Method and device for controlling an industrial gas plant complex

Abstract

A method for controlling an industrial gas equipment complex is provided, wherein the industrial gas equipment complex includes multiple industrial gas equipment powered by one or more renewable power sources, and the method is executed by at least one hardware processor, and the method includes: receiving time-related predicted power data for a predetermined future time period from the one or more renewable power sources; receiving time-related predicted operating characteristic data for each industrial gas equipment; using the predicted power data and predicted characteristic data in an optimization model to generate a set of state variables for the multiple industrial gas equipment; using the generated state variables to generate a set of control set points for the multiple industrial gas equipment; and sending the control set points to a control system to control the industrial gas equipment complex by adjusting one or more control set points of the industrial gas equipment.

Description

Technical Field

[0001] The present invention relates to a method and a system for controlling an industrial gas plant complex comprising a plurality of industrial gas plants powered by one or more renewable power sources. Background Art

[0002] An industrial gas plant complex may include one or more process plants that produce gases or participate in the production of gases. In non-limiting examples, these gases may include: industrial gases, commercial gases, medical gases, inorganic gases, organic gases, fuel gases and green fuel gases in gaseous, liquefied or compressed form.

[0003] There is considerable interest in methods and systems for utilizing renewable energy sources for powering industrial gas plants and industrial gas plant complexes. However, a significant disadvantage of using renewable energy sources such as wind, solar, and tidal energy is the natural variability and instantaneous nature of such energy sources.

[0004] Typically, a constant or substantially constant power supply is preferred for an industrial gas plant or industrial gas plant complex. The variable and intermittent nature of wind, solar and / or tidal energy is therefore problematic and makes it difficult to ensure maximum utilization of an industrial gas plant or industrial gas plant complex utilizing such power sources.

[0005] An exemplary industrial gas is ammonia. Ammonia is produced using hydrogen from water electrolysis and nitrogen separated from air. These gases are then fed into the Haber-Bosch process, where hydrogen and nitrogen react together at high temperatures and pressures to produce ammonia.

[0006] There is considerable interest in producing ammonia using renewable energy sources. This is known as green ammonia. However, ammonia synthesis can be particularly sensitive to variations in input energy from renewable sources. Solutions to these technical problems are therefore needed to enable efficient production of industrial gases in such an arrangement. Summary of the invention

[0007] Some concepts are introduced in simplified form below to provide a basic understanding of some aspects of the present disclosure. The following is not a broad overview of the present disclosure and is not intended to identify the key or important elements of the present disclosure or to describe the scope of the present disclosure. The following only summarizes some concepts of the present disclosure as a preface to the more detailed description provided thereafter.

[0008] According to a first aspect, a method is provided for determining and utilizing predicted available power resources from one or more renewable power sources for one or more industrial gas devices including one or more storage resources, the method being executed by at least one hardware processor, the method comprising: obtaining historical time-related environmental data associated with one or more renewable power sources; obtaining historical time-related operating characteristic data associated with the one or more renewable power sources; training a machine learning model based on the historical time-related environmental data and the historical time-related operating characteristic data; executing the trained machine learning model to predict the available power resources of one or more industrial gas devices within a predetermined future time period; and controlling the one or more industrial gas devices in response to the predicted available power resources for the predetermined future time period.

[0009] In an embodiment, controlling the one or more industrial gas devices includes maximizing usage of a predicted available power resource for a predetermined future time period.

[0010] In an embodiment, the storage resource comprises one or more industrial gas storage vessels and / or one or more energy storage resources.

[0011] In an embodiment, the one or more energy storage resources include one or more of: a battery energy storage system; a compressed air energy storage; a liquid air energy storage; or a pumped hydroelectric energy storage.

[0012] In an embodiment, maximizing usage of the predicted power resource further comprises controlling utilization of the industrial gas storage vessel and / or one or more energy storage resources in response to the predicted available power resource.

[0013] In an embodiment, controlling utilization includes utilizing an algorithm to select one or more storage resources from a set of storage resources for a given pattern of predicted power availability as a function of time.

[0014] In an embodiment, the selection of a storage resource is based on physical characteristics of the storage resource.

[0015] In an embodiment, the one or more renewable power sources include one or more of: a solar power source; a wind power source; a tidal power source; a hydroelectric power source; or a geothermal power source.

[0016] In an embodiment, the environmental data is selected from one or more of: wind speed; cloud cover; precipitation; humidity; air temperature; atmospheric pressure; solar intensity; and tidal time.

[0017] In an embodiment, the operational characteristic data includes power output from one or more renewable power sources.

[0018] In an embodiment, the step of training the machine learning model is performed periodically at a predetermined training time.

[0019] In an embodiment, at training time, a machine learning model is trained based on historical time-related environmental data and historical time-related operating characteristic data obtained within one or more predetermined historical time windows.

[0020] In an embodiment, the method further comprises comparing the value of the predicted power resource for a predetermined future time period with the actual power resource at the end of the predicted time period to generate a prediction error value.

[0021] In an embodiment, when the prediction error value exceeds a predetermined threshold, a predetermined training time is selected.

[0022] In an embodiment, the predetermined training time is selected based on a predetermined experience interval unless the prediction error value exceeds a predetermined threshold within the predetermined experience interval.

[0023] In an embodiment, the one or more industrial gas facilities include a hydrogen production facility having at least one electrolyzer.

[0024] In an embodiment, the one or more industrial gas plants include an ammonia production plant complex that includes a hydrogen production plant.

[0025] In an embodiment, the machine learning model includes one or more of: a gradient boosting algorithm; a long short-term memory (LSTM) algorithm; a support vector machine (SVM) algorithm; or a random decision forest algorithm.

[0026] According to a second aspect, a system is provided for determining and utilizing predicted available power resources from one or more renewable power sources for one or more industrial gas devices including one or more storage resources, the system comprising: at least one hardware processor operable to perform: obtaining historical time-related environmental data associated with one or more renewable power sources; obtaining historical time-related operating characteristic data associated with the one or more renewable power sources; training a machine learning model based on the historical time-related environmental data and the historical time-related operating characteristic data; executing the trained machine learning model to predict the available power resources of one or more industrial gas devices within a predetermined future time period; and controlling the one or more industrial gas devices in response to the predicted available power resources for the predetermined future time period.

[0027] According to a third aspect, a computer-readable storage medium storing a program of instructions is provided, wherein the program of instructions can be executed by a machine to perform a method for determining and utilizing predicted available power resources from one or more renewable power sources for one or more industrial gas equipment including one or more storage resources, the method comprising: obtaining historical time-related environmental data associated with one or more renewable power sources; obtaining historical time-related operating characteristic data associated with the one or more renewable power sources; training a machine learning model based on the historical time-related environmental data and the historical time-related operating characteristic data; executing the trained machine learning model to predict the available power resources of the industrial gas equipment within a predetermined future time period; and controlling the one or more industrial gas equipment in response to the predicted available power resources for the predetermined future time period.

[0028] According to a fourth aspect, a method for monitoring the operating characteristics of an industrial gas equipment complex including multiple industrial gas equipment is provided, the method being executed by at least one hardware processor, the method comprising: assigning a machine learning model to each industrial gas equipment forming the industrial gas equipment complex; training a corresponding machine learning model for each industrial gas equipment based on the received historical time-related operating characteristic data of the corresponding industrial gas equipment; executing the trained machine learning model on each industrial gas equipment to predict the operating characteristics of each corresponding industrial gas equipment in a predetermined future time period; and comparing the predicted operating characteristic data of each corresponding industrial gas equipment in the predetermined future time period with the measured operating characteristic data in the corresponding time period to identify deviations in the performance of the industrial gas equipment.

[0029] In an embodiment, the step of comparing is performed at a timestamp at the end of or within a predetermined future time period of the predicted operating characteristic data.

[0030] In an embodiment, the step of comparing comprises comparing predicted operating characteristic data predicted for a predetermined time window with actual measured operating characteristic data for the same time window.

[0031] In an embodiment, the received historical time-dependent operating characteristic data of the respective industrial gas plant includes data obtained from direct measurements of a process or parameter of the respective industrial gas plant.

[0032] In an embodiment, the received historical time-dependent operational characteristic data of the respective industrial gas plant comprises data obtained from a physics-based model representing operational characteristics of the respective industrial gas plant.

[0033] In an embodiment, measurement data related to a process or parameter of a respective industrial gas plant is input into a respective physics-based model.

[0034] In an embodiment, predicted operating characteristics of each industrial gas device are utilized to determine predicted future resources, future failures, and / or predicted future maintenance.

[0035] In an embodiment, the one or more industrial gas facilities include a hydrogen processing facility having a plurality of electrolyzer modules.

[0036] In an embodiment, a machine learning model is assigned to each electrolyser module.

[0037] In an embodiment, the predicted operating characteristics of each respective industrial gas plant are utilized in a further model to generate an operational performance metric for the industrial gas plant complex.

[0038] In an embodiment, the operational performance metric comprises an efficiency value of the industrial gas plant complex.

[0039] In an embodiment, the industrial gas plant complex comprises an ammonia plant complex, and the determined efficiency value enables a predictive determination of ammonia produced for a given level of energy input.

[0040] According to a fifth aspect, a system for monitoring the operating characteristics of an industrial gas equipment complex including multiple industrial gas equipment is provided, the system comprising at least one hardware processor, which is operable to perform the assignment of a machine learning model to each industrial gas equipment forming the industrial gas equipment complex; training a corresponding machine learning model for each industrial gas equipment based on the received historical time-related operating characteristic data of the corresponding industrial gas equipment; executing the trained machine learning model on each industrial gas equipment to predict the operating characteristics of each corresponding industrial gas equipment in a predetermined future time period; and comparing the predicted operating characteristic data of each corresponding industrial gas equipment in the predetermined future time period with the measured operating characteristic data in the corresponding time period to identify deviations in the performance of the industrial gas equipment.

