Systems, analyzers, controllers, and associated methods to enhance fluid production of refining operations

Abstract

Embodiments of systems and methods for enhancing control of catalytic reformer operations are disclosed. A method includes receiving product data indicative of a property of a product of a catalytic reformer and receiving feedstock data indicative of a feedstock for processing by the catalytic reformer. A target octane number or target octane-barrels for the reformate may be specified and a machine learning model generates an adjustment to a parameter of the catalytic reformer based on the feedstock data and the product data to achieve an enhanced yield of hydrogen gas while achieving the target octane number of the reformate, wherein the machine learning model is trained with catalytic reformer historical data indicative of operating parameters, operational constraints, the feedstock, and products of the catalytic reformer.

Description

SYSTEMS, ANALYZERS, CONTROLLERS, AND ASSOCIATED METHODS TO ENHANCE FLUID PRODUCTION OF REFINING OPERATIONSCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to, and the benefit of U.S. Provisional Application No. 63 / 786,014, filed April 9, 2025, titled SYSTEMS, ANALYZERS, CONTROLLERS, AND ASSOCIATED METHODS TO ENHANCE FLUID PRODUCTION OF REFINING OPERATIONS, U.S. Provisional Application No. 63 / 778,798, filed March 27, 2025, titled SYSTEMS, ANALYZERS, CONTROLLERS, AND ASSOCIATED METHODS TO ENHANCE FLUID PRODUCTION OF REFINING OPERATIONS, U.S. Patent Application No. 18 / 948,759, filed November 15, 2024, titled “SYSTEMS, ANALYZERS, CONTROLLERS, AND ASSOCIATED METHODS TO ENHANCE FLUID SEPARATION FOR DISTILLATION OPERATIONS,” U.S. Provisional Application No. 63 / 660,196, filed June 14, 2024, titled “SYSTEMS, ANALYZERS, CONTROLLERS, AND ASSOCIATED METHODS TO ENHANCE FLUID PRODUCTION OF REFINING OPERATIONS,” U.S. Provisional Application No. 63 / 658,825, filed June 11, 2024, titled “SYSTEMS, ANALYZERS, CONTROLLERS, AND ASSOCIATED METHODS TO ENHANCE FLUID PRODUCTION OF REFINING OPERATIONS,” and U.S. Provisional Application No. 63 / 655,589, filed June 3, 2024, titled “SYSTEMS, ANALYZERS, CONTROLLERS, AND ASSOCIATED METHODS TO ENHANCE FLUID PRODUCTION OF REFINING OPERATIONS,” the disclosures of which are incorporated herein by reference in their entireties.FIELD OF DISCLOSURE

[0002] The disclosure herein relates to systems, analyzers, controllers, and associated methods to enhance fluid production for refining operations and, more particularly, to systems, analyzers, controllers, and associated methods to enhance fluid production of refining operations and sub-operations using machine learning models during the refining operations and sub-operations.BACKGROUND

[0003] Many and varied operations are executed continuously and simultaneously at a refinery. Each operation affects each subsequent operation or sub-operation. For example,if the product from a first operation is produced based on maximizing a first factor, for example, the research octane number of gasoline, the production of that particular product affects further downstream operations. Additionally, further upstream operations may not be suited for production of that particular product or may, at least, cause other operations to produce that particular product in an inefficient manner. Additionally, a variety of starting feedstock are utilized at a refinery. Even further, different batches or portions of one feedstock may vary over time, for example, different portions of a feedstock may include different properties and / or compositions over time. Optimization (in other words, efficient and accurate production of targeted products) of such operations and feedstock poses a significant challenge when attempting to meet a target product objective, particularly over a period time, as equipment and materials used in the operation change over time. Such problems pose further difficulties since updating one operation affects every other operation at the refinery.

[0004] Controllers and monitoring devices may be utilized at a refinery to attempt to optimize (in other words, efficiently and accurately produce targeted products) those operations. However, those controllers and monitoring devices utilize algorithms that require expert personnel and that take extended amounts of time to execute. For example, first-principle models require expert personnel to ensure that the first-principle model is accurately calculating some formula, in other words, expert personnel are required to maintain the first-principle model. Further still, the equipment utilized at one refinery may experience a different service or maintenance cycle than equipment at another refinery. Such factors further complicate any attempt at uniform optimization at a plurality of refineries.SUMMARY

[0005] Thus, in view of the foregoing, Applicant has recognized these problems and others in the art, and has recognized a need for systems, analyzers, controllers, and associated methods for enhancing fluid production for refinery operations. Particularly, the present disclosure relates to systems, analyzers, controllers, and associated methods to enhance fluid production of refining operations and sub-operations using machine learning models during the refining operations and sub-operations. Such fluids may include hydrocarbons and / or renewable hydrocarbons and fluid production may include, for example, production of transportation fuel, among other products.

[0006] The disclosure herein provides embodiments of systems, analyzers, controllers, and associated methods for enhancing fluid production of ongoing and / or continuous refining operations, as well as refining sub-operations. Such systems, analyzers, controllers, and associated methods may include obtaining data corresponding to a refinery operation from one or more sources, such as sensors, analyzers, refining equipment, devices, and / or other sources. The data, along with, in some embodiments, a target product, may then be applied to a machine learning model to produce an output indicative of or including parameters that indicate settings for the devices and / or refining equipment to be set to, to accurately achieve or produce the targeted product.

[0007] Accordingly, an embodiment of the disclosure is directed to a method including receiving product data indicative of a product component of a product of a catalytic reformer and receiving feedstock data indicative of feedstocks for processing by the catalytic reformer. The method further includes receiving a target octane number for the reformate produced by the catalytic reformer and generating, by the machine learning model, an adjustment to a parameter of the catalytic reformer based on the feedstock data and the product data to achieve an enhanced yield of hydrogen gas while achieving the target octane number of the reformate. The machine learning model is trained with catalytic reformer historical data indicative of operating parameters, operational constraints, feedstocks, and products of the catalytic reformer.

[0008] An embodiment may include a method including method comprising receiving product data indicative of a product component of a product of a catalytic reformer and receiving feedstock data indicative of feedstocks for processing by the catalytic reformer. The method further includes receiving a target octane number for the reformate produced by the catalytic reformer and generating, by the machine learning model, an adjustment to a parameter of the catalytic reformer based on the feedstock data and the product data to achieve the target octane number of the reformate while maximizing yield of the reformate. The machine learning model is trained with catalytic reformer historical data indicative of operating parameters, operational constraints, feedstocks, and products of the catalytic reformer.

[0009] An embodiment may include a method including analyzing a naphtha sample to provide naphtha sample properties. The naphtha sample is taken from naphtha being supplied to a reformer and predicting one or more naphtha properties associated with the naphtha sample based on the naphtha sample properties and a first output from application of the naphtha sample properties to a machine learning model trained on historical dataincluding operational parameters of the reformer, naphtha sample properties, and unit materials sample properties. The method further includes operating the reformer to produce unit materials, the unit materials having unit materials properties, and the unit materials including reformate and hydrogen and providing operating parameters of the operating reformer to the machine learning model. The method then analyzes a unit material sample to provide unit material sample properties and predicts one or more unit material properties associated with the unit material sample based on the unit material sample properties and a second output from application of the unit material sample properties to the machine learning model. The method then includes generating, by the machine learning model, an adjustment to a parameter of the reforming operation based on the naphtha sample properties and the unit material sample properties to optimize reformer production of unit materials having a property at or exceeding a target property for the unit materials.

[0010] Another embodiment may include a method including receiving product data indicative of a property of a product of a catalytic reformer and receiving feedstock data indicative of feedstocks for processing by the catalytic reformer. The catalytic reformer includes a first reactor in series with a second reactor. The method further includes generating, by the machine learning model, an adjustment to a parameter of the first reactor based on the feedstock data and the product data to improve the property of the product of the catalytic reformer while reducing yield loss of the product. The machine learning model is trained with catalytic reformer historical data indicative of operating parameters, operational constraints, feedstocks, and products of the catalytic reformer. The method further includes generating, by the machine learning model, an adjustment to a parameter of the second reactor based on the adjustment to the parameter of the first reactor, feedstock data, and the product data.

[0011] Another embodiment of the disclosure is directed to a system for enhancing reformate production in a reformer operation. The system may include a reformer to receive a feedstock and convert the feedstock into reformate and by-product hydrogen. The system may include a plurality of sensors to measure a parameter associated with the reformer and each positioned at one of (a) proximate the reformer or (b) within the reformer. The system may include a plurality of refining operation control devices each positioned proximate and downstream or upstream of the reformer and to control an aspect and / or property of fluid flowing to or from the reformer. The system may include one or more sample collection assemblies to collect samples of the fluid associated with the reformer. The system may include one or more sample analysis assemblies to analyze each collected sample to provideproperties of collected samples. The system may include a reformer controller in signal communication with the reformer, the plurality of sensors, the plurality of refining operation control devices, and the one or more sample analysis assemblies, and storing a trained machine learning model. The reformer controller may be configured to determine an output including predicted properties of the feedstock and parameter settings of the plurality of refining operation control devices and the reformer based on application of one or more of (i) data measured by the plurality of sensors, (ii) data corresponding to analysis from the one or more sample analysis assemblies, or (iii) a target product and corresponding target properties to the trained machine learning model, and adjust one or more of the amount of feedstock or type of feedstock and parameters associated with the plurality of refining operation control devices and reformer based on the output to enhance production of the reformate, hydrogen, sand other gasses.

[0012] Another embodiment of the disclosure is directed to a method for enhancing control of a reforming operation associated with a petroleum refining operation. The method may include supplying naphtha to a reformer associated with the petroleum refining operation, the naphtha having one or more naphtha properties. The method may include analyzing a naphtha sample via a first analyzer to provide naphtha sample properties. The method may include predicting one or more naphtha sample properties associated with the naphtha sample based on (A) the naphtha sample properties and (B) a first output from application of the naphtha sample properties to a first trained machine learning model. The method may include operating the reformer to produce one or more unit materials, the one or more unit materials having one or more unit materials properties, and the one or more unit materials comprising one or more of reformate, hydrogen, or reformer gas. The method may include analyzing the unit material sample via a second analyzer to provide unit material sample properties. The method may include predicting one or more unit material sample properties associated with the unit material sample based on (C) the unit material sample properties and (D) a second output from application of the unit material sample properties to a second trained machine learning model. The method may include controlling, during the reforming operation, based on the naphtha sample properties and the one or more unit material sample properties, one or more of: (a) one or more naphtha properties associated with the naphtha supplied to the reformer; (b) one or more unit product materials properties associated with the unit product materials; (c) operation of the reformer; or (d) operation of one or more upstream equipment or downstream equipment, so that the controlling, during the reforming operation, causes the reforming operation toproduce one or more of: (i) one or more intermediate materials each having one or more properties within a range of one or more target properties of the one or more intermediate materials, (ii) one or more unit product materials each having one or more properties within a range of one or more target properties of the one or more unit product materials, or (iii) one or more downstream materials each having one or more properties within a range of one or more target properties of the one or more downstream materials, thereby to cause the reforming operation to achieve material outputs that more accurately and responsively converge on one or more of the target properties.

