a high value LP H1 Furthermore, the power consumption is calculated according to the model training process to enter the standby state, by adjusting the load power standby value LP SB It descends again to

[0030]

[0038] The present disclosure is directed to a system architecture and control method using fuel cell-based power generation as a primary energy source. The system includes various support modules and technologies to accommodate the power load demands of various applications, such as large-scale model training data centers. That is, the system provides a microgrid architecture that combines a fuel cell system as a primary power source with other support technologies for energy storage and consumption. The support technologies maintain a constant power draw from the fuel cell system or at least reduce the amplitude of power swings experienced by the fuel cell system.

[0031]

[0039] 3 illustrates a system 14 according to one embodiment disclosed herein, which provides a fuel cell-based microgrid that accommodates processes with highly fluctuating power consumption levels.

[0032]

[0040] System 14 includes a fuel cell power system 16, a first energy storage system 18, a second energy storage system 20, a load bank 22, a power converter 24, a load interface converter 26, and a controller 28. Fuel cell power system 16, first energy storage system 18, second energy storage system 20, load bank 22, and load interface converter 26 are coupled to each other and to controller 28 by a power bus 15.

[0033]

[0041] The fuel cell power system 16 includes one or more fuel cell power modules that serve as a primary power source for a load connected to the system 14 (e.g., a primary load connected to a load interface converter 26). For example, Figure 4 illustrates a fuel cell power system 16 according to one embodiment disclosed herein.

[0034]

[0042] In this example, fuel cell power system 16 is a modular fuel cell power system that provides flexible system installation and operation. The modules allow for scaling of installed power generation capacity, reliable power generation, fuel processing flexibility, and power supply output voltage and frequency flexibility. The design also provides an easy means of scaling up to meet the special requirements of customer installations. The modular design also allows for the use of available fuels and required voltages and frequencies, which may vary by customer and / or geographic region.

[0035]

[0043] Fuel cell power system 16 includes one or more fuel cell power modules 30 and one or more power conditioning (i.e., electrical output) modules 32. In some embodiments, power conditioning module 32 is configured to deliver direct current (DC). In alternative embodiments, power conditioning module 32 is configured to deliver alternating current (AC). In these embodiments, power conditioning module 32 includes a mechanism (such as an inverter) for converting DC to AC.

[0036]

[0044] 4, the exemplary fuel cell power system 16 includes a row of seven fuel cell power modules 30 and one power conditioning module 32 disposed on a pad 34. However, the fuel cell power system 16 may include any number of fuel cell power modules and power conditioning modules and any number of rows of modules. For example, the fuel cell power system 16 may include two rows of fuel cell power modules 30 arranged back-to-back / end-to-end.

[0037]

[0045] Each fuel cell power module 30 is configured to house one or more hot boxes 36. Each hot box 36 contains one or more stacks or rows of fuel cells, such as one or more stacks or rows of SOFCs made up of fuel cells (anodes, cathodes, and electrolytes) separated by conductive interconnect plates.

[0038]

[0046] Fuel cell stacks can include internally and / or externally manifolded stacks. For example, a fuel cell stack can be internally manifolded for fuel and air, with fuel and air risers passing through openings in the fuel cell layers and / or interconnect plates between the fuel cells. Alternatively, a fuel cell stack can be internally manifolded for fuel and externally manifolded for air, with only fuel inlets and exhaust risers passing through openings in the fuel cell layers and / or interconnect plates between the fuel cells.

[0039]

[0047] Fuel cells may have a cross-flow (oxidant and fuel flow generally perpendicular to each other on either side of the fuel cell), parallel counter-flow (oxidant and fuel flow generally parallel to each other on either side of the fuel cell but in opposite directions) or parallel co-flow (air and fuel flow in the same direction generally parallel to each other on either side of the fuel cell) configuration.

[0040]

[0048] The power conditioning module 32 may include components for converting the fuel cell stack-generated DC power to AC power (e.g., DC / DC and DC / AC converters), electrical connectors for outputting the AC power to the grid, circuitry for managing electrical transients, and a system controller (e.g., a computer or dedicated control logic device or circuitry). The power conditioning module 32 may be designed to convert the DC power from the fuel cell module to a variety of AC voltages and frequencies. Designs for 208V, 60Hz; 480V, 60Hz; 415V, 50Hz, and other common voltages and frequencies may be provided.

[0041]

[0049] The linear array of fuel cell power modules 30 is easily scaled. For example, more or fewer fuel cell power modules 30 may be provided depending on the power requirements of the building or other facility served by the fuel cell power system 16. Other ratios of fuel cell power modules 30 and input / output modules may be provided. For example, in other exemplary embodiments, more or fewer fuel cell power modules 30 may be provided.

[0042]

[0050] The fuel cell power system 16 is configured to facilitate maintenance of the components of the fuel cell power system 16 .

[0043]

[0051] For example, the fuel cell power system 16 may include an access door 38. All of the regularly or frequently serviced parts, such as consumable parts, may also be located in a single module to reduce the amount of time required for maintenance operations.

[0044]

[0052] As another example, when one fuel cell power module 30 is taken offline (i.e., no power is generated by the stack in the hot box 36 in the offline fuel cell power module 30), the remaining fuel cell power modules 30 and power conditioning module 32 are not taken offline. Furthermore, the fuel cell power system 16 may include more than one of each type of module 30 and 32. When at least one module of a particular type is taken offline, the remaining modules of the same type are not taken offline. Thus, in a system including multiple modules, each of the modules 30 or 32 can be electrically disconnected, removed from the fuel cell power system 16, and / or serviced or repaired without stopping operation of the other modules in the system, allowing the fuel cell power system 16 to continue generating electricity. The entire fuel cell power system 16 does not need to be shut down if one stack of fuel cells in one hot box 36 malfunctions or is taken offline for maintenance. Furthermore, the fuel cell system 16 may include redundant modules that can be brought online while other modules are serviced or replaced.

[0045]

[0053] Returning to FIG. 3 , the first energy storage system 18 and the second energy storage system 20 are energy or power storage systems used to reduce the amplitude of power swings experienced by the fuel cell power system 16. The first energy storage system 18 and the second energy storage system 20 act as power buffers that store excess power for delivery to the power bus 15 when load power (e.g., the power consumption of a primary load connected to the load interface converter 26) is lower than the intended power level of the fuel cell power system 16. Additionally, the first energy storage system 18 and / or the second energy storage system 20 provide power to the power bus 15 when there is a power deficiency on the power bus 15 (e.g., the load power is greater than the power output of the fuel cell power system 16). Although two energy storage systems are shown in FIG. 3 , the system 14 may include any number of energy storage systems. Various methods of operation of the first energy storage system 18 and the second energy storage system 20 are discussed in further detail below.

[0046]

[0054] In one embodiment, fuel cell power system 16, first energy storage system 18 and second energy storage system 20 are unified at the DC level, thus making power bus 15 a DC power bus and system 14 a DC coupled system.

[0047]

[0055] Types of storage systems that can be used for first energy storage system 18 and second energy storage system 20 include, for example, ultracapacitors, supercapacitors, various types of battery technologies, rotary storage systems such as flywheels, and thermal storage systems.