[0041] In an embodiment, the step of comparing is performed at a timestamp at the end of or within a predetermined future time period of the predicted operating characteristic data.

[0042] In an embodiment, the step of comparing comprises comparing predicted operating characteristic data predicted for a predetermined time window with actual measured operating characteristic data for the same time window.

[0043] In an embodiment, the received historical time-dependent operating characteristic data of the respective industrial gas plant includes data obtained from direct measurements of a process or parameter of the respective industrial gas plant.

[0044] In an embodiment, the received historical time-dependent operational characteristic data of the respective industrial gas plant comprises data obtained from a physics-based model representing operational characteristics of the respective industrial gas plant.

[0045] In an embodiment, predicted operating characteristics of each industrial gas device are utilized to determine predicted future resources, future failures, and / or predicted future maintenance.

[0046] In an embodiment, the predicted operating characteristics of each respective industrial gas plant are utilized in a further model to generate an operational performance metric for the industrial gas plant complex.

[0047] According to the sixth aspect, a computer-readable storage medium storing a program of instructions is provided, wherein the program of instructions can be executed by a machine to perform a method for monitoring the operating characteristics of an industrial gas equipment complex including multiple industrial gas equipment, wherein the method is executed by at least one hardware processor, and the method includes: assigning a machine learning model to each industrial gas equipment forming the industrial gas equipment complex; training a corresponding machine learning model for each industrial gas equipment based on the received historical time-related operating characteristic data of the corresponding industrial gas equipment; executing the trained machine learning model on each industrial gas equipment to predict the operating characteristics of each corresponding industrial gas equipment in a predetermined future time period; and comparing the predicted operating characteristic data of each corresponding industrial gas equipment in the predetermined future time period with the measured operating characteristic data in the corresponding time period to identify deviations in the performance of the industrial gas equipment.

[0048] According to the seventh aspect, a method for controlling an industrial gas equipment complex is provided, wherein the industrial gas equipment complex includes a plurality of industrial gas equipment powered by one or more renewable power sources, the method being executed by at least one hardware processor, the method comprising: receiving time-related predicted power data for a predetermined future time period from the one or more renewable power sources; receiving time-related predicted operating characteristic data for each industrial gas equipment; generating a set of state variables for the plurality of industrial gas equipment using the predicted power data and predicted characteristic data in an optimization model; generating a set of control set points for the plurality of industrial gas equipment using the generated state variables; and sending the control set points to a control system to control the industrial gas equipment complex by adjusting one or more control set points of the industrial gas equipment.

[0049] In an embodiment, the optimization model defines the predicted power data and the predicted characteristic data as a set of non-linear equations.

[0050] In an embodiment, the state variables are generated by solving a set of non-linear equations.

[0051] In an embodiment, time-dependent predicted power data is generated from a trained machine learning model.

[0052] In an embodiment, time-dependent predictive power data is obtained by: obtaining historical time-dependent environmental data associated with one or more renewable power sources; obtaining historical time-dependent operational characteristic data associated with the one or more renewable power sources; training a machine learning model based on the historical time-dependent environmental data and the historical time-dependent operational characteristic data; and executing the trained machine learning model to predict available power resources of the one or more industrial gas equipment within a predetermined future time period.

[0053] In an embodiment, time-dependent predicted operating characteristic data is generated from a trained machine learning model for each industrial gas facility.

[0054] In an embodiment, time-related predicted operating characteristic data for each industrial device is obtained by: assigning a machine learning model to each industrial gas device forming an industrial gas device complex; training a corresponding machine learning model for each industrial gas device based on the received historical time-related operating characteristic data of the corresponding industrial gas device; and executing the trained machine learning model for each industrial gas device to predict the operating characteristics of each corresponding industrial gas device within a predetermined future time period.

[0055] In an embodiment, the industrial gas facility complex includes storage resources including one or more industrial gas storage vessels and / or one or more energy storage resources.

[0056] In an embodiment, the one or more energy storage resources include one or more of: a battery energy storage system; a compressed air energy storage; a liquid air energy storage; or a pumped hydroelectric energy storage.

[0057] In an embodiment, the predicted power data further includes data representative of operating parameters of the storage resource.

[0058] In an embodiment, the data representing an operational parameter of a storage resource includes one or more of: resource memory availability; fill level; and utilization.

[0059] According to an eighth aspect, a system for controlling an industrial gas equipment complex including multiple industrial gas equipment powered by one or more renewable power sources is provided, the system comprising: at least one hardware processor operable to perform: receiving time-related predicted power data for a predetermined future time period from the one or more renewable power sources; receiving time-related predicted operating characteristic data for each industrial gas equipment; generating a set of state variables for the multiple industrial gas equipment using the predicted power data and predicted characteristic data in an optimization model; generating a set of control set points for the multiple industrial gas equipment using the generated state variables; and sending the control set points to a control system to control the industrial gas equipment complex by adjusting one or more control set points of the industrial gas equipment.

[0060] In an embodiment, the optimization model defines the predicted power data and the predicted characteristic data as a set of non-linear equations.

[0061] In an embodiment, the state variables are generated by solving a set of non-linear equations.

[0062] In an embodiment, time-dependent predicted power data is generated from a trained machine learning model.

[0063] In an embodiment, time-dependent predicted operating characteristic data is generated from a trained machine learning model for each industrial gas facility.

[0064] According to the ninth aspect, a computer-readable storage medium storing a program of instructions is provided, wherein the program of instructions can be executed by a machine to perform a method for controlling an industrial gas equipment complex including multiple industrial gas equipment powered by one or more renewable power sources, wherein the method is executed by at least one hardware processor, and the method includes: receiving time-related predicted power data for a predetermined future time period from the one or more renewable power sources; receiving time-related predicted operating characteristic data for each industrial gas equipment; generating a set of state variables for multiple industrial gas equipment using the predicted power data and predicted characteristic data in an optimization model; generating a set of control set points for the multiple industrial gas equipment using the generated state variables; and sending the control set points to a control system to control the industrial gas equipment complex by adjusting one or more control set points of the industrial gas equipment.

[0065] In an embodiment, the optimization model defines the predicted power data and the predicted characteristic data as a set of non-linear equations.

[0066] In an embodiment, the state variables are generated by solving a set of non-linear equations.

[0067] In an embodiment, time-dependent predicted power data is generated from a trained machine learning model, and / or time-dependent predicted operating characteristic data is generated from a trained machine learning model for each industrial gas equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Embodiments of the present invention will now be described, by way of example only, with reference to the accompanying drawings, in which:

[0069] Figure 1 It is a schematic diagram of the industrial gas equipment complex and control system;

[0070] Figure 2 yes Figure 1 Detailed schematic diagram of the control system;

[0071] Figure 3 is a graph showing measured and predicted wind power values ​​using a model according to an embodiment of the present invention;

[0072] Figure 4 is a graph showing measured and predicted solar power values ​​using a model according to an embodiment of the present invention;

[0073] Figure 5 is a flow chart of a method according to an embodiment;

[0074] Figure 6 is a flow chart of a method according to an embodiment;

[0075] Figure 7 is a diagram showing an efficiency model of process variables;

[0076] Figure 8 is a graph of the optimum ammonia plant rate as a function of time over a 48 hour forecast period; and

[0077] Fig. 9 is a flow chart of a method according to an embodiment.

[0078] Embodiments of the present disclosure and their advantages may be best understood by referring to the following detailed description. It should be understood that the same reference numerals are used to identify the same elements shown in one or more of the accompanying drawings, wherein the illustrations are for illustrating embodiments of the present disclosure, rather than for limiting embodiments of the present disclosure. DETAILED DESCRIPTION

[0079] Various examples and embodiments of the present disclosure will now be described. The following description provides specific details for a comprehensive understanding and ability to describe these examples. However, it will be appreciated by those skilled in the relevant art that one or more embodiments described herein may be implemented without many of these details. Similarly, it will be appreciated by those skilled in the relevant art that one or more embodiments of the present disclosure may include other features and / or functions that are not described in detail herein. In addition, some well-known structures or functions may not be shown or described in detail below to avoid unnecessarily obscuring the relevant description.

[0080] Figure 1 A schematic diagram of an industrial gas plant complex 10 and a control system 100 is shown.

[0081] The industrial gas plant complex 10 comprises a hydrogen production plant 12, a hydrogen storage unit 14, an air separation unit (ASU) 16, an ammonia synthesis plant 18 and an ammonia storage unit 20. The ammonia storage unit 20 is connected to an external supply chain 22 for onward distribution of ammonia.

[0082] The electricity used to power the industrial gas plant complex 10 is at least partially generated by renewable energy sources (e.g., wind 24 and / or solar energy 26), although other sources such as diesel, gasoline or hydrogen powered power generators (not shown) or a national grid (not shown) may optionally be utilized.

[0083] To address the intermittent nature of the power supply from renewable sources, a storage resource 28 is provided. The storage resource 28 may include one or more resource storage devices or energy storage devices. For example, one or more resource storage devices may include a hydrogen storage unit 14. The production of hydrogen by electrolysis requires a large amount of power, and using stored hydrogen as a hydrogen source for the production of ammonia can significantly reduce the power consumption of the device 10 during periods of low renewable power supply. In addition, a liquid nitrogen storage 16a may also be provided as part of the storage resource 28, such as Figure 1 shown.