[0013] Still other aspects and advantages of these embodiments and other embodiments, are discussed in detail herein. Moreover, it is to be understood that both the foregoing information and the following detailed description provide merely illustrative examples of various aspects and embodiments, and are intended to provide an overview or framework for understanding the nature and character of the claimed aspects and embodiments. Accordingly, these and other objects, along with advantages and features of the present disclosure herein disclosed, will become apparent through reference to the following description and the accompanying drawings. Furthermore, it is to be understood that the features of the various embodiments described herein are not mutually exclusive and may exist in various combinations and permutations.BRIEF DESCRIPTION OF THE DRAWINGS

[0014] These and other features, aspects, and advantages of the disclosure will become better understood with regard to the following descriptions, claims, and accompanying drawings. It is to be noted, however, that the drawings illustrate only several embodiments of the disclosure and, therefore, are not to be considered limiting of the scope of the disclosure.

[0015] FIG. 1A and FIG. IB are simplified diagrams of a refining control system to enhance fluid production at a refinery, according to an embodiment of the disclosure.

[0016] FIG. 2 is a simplified diagram that illustrates an apparatus for enhance fluid production at a refinery, according to an embodiment of the disclosure.

[0017] FIG. 3 is another simplified diagrams of a refining control system to enhance fluid production at a refinery, according to an embodiment of the disclosure.

[0018] FIG. 4 is a simplified diagram that illustrates training of a machine learning model for enhanced fluid production at refinery, according to an embodiment of the disclosure.

[0019] FIG. 5 is a schematic diagram of an enhanced hydrotreater control system and a distillation control system to enhance fluid production at a portion of a refinery, according to an embodiment of the disclosure.

[0020] FIG. 6 is a schematic diagram of an enhanced catalytic reformer control system to enhance fluid production at a portion of a refinery, according to an embodiment of the disclosure.

[0021] FIG. 7 is a schematic diagram of an enhanced aromatics recovery control system to enhance fluid production at a portion of a refinery, according to an embodiment of the disclosure.

[0022] FIG. 8 is a schematic diagram of hydrogen control system to enhance fluid production at a portion of a refinery, according to an embodiment of the disclosure.

[0023] FIG. 9 is a schematic diagram of a feed control system to enhance fluid production at a portion of a refinery, according to an embodiment of the disclosure.

[0024] FIG. 10 and FIG. 11 are simplified diagrams of control systems to enhance to enhance fluid production at refinery, according to an embodiment of the disclosure.

[0025] FIG. 12 is a flow chart illustrating enhanced fluid production at a refinery, according to an embodiment of the disclosure.

[0026] FIG. 13 is a schematic diagram of processes within a refinery.

[0027] FIG. 14 is a schematic diagram of a machine learning model that may be used with the various processes within a refinery.

[0028] FIG. 15 is a flow chart of a method, according to at least one embodiment of the present disclosure.

[0029] FIG. 16 is a flow chart of another method according to one embodiment, according to at least one embodiment of the present disclosure.

[0030] FIG. 17 is flow chart of yet another method, according to at least one embodiment of the present disclosure.

[0031] FIG. 18 is a flow chart of a method, according to at least one embodiment of the present disclosure.

[0032] FIG. 19 is a flow chart of another method according to one embodiment, according to at least one embodiment of the present disclosure.

[0033] FIG. 20 is flow chart of yet another method, according to at least one embodiment of the present disclosure.

[0034] FIG. 21 is flow chart of a method, according to at least one embodiment of the present disclosure.

[0035] FIG. 22 is flow chart of another method, according to at least one embodiment of the present disclosure.

[0036] FIG. 23 is flow chart of a method, according to at least one embodiment of the present disclosure.

[0037] FIG. 24 is flow chart of another method, according to at least one embodiment of the present disclosure.DETAILED DESCRIPTION

[0038] So that the manner in which the features and advantages of the embodiments of the systems and methods disclosed herein, as well as others that will become apparent, may be understood in more detail, a more particular description of embodiments of systems and methods briefly summarized above may be had by reference to the following detailed description of embodiments thereof, in which one or more are further illustrated in the appended drawings, which form a part of this specification. It is to be noted, however, that the drawings illustrate only various embodiments of the systems and methods disclosed herein and are therefore not to be considered limiting of the scope of the systems and methods disclosed herein as it may include other effective embodiments as well.

[0039] The disclosure herein provides embodiments of systems, analyzers, controllers, and associated methods for enhancing fluid production of ongoing and / or continuous refining operations, as well as refining sub-operations. Such systems, analyzers, controllers, and associated methods may include obtaining data corresponding to a refinery operation from one or more sources, such as sensors, analyzers, refining equipment, refining operation control devices, other devices, and / or other sources. The data, along with, in some embodiments, a target product, may then be applied to a machine learning model to produce an output indicative of or including parameters that indicate settings for the refining operation control devices and / or refining equipment to be set to, to accurately achieve or produce the targeted product. Such an application of data to a trained machine learning model may occur at one or more different layers in the control system of the refinery. Further, a plurality of controllers positioned throughout the refinery may each include a plurality of trained machine learning models that are trained to enhance fluid production of one or more refinery operations or sub-operations.

[0040] In embodiments, the refining operations and / or sub-operations may include the processing, converting, refining, enhancing and / or otherwise altering a fluid via the refining operation or sub-operation. The fluid may be a liquid, vapor, and / or gas and may include ahydrocarbon and final product may include a transportation fuel. “Hydrocarbons” or “hydrocarbon fluids” as used herein, may refer to petroleum fluids, renewable fluids, and other hydrocarbon based fluids. “Petroleum fluids” as used herein, may refer to fluid products containing crude oil, petroleum products, natural gas, renewable liquids and / or gasses, and / or distillates or refinery intermediates. For example, crude oil contains a combination of hydrocarbons having different boiling points that exists as a viscous liquid in underground geological formations and at the surface. Petroleum products, for example, may be produced by processing crude oil and other liquids at petroleum refineries, by extracting liquid hydrocarbons at natural gas processing plants, and by producing finished petroleum products at industrial facilities. For example, a petroleum product may include a transportation fuel, among other products. Refinery intermediates, for example, may refer to any refinery hydrocarbon that is not crude oil or a finished petroleum product (such as gasoline), including all refinery output from distillation (for example, distillates or distillation fractions) or from other conversion units. In some non-limiting embodiments of systems and methods, petroleum fluids may include heavy blend crude oil used at a pipeline origination station, natural gas, and / or other types of crude oil, as will be understood by one skilled in the art. Heavy blend crude oil is typically characterized as having an American Petroleum Institute (API) gravity of about 30 degrees or below. In other embodiments, the petroleum fluids may include lighter blend crude oils, for example, having an API gravity of greater than 30 degrees. “Renewable fluids” as used herein, may refer to fluid products containing plant and / or animal derived feedstock. Further, the renewable fluids may be hydrocarbon based. For example, a renewable fluid may be a pyrolysis oil, oleaginous feedstock, biomass derived feedstock, renewable natural gas or other liquids or gasses, as will be understood by those skilled in the art. The API gravity of renewable liquids may vary depending on the type of renewable liquid.

[0041] As used herein “profit” is the potential financial gain from a potential sale of products from a refinery operation less the costs of feedstocks and operational costs of the refinery operation. The potential value of a sale may be based on current market prices for similar products or current prices paid by contracted customers. Costs of feedstocks may include market prices for similar feedstocks as well as the operational costs of upstream processes preparing the feedstocks for use in the refinery operation. Operational costs may include equipment depreciation, employee wages and benefits, and consumables used to process the feedstocks, such as hydrogen, water, catalysts, steam, natural gas, cooling, electricity, and other utilities. For example, a machine learning model may generateadjustments to one or more operational parameters of a refinery process based on the potential profit from the products it produces. A potential profit may be determined from a generated algorithm or generated simulations based on current market pricing and potential ranges of production rates and the corresponding variable production costs. To continue the example, product A may sell for $2 a barrel and may be produced at a rate ranging from 50 barrels a day to 100 barrels a day. Product B may sell for $4 a barrel and be produced at a rate ranging from 10 barrels a day to 20 barrels a day. However, the more product B produced results in less product A being produced, increased utility costs, and accelerated deactivation of the catalysts of the refinery process. All of these variables and more may be managed by the machine learning model to increase the profitability of the refinery process.

[0042] In an embodiment, the systems and methods may include a computing device, apparatus, and / or controller (referred to hereafter as a controller) to obtain various data points and / or parameters to train a machine learning model. Such data may include a historical data set and / or a currently generated data set including an outcome. As used in this application, “current” means happening now or recently enough to still be relevant to an ongoing process. For example, a current sample data may have been generated from a sample obtained several hours ago from a feedstock that is still being processed by a refinery process. In another embodiment, the data set may include a simulated and / or filled- in data set. For example, a refinery may be modeled based on a first-principle model and synthetic or pseudo-data may be generated for a selected time interval (for example, 1 month, 2 months, 6 months, 1 year, or even longer). For such a data set, random perturbations and / or anomalies are used to simulate a data set. In another embodiment, the data may include a partial data set. In such an embodiment, the partial data set, may be filled in via a first principle model and / or a machine learning model. Each data set may include a series of parameters, properties, spectra, and / or other data points associated with a refining operation or process or sub-operation or sub-process. Each data set may also include target parameters, target properties, and / or an outcome and / or target product. Further, in an embodiment where a supervised machine learning model is utilized, each outcome may be classified as a positive or negative outcome or marked in a manner to indicate desirability of the outcome. In other embodiments, the function generated by a set of data may indicate a desired outcome, based on a maximum or minimum point in that function, thus enabling a machine learning model to determine desired parameters based on that maximum or minimum, or based on some other factor in other embodiments. In yetanother embodiment, a trained machine learning model may learn or be trained based on trends included in the data (in other words, the trained machine learning model may comprise a deep learning model). In another embodiment, any of the trained machine learning models described herein may predict and / or optimize target parameters and / or fluids used within a refinery, refinery operation, or refinery sub-operation.

[0043] Once these data sets have been received by the controller, the controller may pre- process the data. For example, the controller may normalize the data (in other words, remove data points that appear to be outliers), remove data corresponding to abnormal events (for example, data generated during start-up, shut-down, turn-arounds, and / or upsets), remove undesired data, remove invalid measurements, and / or segregate the data set into sequences of contiguous data based on selected time intervals (for example, time intervals of 30 minutes, 1 hour, 2 hours, and / or 3 hours, or more or less than the time intervals listed).

[0044] Once the data has been pre-processed, the controller may begin training a model based on a portion of the data set. For example, the controller may utilize an 80 / 20 training and testing process. Other percentages may be utilized in training, testing, and / or validation. As the model is fed data, the model may compare data received to the outcome (in other words, whether the outcome was desired based on some factor, such as an indicated positive / negative flag or classification, based on some maximum or minimum of a function generated based on the data, or based on a trend within the data). Once the training portion of data has been utilized, the controller may test and / or validate the model using the remining portion of the data set. If such testing or validation does not achieve a selected error rate or reach some other error and / or accuracy based threshold, then the controller may re-train or refine the model using a different and / or randomized portion of the data set and a remaining portion of the data set for testing. Once a model has reached that threshold, then the controller may output the trained machine learning model for further use.