[0048]

[0056] The first energy storage system 18 and the second energy storage system 20 are different types of storage systems for achieving optimal performance of the system 14. In one embodiment, the first energy storage system 18 stores a low total amount of power but may achieve fast charge and discharge times. In contrast, the second energy storage system 20 is an energy storage system that stores a large total amount of power (e.g., greater than the total amount of power of the first energy storage system 18) but has slow charge and discharge times (e.g., slower than the charge and discharge times of the first energy storage system 18). With storage systems having different characteristics, the first energy storage system 18 or the second energy storage system 20 is selected depending on the current power consumption of the primary load. In one embodiment, the first energy storage system 18 is used to supplement the power generated by the fuel cell power system 16 and to store excess power generated by the fuel cell power system 16. The second energy storage system 20 is also used to store excess power generated by the fuel cell power system 16 unless the first energy storage system 18 is fully charged (e.g., charged to a level above a predetermined threshold) or has a large amount of excess power (e.g., in a standby state).

[0049]

[0057] The load bank 22 is a controllable load bank that converts excess power on the power bus 15 into thermal energy. The amount of power consumed by the load bank 22 at any given time depends on the excess power available from the fuel cell power system 16 after serving the primary load connected to the load interface converter 26 and the excess power available from the first energy storage system 18 and the second energy storage system 20. Any thermal energy generated by the load bank 22 and the fuel cell power system 16 can be released to the surrounding environment or reused for various purposes (e.g., captured by using combined heat and power (CHP) equipment for a building's heating, ventilation, and air conditioning (HVAC) and other heating loads). The heat generated by the load bank 22 and the fuel cell power system 16 can be supplied to a steam generator, a thermoelectric generator, a water heater, an absorption chiller, a Rankine cycle device, or a combination thereof. For example, an absorption chiller can use the heat from the load bank 22 and the fuel cell power system 16 to drive a thermodynamic process, whereby water is cooled and distributed for HVAC demand. In this way, absorption chillers can provide needed cooling capacity to various industrial facilities (such as hospitals, universities, hotels, warehouses, and manufacturing facilities) without significantly contributing to peak electrical demand. In this particular case, the absorption chillers can be utilized to cool processing systems used to train AI models. Alternatively, heat generated from the load bank 22 and / or fuel cell power system 16 can be held in a thermal storage unit utilizing molten salts, molten silicon, molten aluminum, graphite heat transfer media, thermochemical materials, phase change materials, and the like. The heat stored in the thermal storage media can be released to generate electricity as needed through the use of a suitable heat engine (e.g., steam turbine, thermophotovoltaic, etc.). The electricity generated by the heat engine can be used for multiple purposes (e.g., supplying the first energy storage system 18, the second energy storage system 20, or any other load as needed).

[0050]

[0058] Power converter 24 is a bidirectional power converter that connects a secondary load to power bus 15 (e.g., power converter 24 is a DC / AC converter if system 14 is a DC-coupled system and the secondary load is an AC system). The secondary load is an additional load (such as a utility power grid or another connected microgrid) other than the primary load connected to load interface converter 26. Fuel cell power system 16 does not serve as the primary power source for the secondary load.

[0051]

[0059] Power converter 24 enables secondary loads to act as energy sources and reservoirs in addition to first energy storage system 18 and second energy storage system 20. For example, when there is excess power on power bus 15, power converter 24 exports the excess power to the secondary loads. Conversely, when there is a power deficit on power bus 15, power converter 24 brings power from the secondary loads to power bus 15.

[0052]

[0060] Load interface converter 26 provides power to the primary load. The primary load can be either an AC or DC type. If system 14 is a DC-coupled system and the load is an AC type, load interface converter 26 is a DC-to-AC converter that converts the DC power on power bus 15 to AC power for the primary load. Conversely, if system 14 is a DC-coupled system and the primary load is a DC type, load interface converter 26 is a DC-to-DC converter that converts the DC power on power bus 15 to a DC voltage used by the primary load.

[0053]

[0061] The primary load connected to the load interface converter 26 is primarily powered by the system 14 (more specifically, the fuel cell power system 16). Stated another way, the primary load receives the majority of its power from the fuel cell power system 16. In one embodiment, the primary load is a data center that performs a training process for an AI model. As discussed above with respect to FIGS. 1 and 2 , the power consumption profile of the training process for the AI ​​model is particularly challenging for the fuel cell power system 16 due to its highly variable load, and would therefore benefit greatly from the system 14. However, the primary load can be any type of load with variable power consumption other than an AI model.

[0054]

[0062] The controller 28 is communicatively coupled to and controls the various functions of the fuel cell power system 16, the first energy storage system 18, the second energy storage system 20, the load bank 22, the power converter 24, and the load interface converter 26. The controller 28 manages the power balance in real time between the various generation modules (e.g., the fuel cell power system 16), the storage modules (e.g., the first energy storage system 18 and the second energy storage system 20), and the loads of the system 14 (e.g., the load bank 22, the secondary loads connected to the power converter 24, and the primary loads connected to the load interface converter 26).

[0055]

[0063] The controller 28 continuously measures the current state of charge (SOC) (e.g., current stored charge) of the first energy storage system 18 and the second energy storage system 20, the power output from the fuel cell power system 16, the first energy storage system 18, the second energy storage system 20, and the secondary load to the power bus 15, and the power input from the first energy storage system 18, the second energy storage system 20, the load bank 22, the secondary load, and the power bus 15 to the primary load. The controller 28 manages the power flow to and from each of the fuel cell power system 16, the first energy storage system 18, the second energy storage system 20, the load bank 22, the secondary load, and the primary load based on the measurements by the controller 28 (e.g., sets the power levels input to and output from each of the fuel cell power system 16, the first energy storage system 18, the second energy storage system 20, the load bank 22, the secondary load, and the primary load). Various operating methods for the controller 28 are discussed in further detail below.

[0056]

[0064] Controller 28 may be any type of processor, processor signal, or controller capable of processing data and may include one or more processors. For example, controller 28 may be an industrial personal computer (PC), a programmable logic controller (PLC), a microcontroller, a digital signal processor (DSP), a field programmable gate array (FPGA), or other similar technology.

[0057]

[0065] The system 14 provides varying levels of support for the fuel cell power system 16 to accommodate the continuously fluctuating load of the primary load (e.g., a data center running a training process for an AI model). The system 14 can be configured to maintain a constant power draw from the fuel cell power system 16, or at least reduce the amplitude of power swings experienced by the fuel cell power system 16.

[0058]

[0066] In one embodiment, the controller 28 controls the fuel cell power system 16 to follow the entire load cycle of the primary load. The output of the fuel cell power system 16 increases or decreases in power along with the primary load. Furthermore, if the fuel cell power system 16 is unable to ramp up or increase as fast as the primary load requires, the energy storage system provides additional energy to supplement the power provided by the fuel cell power system 16. Additionally, during the training process, the fuel cell power system 16 provides power greater than the power consumption of the primary load to recharge the energy storage system.

[0059]

[0067] For example, FIG. 5 shows power signals of the primary load, fuel cell power system 16, and energy storage system as the fuel cell power system 16 follows the entire duty cycle of the primary load, according to one embodiment disclosed herein.

[0060]

[0068] Power signal 40 indicates the power consumption of the primary load, power signal 42 indicates the power output of fuel cell power system 16, power signal 44 indicates the power output of the energy storage system, and power signal 46 indicates the SOC of the energy storage system. The vertical axis is the amplitude axis and the horizontal axis is the time axis. Each of the power signals is plotted for an amplitude value equal to zero.