[0084] Additionally or alternatively, in a non-exhaustive arrangement, the energy storage device may include one or more of: a battery energy storage system (BESS) 28a, a compressed / liquid air energy system (CAES or LAES) 28b, or a pumped hydraulic storage system (PHSS) 28c.

[0085] The BESS 28a utilizes electrochemical technology and may include one or more of the following: lithium-ion batteries, lead-acid batteries, zinc bromide, sodium-sulfur, or redox flow batteries. Electrochemical devices such as batteries have advantages in terms of fast charging rates and fast (almost instantaneous) ramp rates to supply power to respond to sudden drops in energy supply. However, their power capacity tends to be more limited than other systems. Therefore, they may be more suitable for use in situations where, for example, power shortages from renewable sources are expected to be temporary or of short duration.

[0086] The CAES 28b compresses the air and stores it at a high pressure of about 70 bar. It is usually stored in underground caverns. When power is needed, the compressed air is heated and expanded in an expansion turbine in order to drive a power generator.

[0087] The LAES 28b includes an air liquefier to draw air from the environment and compress and cool the air to achieve liquefaction. The liquefied air is then stored in insulated tanks until power is needed. To convert the liquefied air into usable energy, the liquid air is pumped to high pressure and heated through a heat exchanger. The resulting high-pressure gas is used to drive a turbine to generate electricity.

[0088] CAES and LAES are capable of storing significantly more energy than most BESS 28a systems. However, CAES and LAES have slower ramp rates than electrochemical storage devices and require longer to store larger amounts of energy. For example, it may take about 5-10 minutes for the compression stage to operate at full load, and 10-20 minutes to generate full power on demand. Therefore, such storage devices are more suitable for long-term storage and are more suitable for supplying power during long-term shortages of renewable energy.

[0089] The PHSS 28c stores energy in the form of gravitational potential energy by pumping water from a lower altitude reservoir to a higher altitude reservoir. When power is needed, the water is released to drive a turbine. Some PHSS installations utilize a reversible pump-turbine unit.

[0090] Considering the large storage capacity of PHSS configurations, they are generally suitable for long-term storage. Moreover, especially for reversible pump-turbines, time scales of about 5-10 minutes from shutdown to full-load power generation, about 5-30 minutes from shutdown to pumping, and about 10-40 minutes from pumping to load generation, or vice versa, are common. Therefore, this type of storage seems to be more suitable for long-term power shortages.

[0091] Although in Figure 1All of these elements are shown in FIG. 2 , but this is for illustration purposes only. The energy storage resource 20 need not include each of the described elements, and may include only one or more of the described elements. Furthermore, the energy resource 28 may include additional elements.

[0092] The components of the industrial gas plant complex 10 will now be described in detail.

[0093] Hydrogen production equipment12

[0094] The hydrogen production facility 12 is operable to electrolyze water to form hydrogen and oxygen. Any suitable water source may be used. However, in embodiments where seawater is used to produce water for electrolysis, the facility will further include at least one desalination and demineralization facility for treating the seawater.

[0095] The hydrogen production facility 12 includes a plurality of electrolysis units 12a, 12b ... 12n or electrolysis cells. Each unit or cell may be referred to as an "electrolysis cell" 12a, 12b ... 12n.

[0096] The electrolyzers may enable the hydrogen production facility 12 to have a total capacity of at least 1 GW. However, the ultimate capacity of the hydrogen production facility 12 is limited only by practical considerations such as power supply.

[0097] Any suitable type of electrolyser may be used. In an embodiment, multiple electrolysers are typically composed of multiple individual cells grouped into "modules" that also contain process equipment such as pumps, coolers, and / or separators. Hundreds of cells may be used and may be grouped in different buildings. Each module typically has a maximum capacity greater than 10 MW, although this is not intended to be limiting.

[0098] Any suitable type of electrolyzer may be used in the present invention. Typically, three conventional types of electrolyzers are used - alkaline electrolyzers; PEM electrolyzers; and solid oxide electrolyzers. Any of these types may be used in the present invention.

[0099] Alkaline electrolyzers convert hydroxide ions (OH - ) is transported from the cathode to the anode, where hydrogen is generated on the cathode side. Typically, a liquid alkaline solution of sodium hydroxide or potassium hydroxide is used as the electrolyte.

[0100] The PEM electrolyzer utilizes a solid plastic material as an electrolyte, and water reacts at the anode to form oxygen and positively charged hydrogen ions. Electrons flow through an external circuit, and the hydrogen ions selectively move through the PEM to the cathode. At the cathode, the hydrogen ions combine with electrons from the external circuit to form hydrogen gas.

[0101] Solid oxide electrolyzers use solid ceramic materials as electrolytes that selectively conduct negatively charged oxygen ions (O 2- ). Water at the cathode combines with electrons from the external circuit to form hydrogen gas and negatively charged oxygen ions. The oxygen ions pass through the solid ceramic membrane and react at the anode to form oxygen gas and generate electrons for the external circuit.

[0102] The electrolytic cells may be arranged in any suitable groupings. For example, they may be arranged in parallel.

[0103] Hydrogen is produced at about atmospheric pressure by the hydrogen production plant 12. The hydrogen stream thus produced is removed from the electrolyser at a slightly elevated pressure and can be conveyed to the ammonia synthesis plant 18 via a pipeline.

[0104] Alternatively, any hydrogen in excess of demand may be stored in a hydrogen storage unit 14. The storage unit 14 includes multiple short-term and long-term storage options with different sizes, fill / discharge rates, and round-trip efficiencies. A typical storage system may include pressure vessels and / or pipeline segments connected to a common inlet / outlet header. For example, the pressure vessels may be spheres of about 25 m in diameter, or "pellets," which are horizontal containers with a large L / D ratio (typically up to about 12:1) and having a diameter of up to about 12 m. In some areas, underground caverns are included as storage systems to smooth out seasonal variations associated with renewable power.

[0105] Preferably, the hydrogen is compressed by a compressor and stored under pressure in a hydrogen storage unit 14 to reduce volume requirements. At this point, it can be used commercially (e.g., sold for automotive use) or can be used as a reservoir for the ammonia synthesis plant 18 via pipeline 30.

[0106] Optionally, a purification system may be implemented to purify or dry the hydrogen prior to further use. For example, the hydrogen may be dried in an adsorption unit, such as a temperature swing adsorption (TSA) unit for use in a downstream process.

[0107] Air separation unit 16

[0108] In a non-limiting embodiment, the nitrogen required for ammonia production is produced by cryogenically distilling air in an air separation unit (ASU) 16. Typically, the ASU 16 operates at a pressure of about 10 bar. The pressure is then reduced to provide a nitrogen stream in one or more pipelines arranged to deliver the nitrogen to the ammonia synthesis plant 16. However, other nitrogen sources may be used if desired, such as nitrogen storage 16a.

[0109] A nitrogen storage unit 16a may also be provided which may be used as a resource storage as described below. The storage unit 16a may, together with the hydrogen storage unit 14, include a number of short-term and long-term storage options having different sizes, fill / discharge rates, and round-trip efficiencies.

[0110] A typical storage system for nitrogen may include multiple pressure vessels and / or pipe segments connected to a common inlet / outlet header. For example, the pressure vessels may be spheres of about 25 m in diameter, or "shots," which are horizontal vessels with large L / D ratios (typically up to about 12:1) and having diameters of up to about 12 m. In certain regions, underground caverns may be used to smooth out seasonal variations associated with renewable power.

[0111] Preferably, the nitrogen is compressed by a compressor and stored under pressure in a nitrogen storage unit 16a to reduce volume requirements. It can be used as a reservoir for the ammonia synthesis plant 18, which can be fed through a connecting pipeline.

[0112] Ammonia synthesis equipment18

[0113] The ammonia synthesis plant 18 operates in the Haber-Bosch process and comprises an ammonia loop. The ammonia loop is a single unit equilibrium reaction system that processes synthesis gas of nitrogen and hydrogen to produce ammonia.

[0114] Nitrogen is provided by one or more pipelines from ASU 16, in embodiments, the air separation unit can be operated continuously to provide nitrogen. Hydrogen is provided by one or more pipelines from hydrogen production facility 12 (if it is operated based on the availability of renewable power in a given situation). Otherwise hydrogen is supplied from hydrogen storage 14.

[0115] The stoichiometric composition of the syngas is processed by a system of syngas compressors, and the resulting ammonia product is refrigerated by another set of compressors and sent to storage. The performance of the ammonia loop is determined by the equilibrium conversion of the exothermic reaction. The parameters used for this will be discussed below.

[0116] Power generation and management systems

[0117] The power for the device 10 as a whole may be generated by any suitable energy source, including renewable or non-renewable energy sources. Figure 1 As shown, the electricity is generated by at least one renewable energy source of wind energy 24 (via a suitable wind power generation field including a plurality of wind turbines) and / or solar energy 26 (via a solar power generation field including a plurality of solar cells). In addition, other renewable energy sources may be used, such as a hydroelectric power source (not shown) and / or a tidal power source (not shown).

[0118] Additionally, power or resources for device 10 as a whole or for sub-devices of device 10 may be drawn from energy storage resource 28. Figure 1 As mentioned above, the energy storage resource 28 may include one or more storage resources. For example, the one or more storage resources may include a hydrogen storage unit 14 and a nitrogen storage 16a.