[0045] Further, a trained machine learning model may be further refined using new data, as such data is generated. Such a refinement may occur while the trained machine learning model is in use. In another embodiment, the trained machine learning model may be refined in an offline environment. In another embodiment, two instances of a trained machine learning model may exist, one stored as an offline copy, while the other is utilized during refining operations. In such an embodiment, the offline copy may be refined and, if testingand / or error rating meets a selected threshold, in addition to other factors, then a controller may replace the version currently utilized to the refined version.

[0046] It will be understood that such systems and methods described herein may utilize a number of trained machine learning models (also referred to as trained models). For example, a model may be trained for each specific operation or process, as well as each particular piece of equipment, at a refinery, such as fluid catalytic cracking (FCC) operations or processes, hydrocracking operations or processes, reforming operations or processes, alkylation operations or processes, isomerization operations or processes, hydrotreating operations or processes, distillation operations or processes, blending operations or processes, hydrodeoxygenation operations or processes, steam management, hydrogen coordination or management, absorption, propylene splitting operations or processes, aromatic recovery, sulfur recovery, coker unit operations, feed optimization, IMO blending, hydrodeoxygenation, hydrocracker operations, other blending operations, residuum oil supercritical extraction operation, solvent deasphalting (SDA) operation, operations or processes for formation of specific fuels, and / or the refining operation or process overall (which may, in an embodiment, utilize outputs from models associated with each sub-operation or sub-process). The use of terms operation and process refers to the steps taken to produce a particular product from a selected feedstock (and, in some embodiments, other inputs). As such, when referring to a particular refining operation or process, the terms “operation” and “process” may be used interchangeably. Further, such models may be trained specifically for equipment at a particular plant or refinery. For example, a FCC unit at a first plant may exhibit different characteristics than that of a FCC unit at a second plant. Thus, a model trained for one may not work for the other and training a model for either FCC unit may include utilization of historical data corresponding to that FCC unit. Various aspects of one model may be utilized to train other models for other similar equipment though.

[0047] Once a model is available, the controller, one or more sub-operation controllers or sub-controllers, and / or one or more operation controllers including a local enhancement or optimization module or circuitry, predictive controls, and / or equipment and device controls may begin optimizing, enhancing, and / or adjusting an operation and / or parameters associated with that operation, in real-time, near real-time, and / or continuously or substantially continuously, at a refinery. In such embodiments, the controller may obtain data from a plurality of sensors, a plurality of refining operation control devices (such as flow control devices, temperature control devices, pressure control devices, and / or otherdevice configured to control an aspect of a refining operation), equipment at the refinery (in other words, refining equipment), and / or one or more sample analyzers and / or, in some embodiments, one or more sub-operation controllers or sub-controllers. In another embodiment, one or more operation controllers may obtain such data, as well as target products and / or other factors or parameters from a refinery controller or platform.

[0048] As noted, one input to any of the models described herein may include spectra or properties of feedstock, intermediaries, products or outputs, and / or other fluids or materials utilized in a refinery, determined via one or more of a spectrographic analyzer or a chromatographic analyzer. The controller or controllers may work in conjunction with such an analyzer to further enhance fluid production (for example, transportation fuel, hydrocarbon based fluid products, and / or other fluids produced during a refining operation) of the refining operation or sub-operation. As such, spectrographic analyzers may be calibrated or standardized and results may be obtained in a faster than typical timeframe, thus enabling prompt acquisition of fluid properties. For example, for any operation described herein, the controller may obtain spectrographic or chromatographic analysis of any feedstock utilized, any intermediaries produced, and / or any products produced by first initiating sample collection. Once a sample has been obtained, the controller and / or an analyzer may initiate analysis of the sample.

[0049] Once the controller or controllers has / have obtained data related to each operation and / or analysis of one or more fluids associated with the operation, then the controller may apply such data and analysis to a corresponding machine learning model. The output of the model may indicate adjustment of one or more devices or refining operation control devices and / or refining equipment and / or adjustment of a feedstock or intermediary used in the operation or sub-operation. In some embodiments, the output may include targets and / or properties for a feedstock and / or blend of feedstock. In another embodiment, the output may be in the form of a vector, each component of the vector corresponding to a value associated with a parameter of equipment or a device or refining operation control device. In an embodiment, a refining operation control device may comprise or include a temperature control device (such as a furnace, heat exchanger, condenser, boiler, induction coil, fans, a cooling device, and / or other device capable of adjusting the temperature of a fluid and / or the temperature within refining equipment), a flow control device (such as a pump, valve, control valve, and / or other device capable of adjusting the flow rate of a fluid), a pressure control device (such as a compressor, a pump, a let-down station or valve, and / oranother device configured to adjust the pressure of a fluid), and / or other devices configured to adjust some aspect of a fluid and / or aspect of refining equipment.

[0050] Once the controller has the output of the model, the controller may adjust the relevant aspects of the refining operation. For example, the controller may adjust components of a blend utilized in a feedstock, settings for various refining operation control devices (such as temperature, pressure, flow rate, and / or another aspect associated with a fluid and / or device), use of hydrogen, recovery of selected fluids or materials, and / or use of other fluids or materials (for example, a catalyst), among other adjustments.

[0051] In another embodiment, the controller may optimize an operation based on the current demand for selected products. For example, for a particular targeted product, selected amounts of feed and / or intermediaries may be utilized, increasing the demand for that feed and / or intermediaries. In other embodiments, demand may be a factor utilized in training a model. For example, a selected product may experience increased demand at varying times or a particular feedstock, used to produce a particular product, may be in high demand. Data indicating such demand may be utilized in the described trained learning models.

[0052] In yet another embodiment, the controller may compare the output of the model to the current properties for a selected operation. Based on that difference of such a comparison, the controller may adjust various aspects of that operation.

[0053] By utilizing the trained machine learning models, the systems and methods described herein may determine specific adjustments to a plurality of operations and parameters specific to equipment at a refinery to accurately and more frequently (as compared to typical adjustment times) reach a target product. Further, such adjustments may increase efficiency of the refinery equipment and / or reduce energy utilized by the refinery equipment, thus reducing cost of the refinery operation. The target product may be based on a number of factors, such as demand and / or price or cost for the product, cost of the product and / or feedstock, and / or based on a target product provided by a refinery controller or platform. Such adjustments may be determined in real-time or near real-time using data from continuous and / or ongoing refinery operations.

[0054] Thus, rather than attempting to adjust operations at a significant delay, a refinery’s operations may be adjusted in-real time, near real-time, or at time intervals shorter than in typical optimization operations (such typical optimization operations including operations by operating personnel to efficiently and accurately produce a target product). Further, such adjustments may be determined faster than typical adjustments to operations, leading torelevant and timely adjustments. Further, such analysis and adjustment utilizes complex non-linear equations which typically take longer to analyze, however with the use of machine learning, such analysis occurs significantly faster and with comparable accuracy.

[0055] FIG. 1 A and FIG. IB simplified diagrams of a refining control system to enhance fluid production at a refinery, according to an embodiment of the disclosure. As illustrated in FIG. 1A, a refinery 100 may include various refining control operation devices and refining equipment. While selected equipment are illustrated in FIG. 1A, it will be understood by those skilled in the art that additional and / or different equipment may be included in or at a refinery 100, particularly based on the type of feedstock processed at the refinery. For example, the refinery 100 may include a desalter, blending tanks, storage tanks, and / or wastewater treatment units, among other equipment. Further, each unit or equipment at the refinery 100 may be optimized using the machine learning models disclosed herein, using data specific to the equipment from the refinery 100. In other words, the equipment may be operated such that the corresponding refinery operations produce an accurate and / or on-specification target product.

[0056] As illustrated in FIG. 1A, a refinery 100 may include a refinery controller 101 and / or a plurality of operation controllers 102. As will be illustrated in subsequent drawings, additional components may be included, such as a refining enhancer, other circuitry, various other controllers, and / or other computing devices. In an embodiment, the refinery controller 101 and / or the plurality of operation controllers 102 may include, for example, a trained machine learning model, as well as other instructions to adjust various devices and / or operations or processes within the refinery 100. The refinery controller 101 and / or the plurality of operation controllers 102 may connect to or be in signal communication with (a) one or more sensors, meters, transducers, and / or other measurement devices positioned throughout the refinery 100 and / or (b) to the equipment (for example, connected to some control aspect or device associated with the equipment) positioned at the refinery 100. The refinery controller 101 may be configured to receive data via such a connection. Further, the refinery controller 101 may receive such data in real-time or near real-time. In an embodiment, the refinery controller 101 may determine a target product and / or other parameters for a selected period of time. The refinery controller 101 may provide such data to each of the operation controllers 102. In other embodiments, the refinery controller 101 may utilize outputs from each of the operation controllers 102 to determine parameters for a target product. In other embodiments, the refinery controller 101 may apply those outputs, as well as other data, to a machine learning model.

[0057] In another embodiment, each of the operation controllers 102 may include a local enhancement circuitry 184, predictive controls circuitry 194, 191, and / or 195, and / or equipment and device controls 199. In embodiments, the circuitry may be a module or instructions. In embodiments, the operation controller 102 may include one or more varying or different predictive controls. For example, as illustrated, one or more of the predictive controls circuitry 191 may include a trained machine learning model 193. The trained machine learning model 193 may be trained for a specific operation and / or piece of equipment and, in some embodiments, may be trained to recognize an adjustment, maximization, and / or optimization for specified factors of the specific operation and / or piece of equipment. As such, the operation controller 102 may include a plurality of predictive control circuitry 191. The operation controls may also include predictive control circuitry 194, which includes a trained machine learning model 196 and / or a first-principle model (and / or, in some embodiments, another type of model). The trained machine learning model 196 may be trained to fill missing data for the first-principle model 198. The operation controller may also include predictive control circuitry 195, which may include a first-principle model 197. The first-principle model 197 may be a model using a known, physics based equation or formulation.

[0058] In an embodiment, the operation controller 102 may include a local enhancement circuitry 184. The local enhancement circuitry 184 may include a trained machine learning model 190 and target setpoint instructions 192. The trained machine learning model 190 may utilize data associated with a specific refining operation and / or the output from each predictive controls circuitry to produce an output. The target setpoint instructions may utilize the output of the trained machine learning model 190 to determine a set of parameters that equipment and / or devices associated with a specific refining operation should be set to, to reach a target product. The operation controller may also include the equipment and device controls 199. The equipment and device controls 199 may cause equipment and / or devices to adjust to the target setpoints.

[0059] As illustrated in Fig. IB, the operation controllers 102 may further be connected to or in signal communication with a sample collection assembly 186 and / or a sample analysis assembly or sample analyzer 188. The sample analyzers 188 may include spectrographic analyzers, standardized spectrographic analyzers, and / or chromatographic analyzers. The type of spectrographic analyzers utilized may include one or more of near-infrared spectroscopic analyzer, a mid-infrared spectroscopic analyzer, a combination of a nearinfrared spectroscopic analyzer and a mid-infrared spectroscopic analyzer, a Ramanspectroscopic analyzer, or a nuclear magnetic resonance spectroscopic analyzer. The sample analyzer 188 may analyze received samples and provide corresponding spectra indicating properties or other analysis indicating components and / or properties of the sample. The operation controllers 102 may also be connected to one or more sub-controllers or sub-operation controllers that are positioned or configured to manage selected aspects or operations of the refinery 100.