[0061]

[0069] The energy storage system can be either a first energy storage system 18 or a second energy storage system 20, depending on the desired energy storage system type. In one embodiment, the energy storage system of Figure 5 is a first energy storage system 18, where the first energy storage system 18 stores a low total amount of power but has fast charge and discharge times, and the second energy storage system 20 stores a large total amount of power but has slow charge and discharge times.

[0062]

[0070] The controller 28 controls the fuel cell power system 16 to follow the full load of the primary load. For example, referring to Figure 5, the controller 28 instructs the fuel cell power system 16 to output power such that a power signal 42 representing the power output of the fuel cell power system 16 follows a power signal 40 representing the power consumption of the primary load. The power signal 40 indicates a load power high value LP at time T1. H From the load power low value LP L Then, at time T2, the load power drops to a high value LP H Similarly, the power signal 42 returns to the fuel cell power high value FP at time T1. H Fuel cell power low value FP L At time T2, the fuel cell power is set to a high value FP H It is set again to Fuel Cell Power High Value FP H is the load power high value LP H and the fuel cell power low value FP L is the low load power value LP L is essentially equal to

[0063]

[0071] In some cases, the fuel cell power system 16 is unable to ramp up or increase power at the same rate as the primary load requires. That is, the fuel cell power system 16 increases its output power at a rate slower than the rate at which the primary load increases its power consumption. In this situation, the additional energy provided by the energy storage system is used to supplement the power provided by the fuel cell power system 16 during the ramp-up period. For example, referring to FIG. 5 , there is a ramp-up period between times T2 and T3, during which the power output of the fuel cell power system 16, the power signal 42, rises above the fuel cell power high value FP HThe power consumption of the fuel cell power system 16 cannot be immediately increased to 1 / 2. Therefore, the controller 28 commands the energy storage system to output additional power to compensate for the deficiency in the fuel cell power system 16. Between times T2 and T3, the power output of the energy storage system, power signal 44, is set so that the total output power of the fuel cell power system 16 and the energy storage system is substantially equal to the power consumption of the primary load. In other words, between times T2 and T3, power signal 44 is equal to the difference between power signals 40 and 42. The power balance between the primary load and the fuel cell power system 16 and the energy storage system is governed by the following equation: Energy storage system output = Main load power consumption - Fuel cell power system output

[0064]

[0072] During the ramp-up period, energy provided by a secondary load can also be used instead of or in addition to the power provided by the energy storage system to supplement the power provided by the fuel cell power system 16.

[0065]

[0073] As the fuel cell power system 16 ramps up to catch up with the primary load, the output from the energy storage system drops. The energy storage system stops providing power to the primary load as soon as the fuel cell power system 16 matches the primary load. For example, referring to Figure 5, as power signal 42 increases to catch up with power signal 40 between times T2 and T3, power signal 44 drops proportionally. When power signal 42 reaches power signal 40 at time T3, power signal 44 is set to zero.

[0066]

[0074] After supplying power to the primary load, the fuel cell power system 16 provides more power than the primary load's power consumption to recharge the energy storage system. The excess power from the fuel cell power system 16 is used to recharge the energy storage system for later use. In other words, a portion of the power generated by the fuel cell power system 16 is diverted to the energy storage system. For example, referring to FIG. 5 , the energy storage system is discharged between times T2 and T3 by an amount indicated by the discharge region 48. At time T3, the power signal 42, which is the power output of the fuel cell power system 16, is set higher than the power signal 40, which is the primary load's power consumption. The additional power, indicated by the charge region 50, is used to recharge the energy storage system. Discharging the energy storage system is indicated by a positive value on the power signal 44, which is the energy storage system's power output. Charging the energy storage system is indicated by a negative value on the power signal 44. This is also indicated by the power signal 46, which is the energy storage system's SOC. The power signal 46 decreases between times T2 and T3 as the energy storage system is discharged and begins to increase at time T3 as the energy storage system is recharged by the fuel cell power system 16.

[0067]

[0075] The implementation of controlling the fuel cell power system 16 to follow the entire duty cycle of the primary load is straightforward. For example, if the fuel cell power system may take 60 seconds to ramp up from no load to full load, the energy storage system is used to support the fuel cell power system 16 over the relatively short ramp-up period, so that an energy storage system with a relatively low capacity can be used.

[0068]

[0076] In another embodiment, the controller 28 controls the fuel cell power system 16 to follow a portion of the duty cycle of the primary load. The output of the fuel cell power system 16 increases or decreases in power with the primary load, but is allowed to follow a predetermined percentage of the load power swing (i.e., change in power) of the primary load to minimize the impact on the life of the fuel cell power system 16. Additionally, during checkpointing, excess power generated by the fuel cell power system 16 can be stored or consumed to maintain a power balance on the power bus 15.

[0069]

[0077] For example, FIG. 6 shows power signals of the primary load, fuel cell power system 16, and energy storage system when the fuel cell power system 16 follows a portion of the primary load's duty cycle, according to one embodiment disclosed herein.

[0070]

[0078] Power signal 52 indicates the power consumption of the primary load connected to load interface converter 26, power signal 54 indicates the power output of fuel cell power system 16, power signal 56 indicates the power output of the energy storage system, power signal 58 indicates the SOC of the energy storage system, and power signal 60 indicates the excess power generated by fuel cell power system 16. The vertical axis is the amplitude axis and the horizontal axis is the time axis. Each of the power signals is shown for an amplitude value equal to zero.

[0071]

[0079] As discussed above, the energy storage system can be either a first energy storage system 18 or a second energy storage system 20, depending on the desired energy storage system type. In one embodiment, the energy storage system of Figure 6 is a first energy storage system 18 that stores a low total amount of power but has fast charge and discharge times, and a second energy storage system 20 that stores a large total amount of power but has slow charge and discharge times, used in addition to the first energy storage system 18 to store excess power when the first energy storage system 18 is fully charged.

[0072]

[0080] The controller 28 controls the fuel cell power system 16 to follow a portion of the load of the primary load. For example, referring to Figure 6, the controller 28 instructs the fuel cell power system 16 to output power such that the power signal 54 follows a predetermined portion or a predetermined percentage of the swing of the power signal 52, which is the power consumption of the primary load. The power signal 52 outputs a load power high value LP at time T1. H From the load power low value LP L Then, at time T2, the load power drops to a high value LP H Similarly, the power signal 54 returns to the fuel cell power high value FP at time T1. H Fuel cell power low value FP L At time T2, the fuel cell power is set to a high value FP H However, the fuel cell power high value FP H is the load power high value LP H Lower and fuel cell power low value FP L is the low load power value LP L Greater than.

[0073]

[0081] During the checkpointing process between times T1 and T2, the power signal 52, which is the power consumption of the primary load, reaches its lowest value (load power low value LP L ) The power signal 54, which is the power output of the fuel cell power system 16, is set to a low load power value LP L Higher fuel cell power, lower FP L , thereby minimizing the impact on the life of the fuel cell power system 16.

[0074]

[0082] Fuel cell power low value FP L is the low load power value LP L Higher fuel cell power, lower FP value L is the excess power (e.g., fuel cell power low value FP L and low load power value LP Land T2). Power signal 60 indicates the excess power generated by fuel cell power system 16 and increases to above zero between times T1 and T2. The excess power is either stored or consumed to maintain a power balance on power bus 15, thereby keeping the voltage level of power bus 15 within acceptable limits.