[0119] Additionally or alternatively, in a non-exhaustive arrangement, the energy storage device may include one or more of: a battery energy storage system (BESS) 28a, a compressed / liquid air energy system (CAES or LAES) 28b, or a pumped hydraulic storage system (PHSS) 28c.

[0120] These elements are optimally used to store additional resources and / or energy when electricity supply from renewable sources is high or predicted to be high, and then utilize these resources and / or energy when renewable electricity resources are predicted to be low.

[0121] The prediction and control of these facilities will be described below. It is very important to select these facilities under optimal conditions so that, for example, the correct energy source is selected for a specific predicted power shortage period.

[0122] Control System 100

[0123] Figure 1 The control system 100 in Figure 2 It is shown in detail in the schematic diagram.

[0124] The control system 100 includes three main categories: plant complex control systems 110, renewable energy control systems 120, and optimization systems 150. These are non-limiting terms and do not necessarily imply any interconnection or grouping between the components of the systems 110, 120, 150, and are shown in common groupings only for the purpose of clarity.

[0125] The plant complex control system 110 includes a hydrogen production plant control system 112 , a hydrogen storage control system 114 , an ASU control system 116 , and an ammonia synthesis plant control system 118 .

[0126] As an example, the hydrogen production facility control system 112 may be configured to monitor the amount and rate of hydrogen generated from electrolysis by measuring such measurements. Such measurements may be derived from sensor measurements such as direct flow measurements, or alternatively inferred by indirect measurements such as electrolyzer current or power demand.

[0127] As another example, for the hydrogen storage control system 114 , the pressure and flow of compressed hydrogen from the electrolyzer and compression system to the storage system, as well as the pressure and flow of compressed hydrogen to the ammonia synthesis plant 18 may be monitored.

[0128] In each case, the control system 110 is operable to control parameters of the respective industrial gas plant and is capable of outputting usage and process data from each industrial gas plant. This will be described in detail below.

[0129] The renewable energy control system 120 in the described embodiment includes a wind control system 124 and a solar control system 126. These control systems control and monitor process parameters of renewable energy sources, such as energy generation, storage, and load. They are also configured to send usage, power, and process data to external systems as needed. If a hydroelectric or tidal renewable power source is used, a similar control system will apply.

[0130] The optimization system 150 includes a computer system including three modules: a power prediction module 152 , a device operation module 154 , and a real-time optimization module 156 .

[0131] A power prediction module (PPM) 152 receives usage and power generation data from the renewable energy control system 120 and a weather and forecast database 160, which includes information related to past (known and historical) environmental and weather data and future (predicted and forecasted) environmental and weather data. The power prediction module 152 includes a machine learning algorithm implemented on a computing system as described below, and is used to generate a model related to future power generation.

[0132] The equipment operation module (POM) 154 is operable to receive equipment operation data from the equipment complex control system 110 and generate a model of equipment operation. The equipment operation module 152 includes a machine learning algorithm implemented on a computing system as will be described below.

[0133] A real time optimization module (RTOM) 156 is arranged to receive inputs from the power prediction module 152 and the plant operation module 154 and derive a plant operation strategy including set point operating parameters. These are then fed to the plant complex control system 110 to control the associated processes controlled thereby.

[0134] The details and operation of each component will now be described.

[0135] Power Prediction Module (PPM) 152

[0136] The power prediction module 152 includes a machine learning algorithm implemented on a computing system and is operable to generate a model to predict future power generation. In an embodiment, one aspect of the power prediction module 152 is the ability to predict future power generation from variable and / or intermittent sources (such as renewable power sources) so that one or more industrial gas equipment (which typically requires a constant power load) can be controlled without the risk of underpowering the equipment.

[0137] In an embodiment, a further aspect of the power prediction module 152 is to use the predicted power generation data to control the equipment complex 10 or aspects thereof. This will be discussed in more detail below.

[0138] The model for predicting future power is based on a machine learning framework. Any suitable machine learning algorithm can be used. For example, the model can utilize techniques such as gradient boosting (using, for example, XGboost), long short-term memory (LSTM), support vector machine (SVM), or random decision forests can be used in such models.

[0139] Gradient boosting is a machine learning technique for regression and classification problems. A strong predictive model is formed that includes an ensemble of weak predictive models such as decision trees. A stepwise procedure can be used to generate models via steepest descent minimization (among other methods).

[0140] LSTM is an artificial recurrent neural network architecture with feedback connections as well as feedforward connections. A common LSTM cell consists of a cell, an input gate, an output gate, and a forget gate. The cell is operable to remember values ​​over arbitrary time intervals where the flow of information into and out of the cell is regulated by the gates.

[0141] A support vector machine utilizes a set of training examples, each of which is included in one of two categories, and generates a model that assigns new examples to a specific category. Thus, an SVM comprises a non-probabilistic binary linear classifier.

[0142] Random decision forests comprise an ensemble machine learning method that operates by building a large number of decision trees during a training process and outputting the class as a pattern of the classes (classification) or median / average predictions (regression) of the individual trees.

[0143] Regardless of the machine learning algorithm used, the model utilizes a two-step process where a training phase is required before the prediction phase. Both phases are implemented as computer programs on one or more computer systems.

[0144] Specialized hardware may also be used. For example, the training phase may involve the use of the central processing unit (CPU) and graphics processing unit (GPU) components of a computer system. In addition, other specialized hardware such as field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), or other stream processor technologies may also be used.

[0145] Training phase

[0146] The model is trained periodically (e.g., on a daily basis) or on demand (if, for example, the accuracy of the model requires a training process) to establish relationships between predictor variables of renewable energy power devices to determine predicted energy availability over a predetermined future period.

[0147] PPM 152 aims to identify the following predictors:

[0148] Wind power WPi

[0149] Solar Power SPi

[0150] Where index i represents the time from period n to n+k (in intervals of fixed duration). In non-limiting embodiments, the intervals may include 15 minutes or 1 hour.

[0151] Predictor variables included the following time-dependent operating characteristic data:

[0152] Wind power WPi

[0153] Solar Power SPi

[0154] Power load Li from period nm to n-1.

[0155] Predictor variables also include measured environmental and meteorological signals, which may include, but are not limited to, time-dependent environmental data including:

[0156] Air temperature Ti,

[0157] Atmospheric pressure Pi

[0158] Wind speed WSi,

[0159] Cloud cover CCi,

[0160] PrecipitationPi

[0161] HumidityHi

[0162] where index i represents the time from period nm to n+k.

[0163] During the training phase, various machine learning algorithms are used to create mathematical relationships between the variables being predicted and the predictor variables. These relationships are stored in a computer in the form of a series of equations that can be accessed and used to make future predictions.

[0164] These models are generated during a training process that is performed periodically at predetermined training times. At each training time, the machine learning model is trained based on historical time-dependent environmental data and historical time-dependent operating characteristic data obtained within one or more predetermined historical time windows. In other words, a moving historical window of time-dependent data (e.g., the previous 2 days, the previous 2 weeks, etc.) is used to train the machine learning model.

[0165] As described above, meteorological measurements come from a weather and forecast database 160, which may include a weather data service or other Internet-connected source. Load, solar power, and wind power are measured by the operator or fed through the renewable energy control system 120 as part of an automated data collection system.

[0166] Prediction Phase

[0167] In the prediction phase, the equations generated and stored in the training phase are used to make predictions about future behavior.

[0168] The predicted variables include forecasts of wind power WPi and solar power SPi for i from c+1 to c+p, where c represents the current time and p represents the forecast horizon, which can be any suitable time scale; for example, non-limiting examples can be 24 hours, 48 ​​hours or longer.

[0169] The predictor variables are:

[0170] Wind power WPi

[0171] Solar Power SPi

[0172] Power load Li

[0173] From cr to c, the meteorological signal is measured (according to the training phase), and from c+1 to c+p the meteorological signal is predicted.

[0174] As for the training phase, in an embodiment, environmental and meteorological measurements are taken from a weather and forecast database 160 , which may include a weather data service or other internet connected source. Load, solar power, and wind power are measured by the operator or automatically fed by the renewable energy control system 120 .

[0175] Figure 3 and 4 The forecasts for wind power and solar power are shown, where the predicted variables are generated from the model forecasting stage described above.

[0176] In an embodiment, the accuracy of the machine learning model is evaluated by comparing the value of the predicted power resource for a predetermined future time period with the actual power resource available to the industrial gas equipment at the end of the predicted future time period. This enables the determination of a prediction error value that provides a measure of the accuracy of the model.

[0177] If the model is not accurate enough, it may need to be trained at an additional training time. The predetermined training time may be selected based on the prediction error value, which in an embodiment may be when the prediction error value exceeds a predetermined threshold.

[0178] In an embodiment, the training time may be selected empirically on a periodic basis; for example, every 24 hours or every 48 hours as appropriate. In an embodiment, the periodic basis may be a default training time policy.

[0179] However, in an embodiment, this may be interrupted when the prediction error value exceeds a predetermined threshold within a predetermined experience interval, in which case the training time is scheduled based on the prediction error value.

[0180] Control based on predicted data

[0181] The PPM 152 may utilize the predicted data to control one or more industrial gas devices in response to the predicted available power resources for a predetermined future time period.For example, in addition to the RTM 156, the PPM 152 may control operational aspects of one or more industrial gas devices.

[0182] In embodiments, the forecast power data may additionally or alternatively control the use of storage resources 28. As described above, storage resources 28 may include resource storage, such as stored hydrogen and / or nitrogen, and energy storage via BESS 28a, CAES / LAES 28b, PHSS 28c, and the like.