[0060] The refinery controller 101 and / or operation controllers 102 may include a processor and a memory or non-transitory machine-readable storage medium storing instructions executable by the processor (as illustrated in subsequent drawings). In some examples, the refinery controller 101 and / or the operation controller 102 may be a computing device. The term “computing device” is used herein to refer to any one or all of programmable logic controllers (PLCs), distributed control systems (DCSs), a proportional integral derivative (PID) controller, a DCS-PID controller, programmable automation controllers (PACs), industrial computers, servers, virtual computing device or environment, desktop computers, personal data assistants (PDAs), laptop computers, tablet computers, smart books, palm-top computers, personal computers, smartphones, virtual computing devices, cloud based computing devices, and similar electronic devices equipped with at least a processor and any other physical components necessarily to perform the various operations described herein. Devices such as smartphones, laptop computers, and tablet computers are generally collectively referred to as mobile devices.

[0061] The term “server” or “server device” is used to refer to any computing device capable of functioning as a server, such as a master exchange server, web server, mail server, document server, or any other type of server. A server may be a dedicated computing device or a server module (e.g., an application) hosted by a computing device that causes the computing device to operate as a server. A server module (e.g., server application) may be a full function server module, or a light or secondary server module (e.g., light or secondary server application) that is configured to provide synchronization services among the dynamic databases on computing devices. A light server or secondary server may be a slimmed-down version of server type functionality that can be implemented on a computing device, such as a smart phone, thereby enabling it to function as an Internet server (e.g., an enterprise e-mail server) only to the extent necessary to provide the functionality described herein.

[0062] As used herein, a “non-transitory machine-readable storage medium” or “memory” may be any electronic, magnetic, optical, or other physical storage apparatus to contain orstore information such as executable instructions, data, and the like. For example, any machine-readable storage medium described herein may be any of random access memory (RAM), volatile memory, non-volatile memory, flash memory, a storage drive (e.g., hard drive), a solid state drive, any type of storage disc, and the like, or a combination thereof. The memory may store or include instructions executable by the processor.

[0063] As used herein, a “processor” or “processing circuitry” may include, for example one processor or multiple processors included in a single device or distributed across multiple computing devices. The processor (such as, processing circuitry 202 shown in FIG. 2, and / or a processor included in, for example, refinery controller 101 and / or the operation controllers 102 (not illustrated)) may be at least one of a central processing unit (CPU), a semiconductor-based microprocessor, a graphics processing unit (GPU), a field- programmable gate array (FPGA) to retrieve and execute instructions, a real time processor (RTP), other electronic circuitry suitable for the retrieval and execution instructions stored on a machine-readable storage medium, or a combination thereof.

[0064] Turning to the equipment positioned at the refinery 100, the refinery 100 may include a reactor 104 and, in some embodiments, a regenerator 120. The reactor 104 may be a catalytic reactor and / or a fluid catalytic cracking unit or reactor. The reactor 104 may include one or more sensors or meters positioned within the reactor 104 (such as sensor or meter 108) and / or proximate the reactor (such as sensors or meters 112, 114, 110, and 144). These sensor or meters may measure some aspect of fluid or material where the sensor or meter is positioned. Further, the reactor 104 may receive feedstock and / or an amount of water / steam at one or more locations of the reactor 104. The reactor 104 may be positioned or configured to convert heavy gas oil, residua, and / or other gas oil blends to an effluent or cracked fluid including smaller molecules, the effluent, in some embodiments, being further separated downstream into different products via distillation or fractionation. The reactor 104 may be operated at or it may be beneficial to operate the reactor 104 at a selected temperature range and / or pressure range to reduce over-cracking, which may cause a loss in valuable products, as well as to reduce over-all energy usage. Further, the amount of catalyst within and / or being fed to (for example, from the regenerator 120 and / or as fresh catalyst) the reactor 104 may impact the value of product produced by the reactor 104. Further, as a catalyst is regenerated within the regenerator 120, degradation may occur and / or coke deposited on the catalyst may not be completely burned off, particularly after multiple uses, thus further impacting product from the reactor 104. The feedstock fed to the reactor 104 may also affect parameters, such as temperature and residence time, amongother parameters. Thus, several factors and / or parameters may impact the product produced by a reactor 104, those factors and / or parameters including temperature, pressure, type of catalyst, quality of catalyst, amount of catalyst, flow rate of feed or feed stock and / or catalyst, amount and temperature of steam injected, and / or properties of feed or feedstock, in addition to the product desired or targeted. The refinery controller 101 and / or operation controllers 102 may gather data related to such factors over time and apply that data, along with, in some embodiments, spectra provided from the sample analyzer 188, to trained machine learning models to produce parameters that enable production of a target product via a minimal amount of energy and / or lowest cost. Such application of data to a model may occur within various modules or circuits of one or more of the operation controllers 102 and / or, in addition to other data generated within the refinery 100, within the refinery controller 101. For example, one trained machine learning model within the predictive controls module 191 may be trained to maximize or be utilized for maximizing operating temperature in relation feedstock and a threshold temperature that may cause overcracking. In another example, another trained machine learning model may be trained to adjust or be utilized for adjusting heater temperature and / or temperature within the reactor in relation to feedstock and process parameters, composition, and / or other aspects to maximize yield or economically optimize yield from the reactor 104. In another embodiment, the trained machine learning model may be trained to adjust or be utilized for adjusting one or more refinery operation control devices to produce a selected yield of a product that also maximizes profit. Such an application of data to such a model may generate a vector that includes parameters or parameter settings corresponding to devices and / or equipment associated with the reactor 104 and, in some embodiments, the regenerator 120. That vector may be utilized by the local enhancement model to further determine, based on the outputs from other models as well as gathered data, parameters or parameter settings that enhance production of effluent from the reactor 104, such parameters or parameter settings being applied to actual equipment and / or devices via the equipment and device controller. Other models may be trained to determine parameters based on other relationships associated with the reactor 104 and / or other equipment.

[0065] In an embodiment, a trained machine learning model may be utilized by predictive controls (which may also be referred to as a prediction model) to optimize targets and / or properties and / or the predictive controls may be utilized by a an online optimization algorithm (which may also be referred to as the local enhancement module) to generate or determine targets.

[0066] The refinery 100 may include a regenerator 120. While a reactor 104 with a side- by-side configuration is illustrated in FIG. 1A, it will be understood that other configurations may be utilized, such as a stacked configuration. In an embodiment, in addition to reactor data and corresponding fluid properties, refinery controller 101 and / or operation controllers 102 may obtain data and fluid properties corresponding to the regenerator 120. For example, the refinery controller 101 and / or operation controllers 102 may obtain data from sensors or meters 116, 124, 126, 128, 132, 134, 138, and 140, flow control devices associated with the regenerator 120 (such as valves 118, 130, 136, and 142), and / or the regenerator 120, as well as properties or spectra associated with spent catalyst, regenerated catalyst, a feed or feedstock (for example, to aid in catalyst regeneration), and / or air (which may include pure oxygen or some combination of oxygen and other elements). Data may be obtained from other devices, such as flow control devices (such as, valves and / or pumps, among other devices configured to control flow of a fluid) and / or temperature control devices (such as boilers, heat exchangers, heating coils, condensers, and / or other heating or cooling devices). In an embodiment, the regenerator 120 may be positioned or configured to burn coke off of spent catalyst, the coke being deposited onto the catalyst in the reactor 104. The regenerator 120 may then provide the regenerated catalyst back to the reactor 104. In embodiments, the refinery controller 101 and / or operation controllers 102 may apply the data from the regenerator 120 to produce parameters or parameter settings to adjust corresponding equipment or devices to. The trained machine learning models may be trained or be utilized to determine parameters to maximize the amount of carbon build up burned from catalyst, to increase temperature within the reactor 104 (for example, via heat from regenerated catalyst), and / or to minimize the amount of resources utilized by the regenerator 120.

[0067] Other equipment may be positioned throughout the refinery 100 and the refinery controller 101 and / or operation controllers 102 may connect to such equipment. The refinery controller 101 and / or operation controllers 102 may obtain or gather data related to that equipment during the refining operation. For example, the refinery controller 101 and / or operation controllers 102 may obtain data from and / or related to a fractionation column 148 or distillation column. Further, the refinery 100 may include and the refinery controller 101 and / or operation controllers 102 and / or one or more of the sub-controllers 184, 191, 194, 195, 199, may obtain data from a hydrotreater (such as hydrotreater 166 and hydrotreater 174) and / or an alkylation unit 158. The refinery controller 101 and / or operation controllers 102 may obtain data from the valve 146, sensors or meters 150, 154,160, 164, 168, 172, 178, and 180, as well as the properties associated with the products from the fractionation column 148 (for example, off gas 152, LPG 156, alkylate 162, gasoline 170, diesel 176, slurry 182, and / or other products), the hydrotreater 166 (for example, gasoline or high-octane gasoline), the hydrotreater 174 (for example, diesel, low- sulfur diesel, and / or higher purity diesel), and / or the alkylation unit 158 (for example, alkylate).

[0068] In an embodiment, the refinery controller 101 and / or operation controllers 102 may obtain data in real time, near real-time, and / or continuously or substantially continuously. In another embodiment, the refinery controller 101 and / or operation controllers 102 may obtain data periodically. In another embodiment, the refinery controller 101 and / or operation controllers 102 may apply data to a trained machine learning model at a selected time interval. Such an interval may be based on the time for samples from various positions within the refinery 100 to be collected and then analyzed. In yet another embodiment, each of the operation controllers 102 may obtain data related to a selected section of the refinery 100. Each of the operation controllers 102 may also obtain properties and / or spectra from the sample analyzer 188. After the operation controllers 102 obtain the properties and / or spectra and data, then the operation controllers 102 may apply the properties and / or spectra and data to a corresponding predictive controls module to produce parameters to produce a target product. The local enhancement module of the operation controller may then apply, to a trained machine learning model of the local enhancement module, the output of each of the predictive controls module, the data obtained throughout the refinery 100, and / or the properties and / or each spectra associated with a collected sample. Such an application may produce an enhanced or optimized set of parameters, which may then be applied to equipment or devices via the equipment and devices controls.

[0069] In yet another embodiment, the refinery controller 101 may first obtain data and the properties and / or spectra and then apply the data and the properties and / or spectra to a trained machine learning model. The refinery controller 101 may transmit the output of the trained machine learning model to each operation controller 102. In another embodiment, the refinery controller 101 may provide target products and corresponding parameters to each of the operation controllers 102, based on user input, previously utilized parameters, current cost of a target product and / or feedstock, and / or other factors. In another embodiment, the refinery controller 102 may facilitate communication between each of the operation controllers 102, facilitate data acquisition for the operation controllers 102, and / or facilitate parameter prediction and / or adjustment among the plurality of operationcontrollers 102. For example, if one operation controller adjusts a process to meet a selected target, that adjustment may affect upstream and / or downstream processes. The refinery controller 101 may facilitate communication and / or perform additional predictions to ensure that such parameter adjustments enable the upstream and / or downstream processes to continue to produce target products.