[0075]

[0083] The first priority for managing excess power is to store the excess power in the energy storage system (e.g., one or both of first energy storage system 18 and second energy storage system 20). Charging the energy storage system is achieved by a negative value of power signal 56, which is the power output of the energy storage system between times T1 and T2, and a negative value of power signal 58, which is the SOC of the energy storage system between times T1 and T2. Min From SOC Max The energy storage system is charged by the amount indicated by the charging field 62.

[0076]

[0084] When the energy storage system is no longer able to store the excess power (e.g., the first energy storage system 18 is fully charged or experiences a partial or complete failure), the excess power is exported to another energy storage system (e.g., the second energy storage system 20).

[0077]

[0085] If all energy storage systems are unable to store any more excess power, the excess power is exported to a secondary load via power converter 24. For example, if the secondary load is a utility grid or another connected microgrid, the excess power is converted by power converter 24 for use by the utility grid or microgrid.

[0078]

[0086] If excess power still exists after the energy storage system and secondary loads, the excess power is diverted to the load bank 22 for conversion to thermal energy. As discussed above, the thermal energy can be released to the surrounding environment or reused for various purposes (e.g., captured by using CHP equipment for the building's HVAC and other heating loads). The amount of excess energy consumed through the secondary loads and load bank 22 is dictated by the consumption area 64.

[0079]

[0087] In the event of a load bank 22 failure, the fuel cell power system 16 may be ramped down to a lower output power.

[0080]

[0088] During the training process before time T1 and after time T2, the power signal 52, which is the power consumption of the primary load, reaches its highest value (load power high value LP H ) The power signal 54, which is the power output of the fuel cell power system 16, is fed back to the load power high value LP to ensure that the energy storage system is not fully charged as discussed above and therefore has the capacity to store excess power from the checkpointing process. H Lower fuel cell power Higher FP H 5, the energy storage system is used to supplement the power provided by the fuel cell power system 16 during the ramp-up period (e.g., between time T2 and time T3 in FIG. 6) because the fuel cell power system 16 cannot ramp up at the same rate as the primary load. However, the energy consumed by the energy storage system during the ramp-up period may not fully discharge the energy storage system. As a result, the energy storage system may not be able to take the excess power from the checkpointing process. To avoid this, the fuel cell power system 16 outputs less power than the primary load (e.g., a fuel cell power high value FP H The load power is high LP H(set to less than 0.05 V) and the difference between the output power of the fuel cell power system 16 and the load of the primary load is compensated for by the energy storage system. As a result, the energy storage system is slightly discharged during the training process (e.g., power signal 56, the power output of the energy storage system, is positive) and then fully discharged in the subsequent checkpoint process (e.g., power signal 58, the SOC of the energy storage system, is negative). Min The energy storage system is discharged by the amount indicated by the discharge area 66.

[0081]

[0089] The energy storage system is fully discharged (e.g., SOC Min , the controller 28 reduces the amount of power consumed by the load bank 22 and then the secondary load. If there is still a power deficit on the primary load power bus 15, the controller 28 reduces the output power of the fuel cell power system 16 to the load power high value LP H Increase to.

[0082]

[0090] When the training process transitions to the checkpoint process, the load of the main load drops (for example, the load power high value LP at time T1 in Figure 6). H From the load power low value LP LThe charge-discharge time (drop to 0.5 V) is nearly instantaneous. Therefore, the response time of the energy storage system is important to properly handle excess power generated by the fuel cell power system 16 during a load drop of the primary load. For optimal performance, the energy storage system should have a large amount of storage to store the excess power generated by the fuel cell power system 16 and a fast response time to immediately store the excess power during a load drop. To achieve this, various types of energy storage are used. For example, in one embodiment, the first energy storage system 18 stores a low total amount of power but has fast charge and discharge times, and the second energy storage system 20 stores a large total amount of power but has slow charge and discharge times. The first energy storage system 18, which is capable of a fast charge time, is used to handle the excess power during a load drop. Once the first energy storage system 18 is fully charged, the second energy storage system 20, which has a large capacity, is used to store any remaining excess power.

[0083]

[0091] Alternatively, the controller 28 controls the fuel cell power system 16 to follow the duty cycle of the primary load, as discussed above with respect to Figure 6. However, the output of the fuel cell power system 16 is ramped down before immediately dropping. For example, Figure 7 shows power signals of the primary load, the fuel cell power system 16, and the energy storage system when the fuel cell power system 16 follows a portion of the duty cycle of the primary load, according to another embodiment disclosed herein.

[0084]

[0092] FIG. 7 shows a power signal 52 indicative of the power consumption of the primary load connected to the load interface converter 26 discussed above with respect to FIG. 6, a power signal 54 indicative of the power output of the fuel cell power system 16, a power signal 56 indicative of the power output of the energy storage system, a power signal 58 indicative of the SOC of the energy storage system, and a power signal 60 indicative of the excess power generated by the fuel cell power system 16.

[0085]

[0093] In contrast to the embodiment shown in FIG. 6, the controller 28 controls the power signal 54 of the fuel cell power system 16 to a high fuel cell power value FP between times T1 and T4. H Fuel cell power low value FP L The fuel cell power system 16 is configured to ramp down to a maximum of 1000 W. Ramping down the fuel cell power system 16 relaxes the demands on the energy storage system, allowing the energy storage system to have a slower charge time. Additionally, excess power during ramp down of the fuel cell power system 16 is diverted to the secondary loads and load bank 22 (shown as dissipation region 64 between times T1 and T4). This approach reduces the overall cost of having multiple, diverse energy storage systems while still improving the performance of the fuel cell power system 16.

[0086]

[0094] In another embodiment, the controller 28 controls the fuel cell power system 16 to output a constant power level, where the output of the fuel cell power system 16 does not scale with the primary load, but remains the same.

[0087]

[0095] For example, FIG. 8 shows power signals of the primary load, fuel cell power system 16, and energy storage system when the fuel cell power system 16 remains at a constant power level, according to one embodiment disclosed herein.

[0088]

[0096] Power signal 68 indicates the power consumption of the primary load, power signal 70 indicates the power output of fuel cell power system 16, power signal 72 indicates the power output of the energy storage system, power signal 74 indicates the SOC of the energy storage system, and power signal 76 indicates the excess power generated by fuel cell power system 16. The vertical axis is the amplitude axis and the horizontal axis is the time axis. Each of the power signals is shown for an amplitude value equal to zero.

[0089]

[0097] As discussed above, the energy storage system can be either a first energy storage system 18 or a second energy storage system 20, depending on the desired energy storage system type. In one embodiment, the energy storage system of Figure 8 is a first energy storage system 18 that stores a low total amount of power but has fast charge and discharge times, and a second energy storage system 20 that stores a large total amount of power but has slow charge and discharge times, used in addition to the first energy storage system 18 to store excess power once the first energy storage system 18 is fully charged.

[0090]

[0098] The operating modes and control sections in this embodiment are similar to the embodiment discussed above with respect to Figure 6. For example, excess power is stored in the energy storage system (indicated by charging region 78) and then in the secondary loads and load bank 22 (indicated by consumption region 80) during the checkpointing process, and the energy storage system is used to supplement the power provided by the fuel cell power system 16 (indicated by discharging region 82) during the training process.