[0183] Thus, the PPM 152 may include control and optimization algorithms that utilize predicted power resources to optimize usage of available storage resources 28 to optimally operate the equipment complex 10 and to plan for future power availability and power usage.

[0184] The PPM 152 system uses the predicted powers WPi and SPi as inputs in a real-time optimization problem and applies an optimization algorithm to propose an optimal rate for operating the industrial gas complex 10 to maximize utilization of available power and stored resources in the resource storage 28.

[0185] For example, expected daily and seasonal variability in predicted power generation may be accounted for by PPM 152 by determining optimal usage of short-term (e.g., BESS 28a) and long-term (e.g., CAES / LAES 28b and PHSS 28c) storage systems among storage resources 28.

[0186] In an embodiment, PPM 152 may determine optimal power-on and power-off times for storage resources 28 to maximize utilization of available power. Examples of this are described below.

[0187] In a non-limiting example, the PPM 152 may determine that sufficient power is available over the next 24 hours to operate the electrolyzers 12a ... n of the hydrogen production facility 12 at a greater load to produce more hydrogen than is required to produce the optimal ammonia production rate over a predetermined period of time. The additional hydrogen may then be stored in the hydrogen storage 14 for use during periods of lower power availability. The same applies to the ASU 16 and the nitrogen storage 16a.

[0188] The PPM 152 may receive operational characteristic data for the hydrogen storage 14 and the nitrogen storage 16a. This may include fill levels and other operational data (e.g., fill pressure, fill volume, density, etc.). This operational data may be used by an algorithm forming part of the PPM 152 to determine optimal storage requirements to account for a predicted future power availability profile as a function of time.

[0189] In an embodiment, the PPM 152 may also determine that sufficient power is available within the predicted time window to store additional power in the energy storage 28a, 28b, 28c. The PPM 152 may utilize an optimization algorithm to select the appropriate energy storage 28a, 28b, 28c based on the prediction of power availability with respect to time.

[0190] In a non-limiting example, the PPM 152 may determine that power availability is available over a relatively short period of time (e.g., 1-2 hours) in excess of the needs of the facility complex 10. Given the short ramp rate, fast charge time, and lower capacity of the BESS 28a solution, it may then be determined that the BESS 28a represents the most appropriate energy storage solution for that period of time.

[0191] In an alternative non-limiting example, the PPM 152 may determine that significant power availability is available over a longer period of time (e.g., 5-10 hours) that exceeds the needs of the equipment complex 10. In this case, it may then be determined that the CAES / LAES 28b and PHSS 28c may be more suitable for storing the available power given the slower ramp rates and higher storage capacities of such energy storage solutions.

[0192] The PPM 152 system solves the optimization problem for the next p time periods and applies the available power forecast to a real-time optimization model to maximize the utilization of power. The model may involve the generation of a power utilization metric, and the optimization seeks to optimize the value of the power generation metric.

[0193] The PPM 152 may also include a tracking system to calculate a predicted power utilization metric as a function of the complementarity of the renewable power resources and adjust generation levels / utilization to maximize the long-term utilization of the renewable resources.

[0194] The PPM 152 system is implemented on a computer and receives various inputs from other computer systems. Instances of the PPM 152 system may utilize mixed integer nonlinear programs (MINLPs) because some decisions require some devices to operate in one of multiple possible patterns that result in integer variables. Figure 5 A method according to one embodiment is shown. It should be noted that the following steps do not need to be performed in the order described below, and some steps can be performed simultaneously with other steps.

[0195] In an embodiment, a method of predicting available power resources from one or more renewable power sources for one or more industrial gas plants is provided. The method is performed by at least one hardware processor.

[0196] At step 200, historical time-dependent environmental data associated with one or more renewable power sources 24, 26 is obtained. Historical refers to past environmental data. This may be collected in any suitable time window and may include data within a window extending up to but not including the current time.

[0197] At step 210, historical time-dependent operational characteristic data associated with one or more renewable power sources is collected. History refers to past operational characteristic data, such as power output. This may be collected in any suitable time window, and may include data within a window extending up to but not including the current time.

[0198] At step 220, a machine learning model is trained based on the historical time-dependent environmental data and the historical time-dependent operational characteristic data captured in steps 200 and 210. The model may be trained any suitable number of times and at predetermined intervals or as needed.

[0199] At step 230 , the trained machine learning model is executed to predict available power resources for one or more industrial gas facilities over a predetermined future time period.

[0200] The predicted data may optionally be used to control one or more industrial gas devices in response to the predicted available power resources for a predetermined future time period at step 240. For example, one or more operating set points of the one or more industrial gas devices may be set based on the predicted available power resources.

[0201] Additionally, as described above, an optimization algorithm may be utilized to determine how to best manage the storage resources 28, including resource storage such as hydrogen storage 14 and nitrogen storage 16, and energy storage resources 28a, 28b, 28c, based on the predicted available power resources for a predetermined future period. The optimization algorithm may select the best schedule and / or select the best resources for a given pattern of predicted power availability as a function of time.

[0202] Selection and scheduling can be based on the operating characteristics of the storage resources; for example, the fill level, pressure and other characteristics of the hydrogen storage 14 and nitrogen storage 16a, or the ramp-up / ramp-down, time dependency, maximum generated power / maximum stored energy operating profiles of the energy storage resources 28a, 28b, 28c.

[0203] At step 250, it is determined whether an additional training process is required. This may be based on an empirical metric, such as a predetermined time period. Alternatively, it may be based on an assessment of the accuracy of the machine learning model by comparing the value of the predicted power resource for a predetermined future time period with the actual power resource available to the industrial gas plant at the end of the predicted future time period. This enables the determination of a prediction error value that provides a measure of the accuracy of the model.

[0204] If it is determined that the model needs to be retrained, then a training time can be scheduled and the method can move back to the training in step 220. It should be noted that this may occur after or during either of steps 230 or 240.

[0205] Equipment Operation Module (POM) 154

[0206] Optimization of plant process control is critical to achieving efficiency. Considering that renewable power sources almost always result in changes in available power, the industrial gas plant complex 10 may frequently operate in a dynamic mode. This requires a real-time performance model to design a robust operating strategy. POM 154 includes machine learning and physics-based models to model plant operations. Physics-based models can be used to generate predictor variables that indicate the operating characteristics of the corresponding industrial gas plant. In an embodiment, these can be used together with other operating plant data (e.g., power input, power output, gas output) as time-dependent data input to the corresponding machine learning model.

[0207] As described above, the industrial gas equipment complex 10 includes a hydrogen production equipment 12, a hydrogen storage unit 14, an air separation unit (ASU) 16, an ammonia synthesis equipment 18 and an ammonia storage unit 20, some of which are controlled by corresponding control systems: a hydrogen production equipment control system 112, a hydrogen storage control system 114, an ASU control system 116 and an ammonia synthesis equipment control system 118.

[0208] In an embodiment, at each time step, the power availability forecast from the PPM 152 is used to define the operating set points for the different industrial gas plants 112 , 114 , 116 , 118 .

[0209] These decisions are actionable in embodiments to achieve high process efficiency. This requires an accurate quantitative understanding of the different process units in terms of their real-time performance, system availability information, and resource availability and impending maintenance issues.

[0210] The POM 154 is configured to capture the time-varying properties of each subsystem (ie, the corresponding industrial gas plant) and predict the production efficiency of the overall process in terms of ammonia produced for a given level of energy consumption.

[0211] The high-fidelity model used for such predictions is based on a real-time machine learning framework that uses time-dependent historical data provided by the distributed control system (DCS) in the form of time series for various process tags. Any suitable machine learning algorithm can be used to build an integrated model for the individual subsystems. For example, the model can utilize techniques such as gradient boosting (using, for example, XGboost), long short-term memory (LSTM), support vector machine (SVM), or random decision forests.

[0212] POM 154-Modeling of Hydrogen Production Equipment 12

[0213] Water electrolysis is an energy-intensive process and a key process step in the production of green hydrogen. Each of the electrolyser modules 12a, 12b ... 12n of the hydrogen production facility 12 consists of hundreds of electrolytic cells that work together to convert renewable power into hydrogen molecules controlled by a time-dependent efficiency η.

[0214] Each electrolyser module 12[k] in the hydrogen production facility 12 is independently modeled based on its historical performance data. The predicted operating characteristic variables utilized are:

[0215] The consumed electrolytic cell power [EP(i,k)]

[0216] Electrolyzer hydrogen produced [EH(i,k)]

[0217] It is based on the number of predictor variables such as:

[0218] Demineralized water flow [ED(i,k)]

[0219] The average pool temperature [ECT(i,k)],

[0220] Average pool pressure [ECP(i,k)],

[0221] The current flowing through the electrode [I(i,k)]

[0222] and other key process indicators.

[0223] The historical time series data of the predictor variables and the response variables are sampled at an appropriate frequency [s] over a period of several months and are used to develop a model of the actual module efficiency. The model is built using equations derived from the predictor variables.

[0224] In addition, an accumulator variable is used to track the functional age of the module, which is one of the predictor variables in the model. Reliability and maintenance event information comes from the asset management system.

[0225] The model will be trained periodically or on demand (to build a relationship between the response variable and the predictor variables at time n at each time instance starting from n-1 to nk at intervals of fixed duration (such as 15 minutes or 1 hour).

[0226] In the training system, various machine learning algorithms are used to create mathematical relationships between the predicted variables and the predictor variables.

[0227] POM 154 - Ammonia Loop Modeling

[0228] The ammonia loop is a single unit balanced reaction system that processes a synthesis gas of nitrogen and hydrogen to produce ammonia. Nitrogen is provided by ASU 16, which in embodiments operates continuously to provide nitrogen.