[0070] FIG. 2 is a simplified diagram that illustrates an apparatus for enhance fluid production at a refinery, according to an embodiment of the disclosure. Such an apparatus 200 may be comprised of a processing circuitry 202, a memory 204, a communications circuitry 206, a modeling circuitry 208, a fluid adjustment circuitry 210, and an equipment and device adjustment circuitry 212, each of which will be described in greater detail below. While the various components are illustrated in FIG. 2 as being connected with processing circuitry 202, it will be understood that the apparatus 200 may further comprise a bus (not expressly shown in FIG. 2) for passing information amongst any combination of the various components of the apparatus 200. The apparatus 200 may be configured to execute various operations described herein, such as those described above in connection with FIGS. 1 A-1B and below in connection with FIGS. 3-24.

[0071] The processing circuitry 202 (and / or co-processor or any other processor assisting or otherwise associated with the processor) may be in communication with the memory 204 via a bus for passing information amongst components of the apparatus. The processing circuitry 202 may be embodied in a number of unusual ways and may, for example, include one or more processing devices configured to perform independently. Furthermore, the processor may include one or more processors configured in tandem via a bus to enable independent execution of software instructions, pipelining, and / or multithreading.

[0072] The processing circuitry 202 may be configured to execute software instructions stored in the memory 204 or otherwise accessible to the processing circuitry 202 (e.g., software instructions stored on a separate storage device). In some cases, the processing circuitry 202 may be configured to execute hard-coded functionality. As such, whether configured by hardware or software methods, or by a combination of hardware with software, the processing circuitry 202 represents an entity (for example, physically embodied in circuitry) capable of performing operations according to various embodiments of the present disclosure while configured accordingly. Alternatively, as another example, when the processing circuitry 202 is embodied as an executor of software instructions, the software instructions may specifically configure the processing circuitry 202 to perform the algorithms and / or operations described herein when the software instructions are executed.

[0073] Memory 204 is non-transitory and may include, for example, one or more volatile and / or non-volatile memories. In other words, for example, the memory 204 may be an electronic storage device (for example, a computer readable storage medium). The memory 204 may be configured to store information, data, content, applications, software instructions, or the like, for enabling the apparatus 200 to carry out various functions in accordance with example embodiments contemplated herein.

[0074] The communications circuitry 206 may be any means such as a device or circuitry embodied in either hardware or a combination of hardware and software that is configured to receive and / or transmit data from / to a network and / or any other device, circuitry, or module in communication with the apparatus 200. In this regard, the communications circuitry 206 may include, for example, a network interface for enabling communications with a wired or wireless communication network. For example, the communications circuitry 206 may include one or more network interface cards, antennas, buses, switches, routers, modems, and supporting hardware and / or software, or any other device suitable for enabling communications via a network. Furthermore, the communications circuitry 206 may include the processing circuitry for causing transmission of such signals to a network or for handling receipt of signals received from a network. The communications circuitry 206, in an embodiment, may enable reception of parameters from various components, devices, and / or sensors (for example, flow control devices, analyzers, sensors, equipment, and / or other components), as well as communication of instructions and / or signals indicative of adjustment to those components and / or devices.

[0075] The apparatus 200 may include a modeling circuitry 208 configured to obtain parameters from one or more components, equipment, devices, sensors, and / or analyzers and / or apply those parameters to a trained machine learning model to obtain parameters that enable equipment to produce a target product. In other embodiments, the modeling circuitry 208 may apply, in addition to the parameters described herein, the output of other similar circuitry (in other words, an additional plurality of modeling circuitry that each correspond to one of a plurality of sub-operations). Obtaining the parameters from the one or more components, equipment, devices, sensors, and / or analyzers may occur periodically, at selected times, continuously, or substantially continuously. In an example, the modeling circuitry 208 may obtain parameters for sub-operations first, generating an output for each sub-operation. Upon generation of an output for each sub-operation, the modeling circuitry 208 may obtain each output and a current data set. The modeling circuitry 208 may poll the components, devices, sensors, and / or analyzers to obtain such parameters or, in anembodiment, receive the parameters without polling. The modeling circuitry 208 may obtain the parameters via the communications circuitry 206. Application of the parameters to the trained machine learning model may determine, generate, or cause generation of an output. The output may be indicative of an adjustment to equipment, fluids, devices, and / or operations to meet or accurately meet a target product and / or to operate the equipment at higher than typical efficiency, for example, utilizing less power or resources such as in a heater or boiler or utilizing a heat exchanger to reduce power usage.

[0076] In another embodiment, the modeling circuitry 208 may train the trained machine learning model prior to use. In such embodiments, the modeling circuitry 208 may obtain historical data, preprocess the historical data, and then train and test the machine learning model. In yet another embodiment, after a refining operation (in other words, after a selected product has been generated via the refining operation), the modeling circuitry 208 may re-train or refine the trained machine learning model, based on the results of the refining operation (in other words, the accuracy of the parameters in reaching the target product’s properties).

[0077] In another, the modeling circuitry 208 may train and / or include a plurality of machine learning models. Each of the plurality of machine learning models may correspond to a selected refinery operation and / or sub-operation.

[0078] The modeling circuitry 208 may utilize processing circuitry 202, memory 204, or any other hardware component included in the apparatus 200 to perform these operations, as described above in connection with FIGS. 1A-1B and below in connection with FIGS. 3-10. The output of the modeling circuitry 208 may be transmitted to other circuitry of the apparatus 200 (such as the fluid adjustment circuitry 210 and / or equipment and device adjustment circuitry 212).

[0079] In addition, the apparatus 200 further comprises the fluid adjustment circuitry 210 that may cause adjustment of feedstock and / or other fluids utilized in a refining operation. In an embodiment the output from the modeling circuitry 208 may be a matrix, a series of parameters, and / or some indicator. In an embodiment, the fluid adjustment circuitry 210 may utilize that output to adjust a blend of feedstock and / or the fluid used in other inputs (for example, an amount of hydrogen, butane, other alkanes, and / or other fluids). The fluid adjustment circuitry 210 may utilize processing circuitry 202, memory 204, or any other hardware component included in the apparatus 200 to perform these operations, as described above in connection with FIGS. 1A-1B and below in connection with FIGS. 3-24. The fluid adjustment circuitry 210 may further utilize communications circuitry 206 to transmit signals to adjust the type and / or amount of feedstock to utilize.

[0080] In addition, the apparatus 200 further comprises the equipment and device adjustment circuitry 212 that may cause adjustment of equipment and / or devices utilized in a refining operation. In an embodiment, the equipment and device adjustment circuitry 212 may utilize that output to adjust temperature, pressure, flow rate, and / or other parameters corresponding to the equipment and / or devices positioned within the refinery (e.g., by setting valve positions, pump speeds, etc.). The equipment and device adjustment circuitry 212 may utilize processing circuitry 202, memory 204, or any other hardware component included in the apparatus 200 to perform these operations, as described above in connection with FIGS. 1A and IB and below in connection with FIGS. 3-10. The equipment and device adjustment circuitry 212 may further utilize communications circuitry 206 to transmit signals to adjust equipment and / or devices utilized.

[0081] Although components 202-212 are described in part using functional language, it will be understood that the particular implementations necessarily include the use of particular hardware. It should also be understood that certain of these components 202-212 may include similar or common hardware. For example, the modeling circuitry 208, the fluid adjustment circuitry 210, and the equipment and device adjustment circuitry 212 may, in some embodiments, each at times utilize the processing circuitry 202, memory 204, or communications circuitry 206, such that duplicate hardware is not required to facilitate operation of these physical elements of the apparatus 200 (although dedicated hardware elements may be used for any of these components in some embodiments, such as those in which enhanced parallelism may be desired). Use of the terms “circuitry,” with respect to elements of the apparatus therefore shall be interpreted as necessarily including the particular hardware configured to perform the functions associated with the particular element being described. Of course, while the terms “circuitry” should be understood broadly to include hardware, in some embodiments, the terms “circuitry” may in addition refer to software instructions that configure the hardware components of the apparatus 200 to perform the various functions described herein.

[0082] Although the modeling circuitry 208, the fluid adjustment circuitry 210, and the equipment and device adjustment circuitry 212 may utilize processing circuitry 202, memory 204, or communications circuitry 206 as described above, it will be understood that any of these elements of apparatus 200 may include one or more dedicated processors, specially configured field programmable gate arrays (FPGA), or application specificinterface circuits (ASIC) to perform its corresponding functions, and may accordingly utilize processing circuitry 202 executing software stored in a memory or memory 204, communications circuitry 206 for enabling any functions not performed by special-purpose hardware elements. In all embodiments, however, it will be understood that the modeling circuitry 208, the fluid adjustment circuitry 210, and the equipment and device adjustment circuitry 212 are implemented via particular machinery designed for performing the functions described herein in connection with such elements of apparatus 200.

[0083] In some embodiments, various components of the apparatus 200 may be hosted remotely (e.g., by one or more cloud servers) and thus need not physically reside on the corresponding apparatus 200. Thus, some or all of the functionality described herein may be provided by third party circuitry. For example, a given apparatus 200 may access one or more third party circuitries via any sort of networked connection that facilitates transmission of data and electronic information between the apparatus 200 and the third party circuitries. In turn, that apparatus 200 may be in remote communication with one or more of the other components describe above as comprising the apparatus 200.

[0084] As will be appreciated based on this disclosure, example embodiments contemplated herein may be implemented by an apparatus 200 (or by a refinery controller). Furthermore, some example embodiments (such as the embodiments described for FIGS. 1A-1B and 3-10) may be a computer program product comprising software instructions stored on at least one non-transitory computer-readable storage medium (such as memory 204). Any suitable non-transitory computer-readable storage medium may be utilized in such embodiments, some examples of which are non-transitory hard disks, CD-ROMs, flash memory, optical storage devices, and magnetic storage devices. It should be appreciated, with respect to certain devices embodied by apparatus 200 as described in FIG. 2, that loading the software instructions onto a computing device or apparatus produces a special-purpose machine comprising the means for implementing various functions described herein.

[0085] FIG. 3 a simplified diagram that illustrates example refining controllers and an example refining enhancer to enhance control of a refining process at a refinery, according to an embodiment of the disclosure. As illustrated in FIG. 3, a refinery may include one or more operation controllers 302. The operation controllers 302 may connect to, for example, a number of feeds and / or processing units (refinery equipment configured to process a feedstock or other input). As illustrated, the operation controllers 302 may connect to and receive data from feed A 303 A, feed B 303B, and up to feed 303N, sensors or other devicesassociated with each feed (such as sensor 304A, 304B, and up to 304N), and various flow control devices (such as valve 306A, valve 306B, and up to valve 306N). In such embodiments, each of feed A 303 A, feed B 303B, and up to feed 303N may flow to a first processing unit 308. A processing unit, for example, in this case, the first processing unit 308, may include one or more refining devices or equipment positioned at a refinery (for example, a FCC unit, a distillation column, and other equipment as described herein). The first processing unit 308 may convert, process, and / or transform a feed into a unit material (such as unit material A 310A, unit material B 31 OB, and up to unit material N 310 N). Additional processing units may be positioned throughout the refinery. As illustrated though, the unit materials may flow to a “Nth” processing unit 314. Sensors (such as sensor 312A, 312B, and up to 312N) and valves (such as valve 313 A, valve 313B, and up to valve 313N) may be positioned between the feed and “Nth” processing unit 314. The final processing device (in other words, the “Nth” processing unit 314) may produce one or more end materials (such as end materials A 316A, end materials 316B, and up to end materials 316N). Sensors (such as sensor 318A, 318B, and up to 318N) and valves (such as valve 320A, valve 320B, and up to valve 320N) may be positioned between the end materials and the material destination 322.