[0091]

[0099] 6, the controller 28 instead controls the fuel cell power system 16 to output a constant power level. For example, referring to FIG. 8, the power signal 68, which is the power consumption of the primary load, reaches a load power high value LP at time T1. H From the load power low value LP L Then, at time T2, the load power drops to a high value LP H Nevertheless, the controller 28 instructs the fuel cell power system 16 to output power such that the power signal 70 remains at the same level FP. In one embodiment, the fuel cell power system 16 returns to the load power high value LP H Smaller and lower load power LP L It outputs a larger constant power.

[0092]

[0100] The fuel cell power system 16 does not follow the load power fluctuations of the primary load, but rather outputs a substantially constant average load power within desired limits through any changes in the load of the primary load (e.g., checkpoint and training processes). By maintaining the fuel cell power system 16 at a substantially constant power output level, the load behavior of the primary load does not have any effect on the life of the fuel cell power system 16.

[0093]

[0101] As discussed above with respect to Figure 2, the model training process also experiences software crashes and recoveries and standby states that cause additional fluctuations in the load. For example, the load drops significantly when a software crash occurs, increases significantly upon recovery, and drops to near zero when the model training enters standby.

[0094]

[0102] Software crashes (and the recovery process) are unplanned events that occur sporadically. Furthermore, the recovery period can last from several seconds to several minutes, depending on the cause of the software crash. As a result, software crashes and recovery, like the training and checkpointing processes discussed above, are difficult to manage proactively. Therefore, during the software crash and recovery process, excess power and output power of the fuel cell power system 16 are handled dynamically. In one embodiment, the controller 28 monitors the power balance between the fuel cell power system 16, the first energy storage system 18, the second energy storage system 20, the load bank 22, the secondary load, and the primary load in real time. In response to detecting excess power on the power bus 15, the controller 28 stores the excess power in the first energy storage system 18 and / or the second energy storage system 20, then directs the excess power to the secondary load if the first energy storage system 18 and / or the second energy storage system 20 are full, or to the load bank 22 if the system is unable to export power to the secondary load. If the secondary loads and / or load banks cannot accommodate the excess power from the fuel cell power system 16 (e.g., due to being full or a failure), the controller 28 reduces the output power of the fuel cell power system 16 to a lower level as a last resort.

[0095]

[0103] The standby state is a planned event that can last from several hours to several days between training jobs. Excess power generated by the fuel cell power system 16 during the standby state is stored in the first energy storage system 18 and, once the first energy storage system 18 is fully charged, in the second energy storage system 20. However, the extended duration of the standby state makes it impractical to store the excess power solely in the energy storage system. Thus, during a standby state having an extended duration, once the first energy storage system 18 and the second energy storage system 20 are fully charged, the controller 28 may direct the excess power to a secondary load, direct the excess power to the load bank 22, reduce the output power of the fuel cell power system 16 to a lower level, or a combination thereof.

[0096]

[0104] 9, 9A, and 9B illustrate a method 84 for managing the power balance of the system 14 according to one embodiment disclosed herein. The method 84 is performed by the controller 28, which monitors and manages the power balance between the fuel cell power system 16, the first energy storage system 18, the second energy storage system 20, the load bank 22, the secondary loads connected to the power converter 24, and the primary loads connected to the load interface converter 26.

[0097]

[0105] In block 86, the controller 28 determines the power and status of all components of the system 14. The controller 28 measures the current SOC of the first energy storage system 18 and the second energy storage system 20, the power output from the fuel cell power system 16, the first energy storage system 18, the second energy storage system 20, and the secondary load to the power bus 15, and the power provided from the power bus 15 to the first energy storage system 18, the second energy storage system 20, the load bank 22, the secondary load, and the primary load.

[0098]

[0106] In block 88, the controller 28 determines the total amount of power on the power bus 15. In one embodiment, the controller 28 determines the total amount of power by adding all power coming in from the fuel cell power system 16, the first energy storage system 18, the second energy storage system 20, and the secondary loads, and subtracting all power going out to the first energy storage system 18, the second energy storage system 20, the load bank 22, the secondary loads, and the primary load.

[0099]

[0107] At block 90, controller 28 determines or detects whether there is excess power on power bus 15. In one embodiment, controller 28 determines that there is excess power on power bus 15 if the total amount of power on power bus 15 from block 88 is greater than or equal to a predetermined excess threshold. If controller 28 determines that there is no excess power on power bus 15 (e.g., the total amount of power is less than the excess threshold), method 84 moves to block 92. Conversely, if controller 28 determines that there is excess power on power bus 15 (e.g., the total amount of power is greater than or equal to the excess threshold), method 84 moves to block 94. In one embodiment, the excess threshold is greater than or equal to a threshold of the current power consumption of the primary load.

[0100]

[0108] At block 92, controller 28 determines or detects whether there is a power deficit on power bus 15. In one embodiment, controller 28 determines that there is a power deficit on power bus 15 if the total amount of power on power bus 15 from block 88 is less than or equal to a predetermined deficit threshold. If controller 28 determines that there is no power deficit on power bus 15 (e.g., the total amount of power is greater than the deficit threshold), method 84 moves to block 96. Conversely, if controller 28 determines that there is a power deficit on power bus 15 (e.g., the total amount of power is less than or equal to the deficit threshold), method 84 moves to block 98. In one embodiment, the deficit threshold is less than the current power consumption of the primary load and less than the excess threshold.

[0101]

[0109] At block 96, method 84 is repeated, beginning again at block 86.

[0102]

[0110] Blocks 98 , 100 , 102 , 104 , 106 , 108 , 110 , 112 and 114 , discussed below, are performed to reduce the power shortage on power bus 15 .

[0103]

[0111] At block 98, the controller 28 determines whether the power output to the load bank 22 may be decreased. In one embodiment, the controller 28 determines that the power output to the load bank 22 may be decreased if the power output to the load bank 22 is greater than or equal to a predetermined load bank threshold. If the controller 28 determines that the power output to the load bank 22 may be decreased (e.g., the power output to the load bank 22 is greater than or equal to the load bank threshold), the method 84 moves to block 100. Conversely, if the controller 28 determines that the power output to the load bank 22 may not be decreased (e.g., the power output to the load bank 22 is less than the load bank threshold), the method 84 moves to block 102.

[0104]

[0112] In block 100, the controller 28 reduces the power output from the power bus 15 to the load bank 22. As a result, power increases on the power bus 15, and this additional power is used to compensate for the power deficit. For example, the additional power originally diverted to the load bank 22 may be used to provide power to the primary load, charge the first energy storage system 18, charge the second energy storage system 20, or a combination thereof. Once the command to reduce the power to the load bank is provided, the method 84 proceeds to block 96, where the method 84 is repeated, beginning at block 86.

[0105]

[0113] At block 102, controller 28 determines whether the power output to a secondary load connected to power converter 24 may be reduced. As discussed above, the secondary load may be a utility grid or another connected microgrid that can both store, use, and / or generate power. In one embodiment, controller 28 determines whether the power output to the secondary load may be reduced if the power output to the secondary load is greater than or equal to a predetermined secondary load threshold. If controller 28 determines that the power output to the secondary load may be reduced (e.g., the power output to the secondary load is greater than or equal to the secondary load threshold), method 84 moves to block 104. Conversely, if controller 28 determines that the power output to the secondary load may not be reduced (e.g., the power output to the secondary load is less than the secondary load threshold), method 84 moves to block 106.