[0229] If the hydrogen production plant 12 is operated in a given case on the basis of the availability of renewable power, hydrogen is provided from the hydrogen production plant, otherwise hydrogen is supplied from the hydrogen storage 14 .

[0230] The stoichiometric composition of the syngas is processed by a syngas compressor system and the product is refrigerated by another set of compressors and sent to storage.

[0231] The performance of the ammonia loop is governed by the equilibrium conversion of the exothermic reaction and is monitored in real time based on a predictive model of the AFi of the ammonia flow into the reservoir as a function of various predictor variables including:

[0232] The power consumed by the ammonia circuit, APi,

[0233] Ammonia circuit pressure and temperature, ALPi, ALTi

[0234] Nitrogen and hydrogen feed flow rates, ANFi, AHFi

[0235] Ammonia plant synthesis gas compressor pressure, ACPi.

[0236] Additional information about the health of the catalyst beds and the timing of maintenance events can be used in the model to obtain the most realistic picture of the conversion loop efficiency.

[0237] Another aspect of the ammonia loop is the different operating modes. In an embodiment, there are two main modes: normal and standby. The normal model involves ramping up and down in response to the amount of available hydrogen. This data is tracked in the DCS and used to view any performance differences or diagnose any process deviations from the production plan. This time-dependent operating characteristic data can be used as input to a trained machine learning model to predict future operating behavior.

[0238] As described below, the model is periodically trained on a longer range of historical data sets (e.g., which may be 6 months to a year) to capture all patterns and different levels of ramp rates.

[0239] POM 154-Hydrogen Storage 14 Modeling

[0240] In an embodiment, the hydrogen storage unit 14 includes a hydrogen purification train, storage, and a set of compressors that are dynamically operable to deliver hydrogen to the ammonia processing equipment and manage hydrogen inventory to avoid downtime due to lack of available gas resources.

[0241] The overall performance of the system is measured by achieving the specified set point of the header pressure in a reliable and energy-efficient manner. The compressors are modeled based on isentropic efficiency based on operating temperature and ambient conditions. Real-time condition monitoring of all compressors is based on adaptive multivariate (Principal Component Analysis (PCA) and Partial Least Squares (PLS) models built on a moving 3-month window of historical data from key process tags.

[0242] Real-time tracking of hydrogen storage can be based on a first main thermodynamic model based on the use of:

[0243] Storage system pressure and temperature, SPi, STi

[0244] Hydrogen compressor pressure and flow, HCPi, HCFi.

[0245] The efficiency of the compressor system is tracked using the consumed compressor power CPi. In addition, the system allows real-time monitoring of the hydrogen purification system in terms of efficiency and reliability.

[0246] POM 154-ASU 16 Modeling

[0247] To model the air separation unit 16, in an embodiment, multivariate partial least squares (PLS) and principal component analysis (PCA) models and engineering models are utilized. These models are operable to diagnose performance impacts and identify preferred operating modes. Data from these models can be used as operational characteristic data input into a trained machine learning model.

[0248] Several key performance indicators (KPIs) may be selected. In non-limiting examples, they may include specific power, N2 recovery, and temperature differential in a heat exchanger forming part of the ASU. The KPIs are tracked in real time to allow for early detection and diagnosis of inefficient operation and emerging equipment health deterioration. The generation of the KPIs is not critical to the present invention, however, these values ​​may be used by a predictive machine learning model to build a predictive model of the performance of the ASU 16.

[0249] Industrial Gas Equipment Model Component Training and Prediction

[0250] For each industrial gas device, a machine learning model is assigned and implemented as described above. For each device, the machine learning model is operable to utilize a training process to generate equations that model the behavior of the corresponding industrial gas device. This is done using the predictor variables described above for each industrial gas device.

[0251] The training process may take historical time-related operational characteristic data as input to train the machine learning model. The operational characteristic data may include physical measurement data related to the corresponding industrial gas equipment, such as, in a non-limiting embodiment, input power, power usage, gas output, measured efficiency, etc.

[0252] In addition, the operational characteristic data may include data generated by one or more physics-based models as described above. The physics-based models may take measured specific industrial gas equipment characteristics (specific to each industrial gas equipment) and may generate one or more metrics indicative of the performance of the industrial gas equipment. These time-dependent metrics may then be used as input operational characteristic data to train a machine learning model assigned to the corresponding industrial gas equipment.

[0253] Once the training process at training time is complete, the model can be used to predict the behavior of the corresponding industrial gas equipment. In the prediction phase, predictions about future behavior are made using the equations generated and stored during the training phase.

[0254] In an embodiment, the accuracy of the machine learning model is evaluated by comparing the value of the predicted future behavior of each industrial gas device at a predetermined future time period with the actual behavior of the industrial gas device at the end of the predicted future time period. This enables the determination of a prediction error value that provides a measure of the accuracy of the model.

[0255] If the model is not accurate enough, it may need to be trained at an additional training time. The predetermined training time may be selected based on the prediction error value, which in an embodiment may be when the prediction error value exceeds a predetermined threshold.

[0256] In an embodiment, the training time may be selected empirically on a periodic basis; for example, every 24 hours or every 48 hours as appropriate. In an embodiment, given that historical data may last for several months, training may not need to be so frequent. In an embodiment, the periodic basis may be a default training time strategy.

[0257] However, in an embodiment, this may be interrupted when the prediction error value exceeds a predetermined threshold within a predetermined experience interval, in which case the training time is scheduled based on the prediction error value.

[0258] POM 154 Summary

[0259] As described above, machine learning models are provided for each industrial gas equipment forming an industrial gas complex. These machine learning models are trained on operating characteristic data generated from measured historical time-related data associated with the relevant equipment and / or from time-related data generated by one or more physics-based models of the corresponding equipment as described above.

[0260] Each of these models generates predictive data related to one or more of: performance; capacity; efficiency; maintenance status; and / or utilization of associated equipment.

[0261] exist Figure 7 An example of generated efficiency curves for process variables is shown in FIG. Figure 7 A graph of the efficiency of an electrolyser versus the load on the electrolyser is shown. The graph is generated from time-dependent operating characteristic data from an electrolyser forming part of the hydrogen plant 12, which data is input into a corresponding machine learning model to generate the operating characteristic data.

[0262] In addition to the predictions of the individual machine learning models assigned to each industrial gas plant, it is possible to utilize an additional model that determines the overall performance of the plant complex in a given situation based on the efficiency of combining the individual modules from each of the above models. An integrated machine learning algorithm is employed to improve the quality of the predictions and select the best model for use in the prediction system.

[0263] In other words, the time-varying operating characteristics of each industrial gas plant forming part of the plant complex over a predetermined future period are predicted by each trained machine learning model and input into a further model to predict the production efficiency of the entire process plant complex. The data from each model may be input into the collective model on a periodic basis; for example, in a non-limiting embodiment, this may be once every 15 minutes.

[0264] With respect to the ammonia production facility of the exemplary embodiment, efficiency determination enables the predicted determination of ammonia produced for a given level of energy input.

[0265] All modeling performed in POM 154 is implemented as a computer program on one or more computers in the device. Regardless of the machine learning algorithm used, the model utilizes a two-step operation process where a training phase is required before the prediction phase. Both phases are implemented as computer programs on one or more computer systems.

[0266] Specialized and non-specialized hardware may also be used. For example, the training phase may involve the use of the central processing unit (CPU) and graphics processing unit (GPU) components of a computer system. In addition, other specialized hardware such as field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), or other stream processor technologies may also be used.

[0267] The model execution computer will be connected to other computer database systems where data from equipment and weather data services will be stored. When the equipment is operating, the performance model is used to make forecasts every 15 minutes to obtain an overview of production.

[0268] All performance models, as well as some unit operation level data, are used at a predetermined frequency in the RTOM 156 to optimize operating efficiency and define set points for the different models.

[0269] Method of operation

[0270] Figure 6 A method according to one embodiment is shown. It should be noted that the following steps do not need to be performed in the order described below, and some steps can be performed simultaneously with other steps.

[0271] In an embodiment, a method of predicting an operational characteristic of an industrial gas plant complex including a plurality of industrial gas plants is provided. The method is performed by at least one hardware processor.

[0272] At step 300, a machine learning model is assigned to each industrial gas equipment forming the industrial gas equipment complex. The model may take any suitable form as described above. It may utilize the historical time-dependent operating characteristics of the industrial gas equipment to generate equations for future predictions.

[0273] At step 310, a respective machine learning model for each industrial gas device is trained based on historical time-dependent operational characteristic data received for the respective industrial gas device. This data may be in any suitable form, and may be specific to a particular type of industrial gas device as described above. History refers to past operational characteristic data. This may be collected in any suitable time window, and may include data within a moving window extending up to, but not including, the current time. In an embodiment, the window may be six months to a year long.

[0274] The historical time-dependent operational characteristic data may include physical measurement data associated with the corresponding industrial gas equipment, such as, in non-limiting examples, input power, power usage, gas output, measured efficiency, etc.

[0275] In addition, the operational characteristic data may include data generated by one or more physics-based models as described above. The physics-based models may take measured specific industrial gas equipment characteristics (specific to each industrial gas equipment) and may generate one or more metrics indicative of the performance of the industrial gas equipment. These time-dependent metrics may then be used as input operational characteristic data to train a machine learning model assigned to the corresponding industrial gas equipment.