[0086] In an embodiment, as each feed is fed to the next processing unit, the operation controller 302 may determine various characteristics and / or properties of the feed. For example, the operation controller 302 may determine temperature, pressure, and / or flow rate, in addition to the composition of the feed and the spectra or properties determined via spectrographic analysis of the feed. For example, as illustrated, the operation controller 302 may determine or obtain feed information 324 (including, at least feed composition 326 and / or feed properties 328, among other data), unit material information 330 (including, at least unit material composition 332 and / or unit material properties 334, among other data), and / or end material information 336 (including, at least end material composition 338 and / or end material properties 340, among other data). Thus, the operation controller 302 may obtain data related to each feed / material in real-time or near real-time, during a refinery operation, and / or directly or indirectly (for example, spectra may be obtained via a sample or spectrographic analyzer).

[0087] Once all the data has been obtained, the operation controller 302 may apply, to the machine learning model 360 of a local enhancer 358, the data including processing unit constraints 350, a target product 352 (including a target composition 354 and target properties 356), and / or material differences 344 (including composition differences 346and properties difference 348) as determined via a comparator 342 (the comparator positioned or configured to compare composition and properties of different materials). The machine learning model 360 may produce material targets 362 which may be utilized to produce target feed ratios 364 and target operation unit parameters 366. In another embodiment, these values may be fed to the comparator and then, after obtaining differences related to another material, reapplied to the machine learning model 360. The machine learning model 360 may then produce adjusted targets 370 (including adjusted target feed ratios 372 and adjusted target operation unit parameters 374).

[0088] In another embodiment, the output of the machine learning model (for example, a vector comprising a plurality of components, each component being a parameter setting for a selected or specific device or equipment) may be compared to current parameter settings for the selected or specific devices or equipment in the comparator 342. In such embodiments, if the comparator 342 determines that there is a difference between an output of the machine learning model, then the parameter settings of the equipment or devices at the refinery may be adjusted.

[0089] In embodiments, the controller 302 may drive the materials to the target by adjusting the valves and / or feed (for example, the blend of different feeds or materials used in the subsequent operation) at one or more points in the overall refining operation.

[0090] The machine learning model 360 may include neural networks, supervised learning models, semi -supervised learning models, unsupervised learning models, or some combination thereof, as will be readily understood by one having ordinary skill in the art. In another embodiment, different types of machine learning algorithms may be utilized for different refinery operations. In further embodiments, some refining operations may use, rather than or in addition to a neural network, decision trees, support vector machines, hidden Markov models, Bayesian networks, linear regression, k-means, and / or tabular reinforcement learning. Specific neural networks that may be utilized include a recurrent neural network, such as a long short-term memory network. Such neural networks may utilize a fixed horizon of historical data to predict future behavior. Additionally, such neural networks may utilize standard active functions, for example a rectified linear unit or a hyperbolic tangent. As noted, in embodiments, different models may be utilized for different operations. The determination for which model to use for each operation may be determined based on error rates associated with a selected model, the R2value, SHAP plots and / or values, gain directions and / or magnitude, and / or gain distributions, among other factors.

[0091] In an embodiment, the operation controller 302, local enhancer 358, and / or comparator 342 may be included in a single controller, a plurality of controllers, one or more computing devices, and / or as one or more modules or as instructions. In other embodiments, a plurality of controllers or computing devices may each include a specific model corresponding to one of the processing units. In another embodiment, the operation controller 302 may include or may be a supervisory controller that considers the predictions of other processing unit specific controllers when generating adjusted targets via the supervisory controller’s machine learning model.

[0092] As noted, data may be obtained in real time or near real-time. In some embodiments, the application of data to a machine learning model may be delayed by the time taken to obtain spectra or properties for a feed or material. Thus, in an embodiment where the operation controller 302 is a supervisory controller, the supervisory controller may generate adjusted targets after each sub-controller generates a target for a specific processing unit. Thus, the overall adjustment targets may be determined at a second time interval, greater than the first time interval, while each sub-adjustment target may be determined at a first time interval.

[0093] In another embodiment, a refinery may include a plurality of operation controllers. Each operation controller 302 may include a plurality of trained machine learning models. Each trained machine learning model may be trained to recognize a specific or selected trend in a set of data. Thus, each operation may be adjusted based on the outputs of a plurality of models, ensuring the operation as a whole produces an accurate target product. Further, each operation controller may interact with each other operation controllers. For example, adjusted parameters from all operation controllers may be provided to each operation controller. Thus, as one operation is adjusted, a downstream and / or upstream operation may be further adjusted based on the adjustment of the one operation.

[0094] FIG. 4 is a simplified diagram that illustrates the training of a machine learning model for enhanced fluid production at refinery, according to an embodiment of the disclosure. Each model described herein may be trained prior to use. Such training may be performed prior to use with a set of historical data specific to a refinery. In a further embodiment, a plurality of machine learning models may be trained, each based on data specific to an operation and selected equipment at the refinery.

[0095] As noted, the machine learning models described herein may be trained using data. As illustrated in FIG. 4, the data may include historical, equipment specific data 402. In other embodiments, the training data may include data related to the entire operation of arefinery, as well as outputs from equipment specific models. In another embodiment, a machine learning model may be re-trained and / or refined via current and marked up equipment specific data 404. The historical equipment specific data 402 and current and marked up equipment specific data 404 may include feed composition, feed properties, material composition, material properties, a target product or products, target composition, target properties, temperatures in equipment, pressure in equipment, flow rates associated with feed and / or materials, and / or equipment parameters. In embodiments, the historical equipment specific data 402 may include or may be utilized to generate a non-linear concave function. In such embodiments, the desired outcome may be determined based on the maximum of such a function. In another embodiment, the desired outcome may be included or added to the data set. In yet another embodiment, training may include the machine learning model learning particular patterns that indicate what the desired outcome may be based on trends within the data. In another embodiment, physics-based data may be provided along with the historical data set to ensure that outputs from a trained machine learning model remain consistent and / or emulate real process / actual possibilities. In yet another embodiment, a plurality of machine learning models may be trained for the same operation. Each of the plurality of machine learning models may utilize different portions of historical data and / or other inputs to cause the model to maximize a specific attribute or parameter. Another model may be trained to utilize the outputs of each of those plurality of models, in addition to data.

[0096] Once the historical data, and any other current data, is available, that data may be pre-processed 406. In such embodiments, the data may be normalized. In other words, outlying data points that are anomalies may be removed from the data set. Further, data corresponding to abnormal events may be removed, such as data generating during startup, shut-down, turn-arounds, maintenance, and / or upsets. Further, undesired data and invalid measurements may be removed. Finally, data may be segregated or separated into sequences based on time. The sequences may comprise data obtained over a consecutive time period, such as time intervals of 30 minutes, 1 hour, 2 hours, and / or 3 hours, or more or less than the time intervals listed. In another embodiment, other factors may be utilized to segregate or separate the data, such as feed used and / or target product being produced.

[0097] Once the data set has been pre-processed, a model may be trained 408. In embodiments, a portion of the data set (for example, 70%, 80%, or 90%) may be fed to the machine learning model. The machine learning model may utilize the inputs versus the known desired outcome (such as target product composition and properties) and / or knownundesired outcome to “learn” what parameters can be utilized to reach the known desired outcome and what parameters lead to the known undesired outcome. Once the data has been used to train the machine learning model, then the remaining portion of the data set may be utilized to test 410 the trained machine learning model. If the trained machine learning model does not meet or achieve a selected error rate, then trained machine learning model the trained machine learning model may be re-trained or refined with a different randomized portion of the data set, and the re-training repeated as necessary, until the selected error rate is met or achieved. In another embodiment, other training schema may be utilized. In another embodiment, readiness of the trained machine learning model may be determined based on how close the trained machine learning model comes to an expected outcome, based on the test data set.

[0098] Once the trained machine learning model 412 meets a selected error rate, then the trained machine learning model may be released for further use. In another embodiment, a separate step may include selection of a type of machine learning model prior to training of the machine learning model. In other embodiments, various types of models may be trained, then tested. The most accurate models, determined by an error rate for each model, may be utilized.

[0099] FIG. 5 is a schematic diagram of an enhanced hydrotreater control system and a distillation control system to enhance fluid production at a portion of a refinery, according to an embodiment of the disclosure. As illustrated in FIG. 5, a distillation / fractionation portion of the refinery may include a distillation controller 602 and / or a hydrotreater controller 608. The distillation controller 602 may obtain data related to a fractionation / distillation column 616. For example, the distillation controller 602 may obtain temperature and / or pressure within the fractionation / distillation column 616. Further, the distillation controller 602 may initiate collection of samples of various fluids associated with the fractionation / distillation column 616 via the sample collection and analysis assembly 614. For example, the sample collection and analysis assembly 614 may obtain samples of off-gas 624, LPG 628, naphtha / gasoline 630, diesel 640, naphtha, and / or slurry 620, among other fluids associated with the fractionation / distillation column 616. Sensor packages 618, 622, 626, 630, 636 may be disposed to measure the temperature, pressure, flow rate, and composition of the products from the fractionation / distillation column 616 including off gas 624, LPG 628, gasoline 634, diesel 640, and slurry 620.

[0100] Once the sample collection and analysis assembly 614 obtains the samples, the sample collection and analysis assembly 614 may analyze the samples to produceproperties or spectra indicative of the properties of each collected sample. The sample collection and analysis assembly 614 may provide the properties and / or spectra to the distillation controller 602. The distillation controller 602 may then apply the data, properties, and / or spectra to the to one or more training machine learning models associated with one or more of a local enhancement module 604 and / or predictive controls module 606. Based on the output of the trained machine learning models, which may indicate parameter and / or feed adjustment of the fractionation / distillation column 616, the distillation controller 602 may adjust the parameters and / or the feed via the local enhancement module 604 and, in some embodiments, an equipment and device control module. The equipment and device control module may comprise a PLC or DCS. In some embodiments, the equipment and device control module may comprise a DCS-PID module, controller, or circuitry.

[0101] In an embodiment, the distillation machine learning model 604 may be trained to maximize lift within the distillation column. Such a model may utilize temperature and / or feed flow rate to maximize such a parameter (in other words, to maximize lift or pressure). Further, such a model may predict parameters based on, in part, the concave objective function for profit to determine a feed and / or temperature input that drive the distillation column to include a maximum lift in a pressure limited column. The parameters for such an application of data to a trained machine learning model may include providing more energy and / or material to the column or increasing material or energy to increase pressure if all pressure control handles are exhausted. For example, the output from one of the trained machine learning models may indicate an increase in temperature and / or feed to maximize the output of a selected one or more products from the fractionation / distillation column 616. Such a maximization of the output may optimize profit for the fractionation / distillation column 616.