[0106]

[0114] In block 104, controller 28 reduces the power output from power bus 15 to the secondary loads. As a result, power increases on power bus 15, and this additional power is used to compensate for the power deficit. For example, the additional power originally diverted to the secondary loads may be used to provide power to the primary load, charge first energy storage system 18, charge second energy storage system 20, or a combination thereof. Once the command to reduce the power output to the secondary loads is provided, method 84 proceeds to block 96, where method 84 is repeated, beginning again at block 86.

[0107]

[0115] In block 106, the controller 28 determines whether the discharge power from the first energy storage system 18 or the second energy storage system 20 may be increased. Stated differently, the controller 28 determines whether additional power may be provided to the power bus 15 from the first energy storage system 18 or the second energy storage system 20. In one embodiment, the controller 28 determines that the discharge power from the first energy storage system 18 may be increased if the SOC of the first energy storage system 18 is greater than or equal to a predetermined first SOC threshold, and determines that the discharge power from the second energy storage system 20 may be increased if the SOC of the second energy storage system 20 is greater than or equal to a predetermined second SOC threshold. If the controller 28 determines that the discharge power from the first energy storage system 18 or the second energy storage system 20 may be increased (e.g., the SOC is greater than or equal to the corresponding SOC threshold), the method 84 moves to block 108. Conversely, if controller 28 determines that the discharge power from first energy storage system 18 or second energy storage system 20 cannot be increased (e.g., the SOC is below the corresponding SOC threshold), method 84 moves to block 110. The first SOC threshold and the second SOC threshold may be equal or different.

[0108]

[0116] In block 108, in response to determining that the discharge power from the first energy storage system 18 may be increased, the controller 28 increases the discharge power from the first energy storage system 18 to the power bus 15, and in response to determining that the discharge power from the second energy storage system 20 may be increased, the controller 28 increases the discharge power from the second energy storage system 20. As a result, power is increased on the power bus 15, and this additional power is used to compensate for the power deficit. For example, the additional power is used to provide power to the primary load. Once a command to increase the discharge power from the first energy storage system 18 and / or the second energy storage system 20 is provided, the method 84 proceeds to block 96, where the method 84 is repeated, beginning again at block 86.

[0109]

[0117] At block 110, controller 28 determines whether the power imported from the secondary load can be increased. Stated differently, controller 28 determines whether additional power can be provided from the secondary load to power bus 15. In one embodiment, controller 28 determines that the power imported from the secondary load can be increased if the available power of the secondary load (e.g., the total amount of power stored by the secondary load or the amount of power currently being generated by the secondary load) is greater than or equal to a predetermined available power threshold. If controller 28 determines that the power imported from the secondary load can be increased (e.g., the available power of the secondary load is greater than or equal to the predetermined available power threshold), method 84 moves to block 112. Conversely, if controller 28 determines that the power imported from the secondary load cannot be increased (e.g., the available power of the secondary load is less than the predetermined available power threshold), method 84 moves to block 114.

[0110]

[0118] In block 112, controller 28 increases the power imported from the secondary load onto power bus 15. As a result, power is increased on power bus 15, and this additional power is used to compensate for the power deficit. For example, the additional power may be used to provide power to the primary load, charge first energy storage system 18, charge second energy storage system 20, or a combination thereof. Method 84 then moves to block 96, where method 84 is repeated, beginning again at block 86.

[0111]

[0119] In block 114, controller 28 increases the power output of fuel cell power system 16 to power bus 15 to compensate for the deficiency on power bus 15. This additional power is used to provide power to the primary load, charge first energy storage system 18, charge second energy storage system 20, or a combination thereof. Method 84 then moves to block 96, where method 84 is repeated, beginning again at block 86.

[0112]

[0120] As a result of blocks 98, 100, 102, 104, 106, 108, 110, 112, and 114, the fuel cell power system 16 is used as a last resort to correct a power shortage on the power bus 15. The power output of the fuel cell power system 16 is increased if the shortage cannot be corrected by adjusting the power of the load bank 22, the secondary load, the first energy storage system 18, and the second energy storage system 20. Thus, power swings at the output of the fuel cell power system 16 are reduced, and degradation of the lifespan of the fuel cell power system 16 is minimized.

[0113]

[0121] Returning to block 90, as discussed above, method 84 moves to block 94 if controller 28 determines that there is excess power on power bus 15. Blocks 94, 116, 118, 120, 122, 124, and 126, discussed below, are performed to manage the excess power on power bus 15.

[0114]

[0122] In block 94, the controller 28 determines whether the charging power to the first energy storage system 18 or the second energy storage system 20 may be increased. Stated differently, the controller 28 determines whether additional power may be stored in the first energy storage system 18 or the second energy storage system 20. In one embodiment, the controller 28 determines that the charging power to the first energy storage system 18 may be increased if the SOC of the first energy storage system 18 is below a predetermined first SOC threshold, and determines that the charging power to the second energy storage system 20 may be increased if the SOC of the second energy storage system 20 is below a predetermined second SOC threshold. If the controller 28 determines that the charging power to the first energy storage system 18 or the second energy storage system 20 may be increased (e.g., the SOC is below a corresponding SOC threshold), the method 84 moves to block 116. Conversely, if controller 28 determines that the charging power to first energy storage system 18 or second energy storage system 20 cannot be increased (e.g., the SOC is above the corresponding SOC threshold), method 84 moves to block 118. The first SOC threshold and the second SOC threshold may be equal or different.

[0115]

[0123] In block 116, in response to determining that the charging power to the first energy storage system 18 may be increased, the controller 28 increases the charging power from the power bus 15 to the first energy storage system 18, and in response to determining that the charging power to the second energy storage system 20 may be increased, the controller 28 increases the charging power to the second energy storage system 20. As a result, excess power on the power bus 15 is utilized and diverted to the first energy storage system 18 and / or the second energy storage system 20. As discussed above, the energy stored in the first energy storage system 18 and the second energy storage system 20 is also used to supplement the power provided by the fuel cell power system 16. Once a command to increase the charging power to the first energy storage system 18 and / or the second energy storage system 20 is provided, the method 84 proceeds to block 96, where the method 84 is repeated, resuming at block 86.

[0116]

[0124] At block 118, controller 28 determines whether the power exported to the secondary load can be increased. Stated differently, controller 28 determines whether additional power can be provided to the secondary load from power bus 15. In one embodiment, controller 28 determines that the power exported to the secondary load can be increased if the available power of the secondary load (e.g., the total amount of power stored by the secondary load or the amount of power currently being generated by the secondary load) is less than or equal to a predetermined available power threshold. If controller 28 determines that the power exported to the secondary load can be increased (e.g., the available power of the secondary load is less than or equal to the predetermined available power threshold), method 84 moves to block 120. Conversely, if controller 28 determines that the power exported to the secondary load cannot be increased (e.g., the available power of the secondary load is greater than the predetermined available power threshold), method 84 moves to block 122.

[0117]

[0125] In block 120, the controller 28 increases the power exported from the power bus 15 to the secondary loads. As a result, excess power on the power bus 15 is diverted to the secondary loads. The energy provided to the secondary loads may later be used to supplement the power provided by the fuel cell power system 16 or may be used by the secondary loads for their own processing. Once a command to increase the power to the secondary loads has been provided, the method 84 proceeds to block 96, where the method 84 is repeated, beginning again at block 86.