[0276] At step 320, the trained machine learning model for each industrial gas device is executed to predict the operating characteristics of each respective industrial gas device over a predetermined future time period. The prediction can optionally be used to control the behavior of the respective industrial gas device, predict possible usage, maintenance planning, resource allocation, or identify process problems and potential problems.

[0277] For example, the predicted data can be used to infer other technical performance of industrial gas equipment. These predictions can be used to determine resource planning, maintenance plans or repair requirements. Such maintenance plans can be combined with the determination of the capacity of power resources and storage units. For example, maintenance of gas generation components (e.g., electrolyzers, ASUs, ammonia production equipment) can be scheduled during time periods when gas storage is high and the predicted available renewable power is low, so as to minimize interruptions and maintain continuity of service provision.

[0278] Additionally, in the following step 330, the data may also be used to determine a set point characteristic.

[0279] At step 330, the predicted operating characteristics determined by the model in step 320 for each respective industrial gas plant are used in another collective model to generate an operational performance metric for the industrial gas plant complex. In the ammonia production plant of the exemplary embodiment, the efficiency determination enables the predicted determination of ammonia produced for a given level of energy input.

[0280] At step 340, the predicted data for a predetermined future time period is compared with the actual measured data at the end of the time period. This comparison is used to infer other technical performance of the industrial gas equipment. These predictions can be used to determine resource planning, maintenance plans or repair requirements.

[0281] In an embodiment, the forecast for a predetermined time period (e.g., a time window of 2 weeks, one month, six months from the generation of the forecasted data or from a timestamp in the forecasted data) is then compared with the actual data for the time window or time period at the end of the time period covered by the forecasted data (e.g., after 2 weeks / one month / six months from the generation of the forecasted data or from a timestamp in the forecasted data). This enables early identification of potential problems in the equipment complex 10, as deviations of any industrial gas equipment or storage system from the forecast model based on past actual behavior may indicate the development of production or maintenance problems.

[0282] By using this approach, potential future problems can be identified early, allowing remedial action to be taken prior to any critical failures or unplanned downtime requiring equipment service for emergency maintenance or repairs.

[0283] Furthermore, in embodiments, scheduling of maintenance may be accomplished in conjunction with determination of the capacity of power resources and storage units. For example, maintenance of gas generating components (e.g., electrolyzers, ASUs, ammonia production equipment) may be scheduled during time periods when gas storage is high and forecasted available renewable power is low in order to minimize disruptions and maintain continuity of service provision.

[0284] In an embodiment, the predicted data may optionally be used to control one or more industrial gas devices in response to the predicted operating characteristics of the industrial gas devices within a predetermined future time period. For example, one or more operating set points of one or more industrial gas devices may be set based on the predicted behavior.

[0285] Control in step 340 may be accomplished in conjunction with the method described in steps 200 to 250, wherein the predicted power availability is utilized in conjunction with the predicted efficiency determination in step 330 to enable a set point to be determined based on the predicted power availability as well as the equipment efficiency.

[0286] At step 350, a determination is made as to whether an additional training process is required. This may be based on an empirical measure, such as a predetermined time period. Alternatively, it may be based on an assessment of the accuracy of the machine learning model by comparing the value of the predicted operating characteristic for a predetermined future time period with the actual operating characteristic of the industrial gas plant at the end of the predicted future time period. This enables the determination of a prediction error value that provides a measure of the accuracy of the model.

[0287] If it is determined that the model needs to be retrained, then a training time can be scheduled and the method can move back to the training in step 310. It should be noted that this may occur after any of steps 320, 330 or 340.

[0288] Real-Time Optimization Module (RTOM) 156

[0289] RTOM 156 includes a system for determining the rates at which various industrial gas plants should operate to best manage the production and storage of renewable hydrogen while maximizing the production of ammonia. In other words, RTOM 156 is capable of real-time optimization of an industrial gas plant complex using renewable power. More specifically, in an embodiment,

[0290] In an embodiment, the system solves an optimization algorithm applied to a dynamic mathematical model of an industrial gas plant complex. The RTOM 156 system uses the predicted powers WPi and SPi and the state of the industrial gas plant, such as efficiency inferred from plant specific factors (such as current, pressure, temperature and flow rate measurements) discussed above with respect to POM 154. These values ​​are taken as inputs and an optimization algorithm is applied to propose an optimal operating rate for the ammonia plant during the time period from c+1 to c+p.

[0291] The rates of other industrial gas equipment such as the hydrogen production equipment 12, the hydrogen compression and storage system 14, the air separation unit 16, the nitrogen storage 16a and the water equipment are linked to the rate of ammonia and are controlled by the low level controller as described above. Only the first value from the list of best values ​​for time c+1 is implemented and the calculation is repeated with the new data at time c+1 as it becomes available.

[0292] The RTOM 156 may also utilize data related to resource storage devices or energy storage devices of the storage resource 28. As described above, the energy storage device may include one or more of the following: a battery energy storage system (BESS) 28a, a compressed / liquid air energy system (CAES or LAES) 28b, or a pumped hydraulic storage system (PHSS) 28c. The status, operating characteristics, availability, resource storage level, and ease of power availability of each unit of the storage resource 28 may be taken into account in the optimization problem.

[0293] The RTOM 156 system is implemented on a computer and receives various inputs from other computer systems. The predicted power comes from the PPM 152. The state of the industrial gas plant is represented by equations that relate the power consumption of various units such as the electrolyzer (ECi), hydrogen compressor (CPi), ammonia plant (APi), and ASU (NPi) to the process variables associated with these units.

[0294] The RTOM 156 system solves an optimization problem that seeks to maximize ammonia production given the total amount of power predicted to be available over the next p time periods, the amount of hydrogen available in storage, and the constraints of physics-based or data-based equations describing the operation of the process equipment.

[0295] Typically, such problems are mixed integer nonlinear programming (MINLP) because the process equipment equations are nonlinear and some decisions require some equipment to operate in one of several possible modes resulting in integer variables. The optimization generates set points from time c+1 to c+p to balance the predicted power generated and the power consumed so that the right amount of hydrogen is produced and the ammonia equipment operates at the correct rate to maximize ammonia production. The RTOM 156 system also takes hydrogen storage into account and based on future power forecasts, hydrogen can be stored or consumed from storage.

[0296] In addition, the RTOM 156 system can choose to recommend that equipment enter standby or shutdown mode based on power availability and equipment availability information. The output of the RTOM 156 system is a recommended ammonia production rate, which is automatically transmitted to the advanced control system that controls the ammonia plant, ASU, electrolyzer, water plant, and hydrogen compression and storage.

[0297] Examples of this are Figure 8 , which shows the optimum ammonia plant rate as a function of time over a 48 hour forecast period.

[0298] Method of operation

[0299] In an embodiment, a method of controlling an industrial gas plant complex including a plurality of industrial gas plants powered by one or more renewable power sources is provided. The method is performed by at least one hardware processor.

[0300] At step 410, time-dependent predicted power data for a predetermined future time period is received from one or more renewable power sources. In a non-limiting embodiment, this may be determined by the PPM 152 according to the process and method discussed in steps 200 to 250. Alternatively, the predicted power data may be obtained according to any other suitable process.

[0301] At step 420, time-dependent predicted operating characteristic data for each industrial gas plant is received. In this context, "industrial gas plant" may also include industrial gas storage, for example, hydrogen, nitrogen and / or ammonia storage as described above in the present embodiment.

[0302] In an embodiment, the operational characteristic data may be based on information about the POM 156 and Figure 6 The data may be generated by the protocol described in method steps 300 to 350. However, the data may also be determined according to any other suitable process.

[0303] At step 430, the predicted power data and the predicted characteristic data are utilized in an optimization model to generate a set of state variables (which may be optimized state variables) for a plurality of industrial gas devices. In an embodiment, this may be accomplished by solving an optimization problem. For example, in a non-limiting embodiment, the optimization model may define the predicted power data and the predicted characteristic data as a set of nonlinear equations. Storage resource data may optionally be included in the power data. The state variables are then generated by solving the set of nonlinear equations.

[0304] At step 440, a set of control set points for a plurality of industrial gas plants are generated using the generated state variables (which may be optimized state variables). The set points may be defined to achieve any particular goal. For example, the set points may be defined at a particular time to ensure that the industrial gas plants operate efficiently and effectively given available power and storage resources. The set points may alternatively or additionally be used to maximize the production of industrial gas given a predicted power availability.

[0305] As a further example, predicted power availability and predicted efficiency and operating characteristics of individual industrial gas plants and / or industrial gas plant complexes as a whole can be used to prevent power shortages of individual plants and / or plant complexes, or to increase production when power availability is sufficient at the current time but a future shortage is predicted.

[0306] At step 450, the control set points are sent to a control system to control the industrial gas plant complex by adjusting one or more control set points of the industrial gas plant.

[0307] In summary, in an exemplary embodiment, a control and optimization system for an industrial gas plant complex is provided. The industrial gas plant complex includes a plurality of industrial gas plants, including a hydrogen production plant using electrolysis of water, an air separation plant for producing nitrogen, an ammonia synthesis plant for producing ammonia, a hydrogen storage system, and an ammonia storage and transportation system.

[0308] In an exemplary embodiment, the system seeks to maximize ammonia production given an uncertain future power input to the system. The system also includes a machine learning based program to predict future power input based on weather forecasts. The program continuously learns from environmental and weather patterns and generated power to make future power generation predictions. The system also includes another machine learning program to learn operational data from the industrial gas equipment and modify the mathematical model used for optimization.

[0309] Although the invention has been described with reference to the preferred embodiments shown in the drawings, it will be understood that various modifications are possible within the spirit and scope of the invention as defined in the appended claims.