[0102] The fractionation / distillation column 616 may include a vacuum column. Such issues described above may also occur for a vacuum column when the vacuum tower may no longer be cooled (for example, due to cooling water limitations) or when a vacuum ejector system is limited and may no longer decrease pressure. The concave objective function in this case can be used to determine what feed and temperature in the column with uncontrolled pressure give the most profitable light vacuum gas oil (LVGO), middle vacuum gas oil (MVGO), heavy gas oil (HVGO) lift from resid.

[0103] In another example, the fractionation / distillation column 616 may comprise a vacuum column and / or crude unit. In such examples, one trained machine learning modelmay be trained or configured to utilize a resid viscosity, asphalt m-value, and / or other asphalt property as an input and / or, in other embodiments, sample analysis of the outputs and inputs of the vacuum column. Further, the trained machine learning model may utilize as an input, current valve settings, heater outlet temperature (for example a vacuum heater outlet temperature), and / or vacuum operating parameters. The output of such a model may include updated settings or set points for the valve settings, heater outlet temperature (for example a vacuum heater outlet temperature), and / or vacuum operating parameters. Such a trained machine learning model may minimize the amount of gas oil in the resid, when the resid is being utilized to make asphalt. The distillation controller 602 may utilize such an output to set equipment to parameters or settings output from the trained machine learning model.

[0104] The issues described above may also occur for an atmospheric vacuum column, where the feed and temperature in the column with uncontrolled pressure may be determined with a neural network model and controller to give higher lift of naphtha, distillate, and atmospheric gas oil from a crude column feed. For an atmospheric vacuum column, several constraints may be considered, such as product quality, hydraulic constraints, operating limits, and column differential pressure limits. Alternatively, problems associated with operation of the atmospheric vacuum column may be formulated to manipulate other variables such as pump-around return flow, pump-around return temperatures, and stripping steam that optimize the lift in the column when the column pressure is not being controlled.

[0105] In an embodiment, properties of feedstock utilized in a distillation operation may and / or other operations described herein may include boiling point, viscosity, composition, API gravity, distillation points, Coker gas oil content, carbon residue content, nitrogen content, sulfur content, saturates content, thiophene content, single-ring aromatics content, dual-ring aromatics content, triple-ring aromatics content, or quad-ring aromatics content. In another embodiment, fluids, target products, material, and / or unit materials produced by a fractionation / distillation column 616 may include one or more of an amount of butane- free gasoline, an amount of total butane, an amount of dry gas, an amount of coke, an amount of gasoline, octane rating, an amount of light fuel oil, an amount of heavy fuel oil, an amount of hydrogen sulfide, an amount of sulfur in light fuel oil, or an aniline point of light fuel oil. In another embodiment, properties of the fluids, target products, material, and / or unit materials produced by the fractionation / distillation column 616 may include one or more of pentane content, raw crude water content, desalted crude water content,heavy atmospheric gas oil (HAGO) content, light atmospheric gas oil (LAGO) flash, or kerosene flash point. In yet another embodiment, the distillation controller 602 may control the pentane content, the raw crude water content, the desalted crude water content, the heavy atmospheric gas oil (HAGO) content, the light atmospheric gas oil (LAGO) flash, or the kerosene flash point, one or more of: crude blend, make-up water, desalter severity, HAGO wash rate, stripping, LAGO draw rate, stripping steam, or kerosene draw of the one or more of the first processing units.

[0106] In yet another embodiment, the properties of the fluids, target products, material, and / or unit materials produced by the fractionation / distillation and absorber column 616 may include one or more of ethane content, propane content, propene content, isobutane content, or n-butane content. In such embodiments, the distillation controller 602 may control one or more of the ethane content, the propane content, the propene content, the isobutane content, or the n-butane content, one or more of: absorber pressure, lean oil flow rate, lean oil temperature, high-pressure separator temperature, reactor conversion, or stripper reboiler duty.

[0107] In yet another embodiment, the properties of the fluids, target products, material, and / or unit materials produced by the fractionation / distillation column 616 may include one or more of high-pressure separator water content or stripper bottoms water content. In such embodiments, the distillation controller 602 may control a temperature of a high- pressure separator.

[0108] In another embodiment, the fractionation / distillation column 616 may comprise a vacuum tower (in other words, a distillation tower operating under reduced pressure). In such embodiments, one of the trained machine learning models may infer or predict a micro-carbon residue (MCR) or HVGO MCR based on various parameters and / or feeds (such as, for example, wash rate, c-factor, bed distribution, lift drive-up entrainment, and / or feed properties, among others), enabling a controller (for example, the distillation controller 602) to determine adjustments to a vacuum distillation operation based on the inferred or predicted MCR. In another embodiment, rather than or in addition to utilizing HVGO MCR, the trained machine learning models may infer or predict nickel and / or vanadium (and / or other metals) content in HVGO. In yet another embodiment, rather than or in addition to utilizing HVGO MCR, the trained machine learning models may infer or predict a distillation cut point, a gas oil lift from resid, and / or HVGO wash bed lifecycle for HVGO production and / or to prevent premature shutdown.

[0109] In another embodiment, the trained machine learning model may also optimize deisopentanizer fractionation, such an optimization increasing octane on, in some embodiments, an isomerization unit. Such a trained machine learning model may be utilized to optimize DIP feed limit based on various factors, such factors being applied to the trained machine learning model. Further, such factors may include amount of steam utilized, temperature from a reboiler, a reflux rate, feed properties, and / or output properties.

[0110] In another embodiment, the distillation controller 602 (and / or, in embodiments, the local enhancement module 604 and / or predictive controls module 606) may include a trained machine learning model trained and / or configured to determine a salt point temperature for the fractionation / distillation column 616 (or, in some embodiments, a crude atmospheric distillation column). In such embodiments, the distillation controller 602 may input fractionation / distillation column 616 setpoints, valve setpoints, overhead temperature, and / or overhead reflux, among other factors, to the model. The trained machine learning model may output updates to the setpoints for the fractionation / distillation column 616 and / or valves, as well as temperature and / or overhead reflux setpoints. Such a trained machine learning model may prevent salt deposition in overhead piping and / or other downstream mechanical equipment. These salt depositions may lead to a loss of containment, premature damage of equipment, and / or premature equipment or plant shutdown.[OHl] Further, FIG. 5 illustrates a hydrotreater controller 608. The hydrotreater controller 608 may obtain data related to each hydrotreater 632 and 638. It will be noted that data obtained from each hydrotreater may be analyzed separately, as each hydrotreater performs a different function (for example, increase gasoline octane or remove sulfur and / or impurities from diesel). In yet another embodiment, the hydrotreater controller 608 may include a machine learning model specific for and trained for each specific hydrotreater, due to the changes each piece of equipment may experience over time.

[0112] The hydrotreater controller 608 may obtain data related to each hydrotreater and / or initiate capture of samples associated with each hydrotreater. The sample collection and analysis assembly 614 may then analyze the samples to produce properties or spectra indicative of properties of the fluids associated with the hydrotreater. Once the hydrotreater controller 608 obtains data, properties, and / or spectra related to a hydrotreater, the hydrotreater controller 608 may apply the data, properties, and / or spectra to a machine learning model within one or more of the local enhancement module 610 or the predictive controls module 612. The output of the hydrotreater machine learning model may indicateadjustments to parameters and / or feed associated with the hydrotreater. Using such an output, the local enhancement module 610 may adjust the parameters and / or feed associated with the hydrotreater, for example via a PID controller, DCS controller, PLC controller, or a DCS-PID controller.

[0113] In some embodiments, the local enhancement module 610 may include programming and an algorithm configured to facilitate optimization of the hydrotreaters 632, 638 to achieve the target parameters based on the inputs, the target parameters, unit constraints, and the outputs of the hydrotreaters 632, 638 from one or more machine learning models (e.g., of the predictive controls modules 612), and inputs. The algorithm may be modeled by a machine learning model and configured to facilitate optimization of the hydrotreaters 632, 638 based on a machine learning model. In some embodiments, the local enhancement module 610 includes an optimizer comprising an algorithm configured to achieve target parameters based on the optimization (e.g. maximization) of an objective function based on the outputs from the first machine learning models (e.g., of the predictive controls modules 612), the target parameters, unit constraints, and inputs.

[0114] In an embodiment, the machine learning model utilized in the hydrotreater controller 608 may be trained to maximize sulfur removal at a lowest possible temperature. In an example, as a targeted product’s sulfur level or amount becomes lower, the hydrotreater temperature setting increase becomes even higher to achieve the same amount of sulfur removal. Thus, the output of the machine learning model of the hydrotreater controller 608 may indicate adjustment of hydrotreater severity to achieve maximum aromatic saturation of hydrotreater feed against hydrotreater operation constraints. Such an output may comprise a vector or other list of values indicative of equipment and / or device parameter settings.

[0115] In an embodiment, one of the hydrotreaters may be a naphtha hydrotreater. The naphtha hydrotreater may be utilized, in conjunction with a distillate hydrotreater, to treat wild naphtha. The treated wild naphtha may be utilized by one or more different refinery equipment, such as, but not limited to, a hydrocracker, a crude unit, and / or a gasoline desulfurization unit. In such embodiments, one trained machine learning model may be trained or configured to maximize the amount of treated naphtha produced in relation to the amount of naphtha used in the refinery equipment. The trained machine learning model may adjust the kerosene or flash target in a crude column to increase or adjust distillate transported to the naphtha or distillate hydrotreater. Further, the trained machine learning model may also determine an amount of heavy coker naphtha to transport to distillatehydrotreaters to thereby produce a minimum amount of wild naphtha via the naphtha hydrotreater.

[0116] FIG. 6 is a schematic diagram of an enhanced catalytic reformer control system 700 to enhance fluid production at a portion of a refinery, according to an embodiment of the disclosure. As shown, the catalytic reformer system 700 receives feedstocks 701 from upstream distillation processes as illustrated in FIG. 14. The principal objectives of the catalytic reforming operation are to produce high octane gasoline blending component, produce hydrogen, and to produce aromatic hydrocarbons, including benzene, toluene, and xylenes for use in the chemical industry as solvents and as precursors for other products.

[0117] Upstream distillation processes may include one or more atmospheric distillation columns, splitters, separators, depentanizers, and other equipment capable of removing C5 and lighter hydrocarbons from the feedstocks 701. In some applications, the upstream distillation processes may be used to remove or limit the amount of C6 hydrocarbons to limit the formation and presence of benzene in the reformate of the catalytic reformer system 700. In general, the feedstocks are principally composed of C7 to C9 naphtha and alkanes, with the best feed being rich in naphthenes. Polynuclear aromatic hydrocarbons are coke precursors may deactivate the catalysts and may be removed during regeneration. Contaminants, including nitrogen, sulfur, and metals, should be minimized. Nitrogen strips chloride off reformer catalysts resulting in yield loss and forming salt deposits downstream of the reactor in other equipment if not removed by water wash. Sulfur is a temporary poison for reformer catalyst resulting in temporary yield loss. Higher reactor severity may be required to compensate for sulfur poisoning, which may result in more coke formation on the catalysts to meet a target octane number for a property of the reformate. Metals deposit on the catalysts and permanently block active catalyst sites resulting in deactivated catalyst.