[0118]

[0126] At block 122, the controller 28 determines whether the power output from the power bus 15 to the load bank 22 may be increased. In one embodiment, the controller 28 determines that the power output to the load bank 22 may be increased if the power output to the load bank 22 is less than or equal to a predetermined load bank threshold. If the controller 28 determines that the power output to the load bank 22 may be increased (e.g., the power output to the load bank 22 is less than or equal to the load bank threshold), the method 84 moves to block 124. Conversely, if the controller 28 determines that the power output to the load bank 22 may not be increased (e.g., the power output to the load bank 22 is greater than the load bank threshold), the method 84 moves to block 126.

[0119]

[0127] In block 124, controller 28 increases the power output from power bus 15 to load bank 22. As a result, excess power on power bus 15 is diverted to load bank 22. As discussed above, load bank 22 converts the excess power into thermal energy that can be released to the surrounding environment or captured by using CHP equipment for the building's HVAC and other heating loads. Next, method 84 moves to block 96, where method 84 is repeated, beginning again at block 86.

[0120]

[0128] In block 126, controller 28 reduces the power output of fuel cell power system 16 to power bus 15 to reduce the amount of power on power bus 15. Method 84 then moves to block 96 where method 84 is repeated, beginning again at block 86.

[0121]

[0129] As a result of blocks 94, 116, 118, 120, 122, 124, and 126, the fuel cell power system 16 is used as a last resort to reduce excess power on the power bus 15. The power output of the fuel cell power system 16 is reduced if the excess power cannot be reduced by adjusting the power of the load bank 22, the secondary load, the first energy storage system 18, and the second energy storage system 20. Thus, power swings at the output of the fuel cell power system 16 are reduced, and degradation over the life of the fuel cell power system 16 is minimized.

[0122]

[0130] Various embodiments disclosed herein provide system architectures and control methods that use fuel cell-based power generation as a primary energy source. The systems utilize energy storage systems, load banks, and other types of loads to supplement the power output by the fuel cell system and to store any excess power generated by the fuel cell system. As a result of various support modules and techniques, swings in the power output by the fuel cell system can be minimized and the life of the fuel cell power modules can be extended.

[0123]

[0131] The system may be summarized as including a fuel cell power system configured to output a first power signal having a first power level to a load having a power consumption level, and a first energy storage system configured to output a second power signal to the load when the first power level is less than the power consumption level, and to receive a third power signal from the fuel cell power system when the first power level is greater than the power consumption level.

[0124]

[0132] The workload may include a processing system for training an artificial intelligence model.

[0125]

[0133] The system may further include a controller configured to determine a first power level of the first power signal and a power consumption level of the load, and set power levels of the second power signal and the third power signal based on the first power level and the power consumption level.

[0126]

[0134] The system may further include a load bank configured to convert electrical power generated by the fuel cell power system into thermal energy when the first power level exceeds the power consumption level.

[0127]

[0135] The system may further include a converter configured to convert power generated by the fuel cell power system into power for a grid external to the system when the first power level exceeds the power consumption level.

[0128]

[0136] The system may further include a converter configured to convert power generated by the fuel cell power system into power for the load.

[0129]

[0137] The system may further include a second energy storage system having slower discharge and charge times and a larger power capacity than the first energy storage system, the second energy storage system configured to receive a fourth power signal from the fuel cell power system when the first power level exceeds the power consumption level and the first energy storage system is charged.

[0130]

[0138] The fuel cell power system may include a plurality of power modules, each of the plurality of power modules including a hot box.

[0131]

[0139] Each hot box may contain one or more fuel cell stacks.

[0132]

[0140] One or more fuel cell stacks may include solid oxide fuel cells interleaved with conductive interconnects.

[0133]

[0141] The system may be summarized as including a fuel cell power system configured to output a first power signal to a load, and an energy storage system configured to output a second power signal to the load when the power level of the first power signal is below the power consumption of the load, and to store power generated by the fuel cell power system when the power level of the first power signal is above the power consumption of the load.

[0134]

[0142] The load may include a processing system for training one or more artificial intelligence models, the energy storage system may output a second power signal to the load in response to checkpoint processing being performed for the one or more artificial intelligence models, and the energy storage system may store power generated by the fuel cell power system in response to training processing being performed for the one or more artificial intelligence models.

[0135]

[0143] The fuel cell power system may be configured to decrease the power level of the first power signal in response to a decrease in power consumption of the load, and to increase the power level of the first power signal in response to an increase in power consumption of the load.

[0136]

[0144] The energy storage system may output a second power signal when the power level of the first power signal is increased at a first rate and the power consumption of the load is increased at a second rate faster than the first rate, and the power level of the second power signal may be set based on the difference between the power level of the power consumption of the load and the power level of the first power signal.

[0137]

[0145] The energy storage system can store power generated by the fuel cell power system when the power level of the first power signal is reduced at a first rate and the power consumption of the load is reduced at a second rate that is faster than the first rate.

[0138]

[0146] The system may further include a load bank configured to convert electrical power generated by the fuel cell power system into thermal energy when a power level of the first power signal exceeds the power consumption of the load.

[0139]

[0147] The system may further include a converter configured to convert power generated by the fuel cell power system into power for a grid external to the system when the power level of the first power signal exceeds the power consumption of the load.

[0140]

[0148] The first power signal may have a substantially constant power level.

[0141]

[0149] The thermal energy produced by the fuel cell power system can be supplied to an absorption chiller for cooling the processing system.

[0142]

[0150] The thermal energy generated by the load bank or the fuel cell power system, or both the load bank and the fuel cell power system, may be supplied to a steam generator, a thermoelectric generator, a water heater, an absorption chiller, a Rankine cycle device, or a combination thereof.

[0143]

[0151] The method may be summarized as including: determining, by a controller, a total amount of power on a power bus connecting the fuel cell power system, one or more energy storage systems, and the load to each other; detecting, by the controller, a shortage of power on the power bus based on the total amount of power and the power consumption of the load; increasing, by the controller, the power on the power bus in response to detecting the shortage; detecting, by the controller, excess power on the power bus based on the total amount of power and the power consumption of the load; and decreasing, by the controller, the power on the power bus in response to detecting the excess power.

[0144]

[0152] The workload may include a processing system for training an artificial intelligence model.

[0145]

[0153] A load bank configured to convert electrical power to thermal energy may be connected to the power bus, and increasing the power on the power bus may include reducing the power output from the power bus to the load bank.

[0146]

[0154] An external grid may be connected to the power bus, and increasing the power on the power bus may include reducing the power output from the power bus to the external grid.

[0147]

[0155] Increasing the power on the power bus may include increasing the power output from one or more energy storage systems onto the power bus.

[0148]

[0156] An external grid may be connected to the power bus, and increasing the power on the power bus may include increasing the power output from the external grid to the power bus.

[0149]

[0157] Increasing the power on the power bus may include increasing the power output from the fuel cell power system onto the power bus.

[0150]

[0158] Reducing the power on the power bus may include increasing the power output from the power bus to one or more energy storage systems.

[0151]

[0159] An external grid may be connected to the power bus, and reducing power on the power bus may include increasing power output from the power bus to the external grid.

[0152]

[0160] A load bank configured to convert electrical power into thermal energy may be connected to the power bus, and reducing the power on the power bus may include increasing the power output from the power bus to the load bank.

[0153]

[0161] Reducing the power on the power bus may include reducing the power output from the fuel cell power system onto the power bus.