[0310] In the specification and claims, the term "industrial gas plant" is intended to mean process equipment that produces or participates in the production of industrial gases, commercial gases, medical gases, inorganic gases, organic gases, fuel gases and green fuel gases in gaseous, liquefied or compressed form.

[0311] For example, the term "industrial gas equipment" may include process equipment for manufacturing gases such as those described in NACE 20.11 class and including, but not limited to: elemental gases; liquids or compressed air; refrigeration gases; mixed industrial gases; inert gases, such as carbon dioxide; and barrier gases. In addition, the term "industrial gas equipment" may also include process equipment for manufacturing industrial gases (such as ammonia) in NACE 20.15 class, process equipment for extracting and / or manufacturing methane, ethane, butane or propane (NACE 06.20 class and 19.20 class) and manufacturing gaseous fuels as defined by NACE 35.21 class. The above has been described for the European NACE system, but is intended to cover equivalent categories under the North American classifications SIC and NAICS. In addition, the above list is non-limiting and non-exhaustive.

[0312] In some examples, a hydrogen storage system is shown, and in some cases a purification unit is shown. However, it should be understood that the present invention can be implemented without the use of a hydrogen storage system or purification unit, which are shown here only for the sake of completeness.

[0313] In this specification, unless explicitly stated otherwise, the word "or" is used in the sense of an operator that returns a true value when either or both of the conditions are met, rather than the operator "exclusive or" which requires only one of the conditions to be met. The word "comprising" is used in the sense of "including", rather than meaning "consisting of".

[0314] In the discussion of the embodiments of the present invention, the pressures given are absolute pressures unless otherwise indicated.

[0315] All prior teachings mentioned above are hereby incorporated herein by reference. Acknowledgment of any previously published document in this document is not to be taken as an acknowledgment or representation that its teachings were common general knowledge in Australia or elsewhere at that time.

[0316] Where applicable, the various embodiments provided by the present disclosure can be implemented using a combination of hardware, software, or hardware and software. In addition, where applicable, the various hardware components and / or software components set forth herein can be combined into composite components comprising software, hardware, and / or both, without departing from the spirit of the present disclosure. Where applicable, the various hardware components and / or software components set forth herein can be divided into subcomponents comprising software, hardware, or both, without departing from the scope of the present disclosure. In addition, where applicable, it is expected that software components can be implemented as hardware components, and vice versa.

[0317] Software according to the present disclosure (such as program code and / or data) may be stored on one or more computer-readable media. It is also contemplated that the software identified herein may be implemented using one or more general or special-purpose computers and / or networked and / or otherwise computer systems. Where applicable, the order of the various steps described herein may be changed, combined into composite steps, and / or divided into sub-steps to provide the features described herein.

[0318] Although various operations have been described herein in terms of "modules," "units," or "components," it should be noted that these terms are not limited to a single unit or function. In addition, the functions attributed to some modules or components described herein may be combined and attributed to fewer modules or components. In addition, although the present invention has been described with reference to specific examples, these examples are merely illustrative and are not intended to limit the present invention. It will be apparent to those of ordinary skill in the art that the disclosed embodiments may be changed, added, or deleted without departing from the spirit and scope of the present invention. For example, one or more parts of the above method may be performed in a different order (or simultaneously) and still obtain the desired results.

Claims

1. A method of controlling an industrial gas plant complex, the industrial gas plant complex comprising a plurality of industrial gas plants powered by one or more renewable power sources, the method being performed by at least one hardware processor, the method comprising: receiving time-dependent predicted power data from the one or more renewable power sources for a predetermined future time period; Time-dependent predicted operating characteristic data is received for each industrial gas device, the time-dependent predicted operating characteristic data including data for: Hydrogen storage, for hydrogen storage of hydrogen output from a hydrogen production facility; Nitrogen storage, nitrogen storage for air separation units (ASU), Electrolyser hydrogen produced for hydrogen production facilities: and Ammonia flows into the storage and is used in the ammonia circuit of the ammonia production equipment; utilizing the predicted power data and the predicted characteristic data in an optimization model to generate a set of state variables for the plurality of industrial gas plants, the industrial gas plants including a hydrogen production plant and an ammonia production plant; utilizing the generated state variables to generate a set of control set points for the plurality of industrial gas plants for maximizing ammonia production of the ammonia production plant; as well as The control set points are sent to a control system to control the industrial gas plant complex by adjusting one or more control set points of the industrial gas plant. 2 . The method of claim 1 , wherein the optimization model defines the predicted power data and the predicted characteristic data as a set of nonlinear equations.

3. The method of claim 2, wherein the state variables are generated by solving the set of nonlinear equations.

4. The method of claim 1, wherein the time-dependent predicted power data is generated from a trained machine learning model.

5. The method according to claim 4, wherein the time-dependent predicted power data is obtained by: obtaining historical time-dependent environmental data associated with the one or more renewable power sources; obtaining historical time-dependent operating characteristic data associated with the one or more renewable power sources; training a machine learning model based on the historical time-related environmental data and the historical time-related operational characteristic data; as well as The trained machine learning model is executed to predict available power resources for the one or more industrial gas facilities over a predetermined future time period.

6. The method of claim 1, wherein the time-dependent predicted operating characteristic data is generated from a trained machine learning model for each of the industrial gas equipment.

7. The method of claim 6, wherein the time-dependent predicted operating characteristic data for each industrial device is obtained by: assigning a machine learning model to each of the industrial gas plants forming the industrial gas plant complex; training a corresponding machine learning model for each industrial gas device based on historical time-dependent operating characteristic data received about the corresponding industrial gas device; and The trained machine learning model is executed for each industrial gas device to predict operating characteristics of each respective industrial gas device over a predetermined future time period.

8. The method of claim 1, wherein the industrial gas equipment complex includes storage resources, the storage resources comprising one or more industrial gas storage containers for hydrogen storage, one or more industrial gas storage containers for nitrogen storage, and / or one or more energy storage resources.

9. The method of claim 8, wherein the one or more energy storage resources include one or more of: a battery energy storage system; a compressed air energy storage; a liquid air energy storage; or a pumped hydroelectric energy storage.

10. The method of claim 8, wherein the predicted power data further comprises data representative of an operating parameter of the storage resource.

11. The method of claim 10, wherein the data representing operating parameters of the storage resource comprises one or more of: resource memory availability; fill level; and utilization.

12. A system for controlling an industrial gas plant complex, the industrial gas plant complex comprising a plurality of industrial gas plants powered by one or more renewable power sources, the system comprising: At least one hardware processor operable to perform: receiving time-dependent predicted power data from the one or more renewable power sources for a predetermined future time period; Time-dependent predicted operating characteristic data is received for each industrial gas device, the time-dependent predicted operating characteristic data including data for: Hydrogen storage, for hydrogen storage of hydrogen output from a hydrogen production facility; Nitrogen storage, nitrogen storage for air separation units (ASU), Electrolyser hydrogen produced for hydrogen production facilities: and Ammonia flows into the storage and is used in the ammonia circuit of the ammonia production equipment; generating a set of state variables for the plurality of industrial gas plants, the industrial gas plants comprising a hydrogen production plant and an ammonia production plant, using the predicted power data and the predicted characteristic data in an optimization model; generating a set of control set points for the plurality of industrial gas plants using the generated state variables to maximize ammonia production of the ammonia production plant; as well as The control set points are sent to a control system to control the industrial gas plant complex by adjusting one or more control set points of the industrial gas plant.

13. The system of claim 12, wherein the optimization model defines the predicted power data and predicted characteristic data as a set of non-linear equations.

14. The system of claim 13, wherein the state variables are generated by solving the set of nonlinear equations.

15. The system of claim 12, wherein the time-dependent predicted power data is generated from a trained machine learning model.

16. The system of claim 12, wherein the time-dependent predicted operating characteristic data is generated from a trained machine learning model for each of the industrial gas equipment.

17. A computer readable storage medium storing a program of instructions executable by a machine to perform a method of controlling an industrial gas plant complex including a plurality of industrial gas plants powered by one or more renewable power sources, the method being performed by at least one hardware processor, the method comprising: receiving time-dependent predicted power data from the one or more renewable power sources for a predetermined future time period; Time-dependent predicted operating characteristic data is received for each industrial gas device, the time-dependent predicted operating characteristic data including data for: Hydrogen storage, for hydrogen storage of hydrogen output from a hydrogen production facility; Nitrogen storage, nitrogen storage for air separation units (ASU), Electrolyser hydrogen produced for hydrogen production facilities: and Ammonia flows into the storage and is used in the ammonia circuit of the ammonia production equipment; utilizing the predicted power data and the predicted characteristic data in an optimization model to generate a set of state variables for the plurality of industrial gas plants, the industrial gas plants including a hydrogen production plant and an ammonia production plant; utilizing the generated state variables to generate a set of control set points for the plurality of industrial gas plants for maximizing ammonia production of the ammonia production plant; as well as The control set points are sent to a control system to control the industrial gas plant complex by adjusting one or more control set points of the industrial gas plant.

18. The computer-readable storage medium of claim 17, wherein the optimization model defines the predicted power data and predicted characteristic data as a set of non-linear equations.

19. The computer-readable storage medium of claim 18, wherein the state variables are generated by solving the set of nonlinear equations.

20. The computer-readable storage medium of claim 17, wherein the time-dependent predicted power data is generated from a trained machine learning model, and / or the time-dependent predicted operating characteristic data is generated from a trained machine learning model for each of the industrial gas equipment.

Images