[0154]

[0162] The various embodiments described above can be combined to provide further embodiments. These and other modifications can be made to the above-described embodiments in light of the above detailed description. Generally, in the following claims, the terms used should not be construed to limit the claims to the specific embodiments disclosed in the specification and the claims, but rather to include all possible embodiments, along with the full range of equivalents to which such claims are entitled. Accordingly, the claims are not limited by the present disclosure. [Explanation of symbols]

[0155] 10 Power Signal 12 Power Signal 14 Systems 15 Power Bus 16 Fuel Cell Power System 18 First Energy Storage System 20 Secondary Energy Storage System 22 Load Bank 24 Power Converter 26 Load Interface Converter 28 Controller 30 Fuel Cell Power Module 32 Power Conditioning Module 34 Pad 36 Hot Box 38 Access Door 40 Power Signal 42 Power Signal 44 Power Signal 46 Power Signal 48 Discharge area 50 charging area 52 Power Signal 54 Power Signal 56 Power Signal 58 Power Signal 60 Power Signal 62 Charging area 64 consumption area 66 Discharge area 68 Main load power consumption 70 Power Signal 72 Power Signal 74 Power Signal 76 Power Signal 78 charging area 80 consumption area 82 Discharge area 84 method 86, 88, 90, 92, 94, 96, 98, 100, 102, 104, 106, 108, 110, 112, 114, 116, 118, 120, 122, 124, 126 blocks FP H High fuel cell electricity prices FP L Low fuel cell power value LP H High load power LP H1 High load power LP H2 High load power LP L Low load power LP L1 Low load power LP L2 Low load power LP SB Load power standby value LP SCR Load power software recovery value T1 time T2 time T3 time t CP Checkpoint Period t CP1 Checkpoint Period t CP2 Checkpoint Period t SCR Software crash recovery period t SB Standby Period t T Training period t T1 Training period t T2 Training period

Claims

1. a fuel cell power system configured to output a first power signal having a first power level to a load having a power consumption level; 1. A first energy storage system, comprising: outputting a second power signal to the load if the first power level is less than the power consumption level; receiving a third power signal from the fuel cell power system if the first power level exceeds the power consumption level; a first energy storage system configured to: A system including:

2. The system of claim 1 , wherein the workload comprises a processing system for training an artificial intelligence model.

3. determining the first power level of the first power signal and the power consumption level of the load; setting power levels of the second power signal and the third power signal based on the first power level and the power consumption level; The system of claim 1 , further comprising a controller configured to:

4. The system of claim 1 , further comprising a load bank configured to convert electrical power generated by the fuel cell power system into thermal energy when the first power level exceeds the power consumption level.

5. 10. The system of claim 1, further comprising a converter configured to convert power generated by the fuel cell power system into power for a grid external to the system when the first power level exceeds the power consumption level.

6. The system of claim 1 further comprising a converter configured to convert power generated by the fuel cell power system into power for the load.

7. 10. The system of claim 1, further comprising a second energy storage system having slower discharge and charge times and a larger power capacity than the first energy storage system, the second energy storage system configured to receive a fourth power signal from the fuel cell power system when the first power level exceeds the power consumption level and the first energy storage system is being charged.

8. The system of claim 1 , wherein the fuel cell power system includes a plurality of power modules, each of the plurality of power modules including a hot box.

9. The system of claim 8 , wherein each hot box includes one or more fuel cell stacks.

10. 10. The system of claim 9, wherein the one or more fuel cell stacks include solid oxide fuel cells interleaved with electrically conductive interconnects.

11. a fuel cell power system configured to output a first power signal to a load; 1. An energy storage system comprising: outputting a second power signal to the load when a power level of the first power signal is below a power consumption of the load; storing power generated by the fuel cell power system when the power level of the first power signal exceeds the power consumption of the load; an energy storage system configured to: A system including:

12. the workload includes a processing system for training one or more artificial intelligence models; the energy storage system outputs the second power signal to the load in response to a checkpoint being performed for the one or more artificial intelligence models; 12. The system of claim 11, wherein the energy storage system stores the power generated by the fuel cell power system in response to a training process being performed for the one or more artificial intelligence models.

13. The fuel cell power system includes: decreasing the power level of the first power signal in response to the decrease in the power consumption of the load; increasing the power level of the first power signal in response to an increase in the power consumption of the load; The system of claim 11 configured to:

14. the energy storage system outputs the second power signal when the power level of the first power signal is increased at a first rate and the power consumption of the load is increased at a second rate that is faster than the first rate; The system of claim 13 , wherein a power level of the second power signal is set based on a difference between a power level of the power consumption of the load and a power level of the first power signal.

15. 14. The system of claim 13, wherein the energy storage system stores the power generated by the fuel cell power system when the power level of the first power signal is decreased at a first rate and the power consumption of the load is decreased at a second rate that is faster than the first rate.

16. 12. The system of claim 11, further comprising a load bank configured to convert electrical power generated by the fuel cell power system into thermal energy when the power level of the first power signal exceeds the power consumption of the load.

17. 12. The system of claim 11, further comprising a converter configured to convert power generated by the fuel cell power system into power for a grid external to the system when the power level of the first power signal exceeds the power consumption of the load.

18. The system of claim 11 , wherein the first power signal has a substantially constant power level.

19. The system of claim 12 , wherein thermal energy generated by the fuel cell power system is supplied to an absorption chiller for cooling a processing system.

20. 17. The system of claim 16, wherein thermal energy generated by the load bank, the fuel cell power system, or both the load bank and the fuel cell power system is supplied to a steam generator, a thermoelectric generator, a water heater, an absorption chiller, a Rankine cycle device, or a combination thereof.

21. determining, by a controller, a total amount of power on a power bus interconnecting the fuel cell power system, the one or more energy storage systems, and the load; detecting, by the controller, a power deficiency on the power bus based on the total power amount and the power consumption of the load; increasing, by the controller, power on the power bus in response to detecting the shortage; detecting, by the controller, excess power on the power bus based on the total power amount and the power consumption of the load; reducing power on the power bus in response to detecting the excess power by the controller. A method comprising:

22. 22. The method of claim 21, wherein the workload comprises a processing system for training an artificial intelligence model.

23. a load bank configured to convert electrical power into thermal energy connected to the electrical power bus; and 22. The method of claim 21, wherein the increasing the power on the power bus comprises reducing power output from the power bus to the load bank.

24. an external grid connected to the power bus; 22. The method of claim 21, wherein the increasing the power on the power bus comprises reducing power output from the power bus to the external grid.

25. 22. The method of claim 21, wherein the increasing the power on the power bus comprises increasing power output from the one or more energy storage systems to the power bus.

26. an external grid connected to the power bus; 22. The method of claim 21, wherein the increasing the power on the power bus comprises increasing power output from the external grid to the power bus.

27. 22. The method of claim 21, wherein the increasing the power on the power bus comprises increasing the power output from the fuel cell power system onto the power bus.

28. 22. The method of claim 21, wherein the decreasing the power on the power bus comprises increasing the power output from the power bus to the one or more energy storage systems.

29. an external grid connected to the power bus; 22. The method of claim 21, wherein the decreasing the power on the power bus comprises increasing the power output from the power bus to the external grid.

30. a load bank configured to convert electrical power into thermal energy, connected to the electrical power bus; 22. The method of claim 21, wherein the decreasing the power on the power bus comprises increasing the power output from the power bus to the load bank.

31. 22. The method of claim 21, wherein the reduction of the power on the power bus comprises reducing the power output from the fuel cell power system to the power bus.

Images