[0118] The feedstocks 701 are fed into a feed heater 710. The feed heater 710 may include a surge drum (not shown) to facilitate control over the feed rate of feedstocks into the reactor 712. The feed heaters 710 preheats the feedstocks 701 in preparation for the catalytic reforming reaction within a first reactor 712. The feedstocks 701 are processed in each reactor 712, 716, 720 and may be preheated in feed heaters 710, 714, 718 before entering the reactors 712, 716, 720. Depending on the configuration, a catalytic reformer may have three to five reactors in series with each reactor having operational parameters to process a portion of the feedstocks into reformate. As shown, the one or more reactors 720 may represent a third reactor, a fourth reactor, or a fifth reactor. The catalytic reformersystem 700 disperses the processing of the feedstocks 701 across the reactors 712, 716, 720.

[0119] A machine learning model may be used to optimize the operational parameters of each reactor 712, 716, 720 to achieve a desired product property. Desired product properties include octane number of the reformate, octane-barrels of reformate, yield of reformate, and purity and quantity of hydrogen gas produced. Octane number may range from about 82 to 103 road octane number. Yields of reformate may range from about 66% to about 95% by volume. Hydrogen gas purity may range from about 0.2 to 0.4 recycle gas gravity. Reformate Reid vapor pressure may range from about 3 psi / 21 kpa to 6 psi / 41 kpa. In some applications, octane number may be obtained from a user, a customer specification, a machine learning model managing a gasoline blending pool, or determined by the machine learning model based on a gasoline pool target and the production rates of other components of the gasoline pool. In other applications, the machine learning model may manage each reactor to produce a maximum octane number constrained by a target reformate yield. In some applications, a target quantity or target quality of hydrogen gas may be used by the machine learning model to manage each reactor. In yet other applications, the machine learning model may manage octane-barrels constrained by a desired or target catalyst life. Further, a machine learning model may be programed to limit processing severity and cracking.

[0120] Each reactor 712, 716, 720 includes catalysts 732, 734, 736 to facilitate the isomerization and dehydrogenation of the feedstocks 701 into reformate 730. The catalysts 732, 734, 736 may be arranged in fixed beds or arranged as part of a continuous catalyst regeneration reactor. In some applications, the reactors 712, 716, 720 are semi-regenerative or cyclic regenerative. In applications where the reactors 712, 716, 720 are continuous catalyst regeneration reactors, the catalysts may be constantly moved through one of the reactors 712, 716, 720 to a catalyst regenerator.

[0121] In the presence of catalysts 732, 734, 736, the feedstocks are converted into isomerized, cyclic, and aromatic hydrocarbons. The catalysts 732, 734, 736 may include platinum, rhenium, tin, germanium, iridium, alumina, and chloride to provide hydrogenation and dehydrogenation functionality and to provide acid catalyzed isomerization functionality. Further, chloride may be injected into the reactors 712, 716, 720 or mixed into the feedstocks 701 to support the acid catalyzed isomerization function of the catalysts 732, 734, 736. In general, this reaction is endothermic and releases hydrogen gas as double bonds are formed in the hydrocarbons of the feedstocks 701.

[0122] In particular, the chemical reactions promoted by the catalysts 732, 734, 736 include dehydrogenation of cyclohexanes, dehydroisomerization of cyclopentanes, dehydrocyclization of paraffins, isomerization of paraffins, hydrocracking of paraffins, and dealkylation of aromatics. Additional reactions include desulfurization, denitrogenation, and deoxygenation of the feedstocks 701. The dehydrogenation of cyclohexanes is favored by high temperatures and low pressures and is a highly endothermic reaction. The dehydroisomerization of cyclopentanes is also favored by high temperatures and low pressures and is a highly endothermic reaction. The dehydrocyclization of paraffins is also favored by high temperatures and low pressures and is a highly endothermic reaction. The dehydrocyclization of paraffins is a slower reaction and may be favored by a lower feed rate of feedstocks. The hydrocracking of paraffins and the dealkylation of aromatics are promoted by high temperatures and are exothermic reactions resulting in reformate loss.

[0123] The process variables of the reactors 712, 716, 720 of the catalytic reformer system 700 include reactor pressures, reactor inlet temperatures, feed rate of feedstocks 701 (also referred to as liquid hourly space velocity), the molar ratio between hydrogen gas and hydrocarbons, the composition of the feedstocks 701, and catalyst life. Higher inlet temperatures favor dehydrocyclization and cracking resulting in higher aromatic production and higher octane number of the reformate, but lower yields due to the cracking and shrinkage due to the loss of saturation and formation of hydrogen gas. As the feedstocks undergo dehydrogenation and dehydrocyclization, hydrogen gas is produced and pressures increase toward the outlet of a reactor. Further, the molar ratio between hydrogen gas and hydrocarbons also changes throughout a reactor. Slower feed rates results in lower reactor temperatures due to higher residence times. Extremely low feed rates may result in higher cracking rates and a reduction in octane number of the reformate product.

[0124] The composition of the feedstocks directly affect product composition and yields. Contaminants and heavier, high boiling point hydrocarbons may deactivate the catalysts and C5 and lighter hydrocarbons take up space while not participating in the catalyzed reactions of the catalytic reformer. Further, paraffin reactions tend to be slower and result in a smaller proportion of reactions within a reactor, so feedstocks containing a high proportion of paraffins results in a lower octane and yield of reformate.

[0125] From the reactors 712, 716, 720, the products of the reactors is passed through a separator 722 that separates the hydrogen gas 724. The hydrogen gas 724 may include a recycling compressor and a net gas compressor. The recycling compressor compresses a portion of the hydrogen gas to be recycled and mixed with the feedstocks 701 to maintaina desired molar ratio of hydrogen gas to hydrocarbons. The remaining hydrogen gas may be passed through a net gas compressor that pressurizes the remaining hydrogen gas sent to other processes 780 within the refinery including hydrotreating processes or for transportation and sale.

[0126] From the separator 722, the remaining products of the reactors 712, 716, 720 may be passed through a stabilizer 726 where the light ends 728 are separated from the reformate 730. Stabilizer functionality directly affects the properties of the reformate as the light ends remaining in the reformate reduce the octane on the reformate and may negatively affect the Reid vapor pressure of the reformate.

[0127] Catalytic reformer constraints may include heat generation limits, firing rates, draft limits, and tube skin temperature limits of the heaters, including the feed heaters 710, 714, and 718. Catalytic reformer constraints may also include designed pressure and temperature limits of the reactors, as well as limits on the change of pressure across the catalysts within the reactor. Other constraints may include the horsepower, speed, amperage draw, and pressure head of the recycled hydrogen gas lines. Cooling limitations may be due to cooling water supply and availability, water temperature, and weather conditions. Cooling limitations affect separator 722 and stabilizer 726 operation. Other process constraints include the capacity of the stabilizer 726 to process large quantities of one group of hydrocarbons, fouling within the stabilizer, off gas and liquid line size capacities, and the rundown temperature of the stabilizer 726.

[0128] A section of the refinery corresponding to the catalytic reformer system 700 may include a catalytic reformer controller 702. The catalytic reformer controller 702 may include one or more machine learning models. Similar to previously described controllers, the catalytic reformer controller 702 may obtain data associated with the equipment of the reformer unit, such as reactors 712, 716, and 720, feed heaters 710, 714, 718, a separator 722, and stabilizer 726. The catalytic reformer controller 702 may also initiate capture of samples of fluids associated with the reformer and gather data from sensor packages 740, 742, 744, 746, 748, 750, 760, 762, 764, 766, 768, 770, 772, 774, 776 positioned throughout the catalytic reformer system 700. The sensor packages 740, 742, 744, 746, 748, 750, 760, 762, 764, 766, 768, 770, 772, 774, 776 may include temperature sensors; such as resistance temperature detectors, thermocouples, and infrared sensors; pressure sensors, such as hydrostatic pressure sensors and differential pressure sensors; flow rate sensors, such as coriolis flow meters, ultrasonic flow meters, and turbine flow meters for metering flow; density and composition analyzers, such as sulfur analyzers, refractometers, spectrometers,including near infrared spectrometry, and gas chromatographs; and sampling units for analyzing the composition of a sample through testing, distillation, or obtaining physical samples from the feedstocks 701 to be analyzed in a lab or by the sample collection and analysis assembly 708.

[0129] In particular, sensor packages, 740, 744, 748 may be used to determine inlet temperatures of the reactors 712, 716, 720 and a weighted average inlet temperature of the catalytic reformer system 700. Sensor packages 742, 746, 750 may be used to determine outlet temperatures of the reactors 712, 716, 720. Together, sensor packages, 740, 744, 748 and sensor packages 742, 746, 750 may be used to determine the pressure drop across the reactors 712, 716, 720 and the weighted average bed temperature of the catalysts 732, 734, 736 of the reactors 712, 716, 720. Sensor packages 760, 762, 764, 766, 768 may be used to measure, analyze, and sample feedstocks 701 and reactor products to generate feedstock data and product data going into and out of the reactors 712, 716, 720. Sensor packages 770, 772, 774, 776 may be used to measure, analyze, and sample the products of the catalytic reformer system 700.

[0130] The sample collection and analysis assembly 708 may then analyze the samples and produce properties and / or a spectra for each sample. The catalytic reformer controller 702 may apply the data, properties, and / or spectra to one or more machine learning models of the local enhancement module 704 and / or predictive controls module 706 to produce an output indicative of adjustment to parameters and / or feed. The catalytic reformer controller 702 may then utilize the output to adjust various parameters and / or feed associated with the reformer via the local enhancement module 704, and, for example, further via a PID controller, DCS controller, PLC controller, or a DCS-PID controller.

[0131] In applications where continuous catalyst regeneration reactors are used, a catalyst regenerator 752 is disposed to receive and regenerate the catalysts 732, 734, 736 of the reactors 712, 716, 720. Specifically, the catalysts 732, 734, 736 move serially through the reactors from reactor 712 to reactor 716 to reactor 720 and then into the catalyst regenerator 752. The regenerated catalysts may then be fed into reactor 712 to restart the cycle.

[0132] In some applications, a machine learning model may use data gathered from one or more of the sensor packages 740, 742, 744, 746, 748, 750, 760, 762, 764, 766, 768, 770, 772, 774, 776 to generate a prediction of remaining catalyst life or current state of catalyst deactivation. This prediction may be used by the machine learning model to adjust operational parameters of the feed heaters 710, 714, 718, the reactors 712, 716, 720, the separator 722, the stabilizer 726, and where applicable, the catalytic regenerator 752. Themachine learning model may also adjust the operational parameters of the catalytic regenerator 752 in order to better regenerate the catalysts 732, 734, 736 as they progress through the reactors 712, 716, 720. Operating parameters of the catalytic regenerator 752 include temperatures within the catalytic regenerator 752 and feed rate of catalysts 732, 734, 736 through the reactors 712, 716, 720 and the catalytic regenerator 752. Temperature and feed rate may be adjusted to ensure coke deposits are adequately removed during regeneration.

[0133] The separator 722 may be managed by a machine learning model to adjust and control the operating pressure within the reactors 712, 716, 720. As the separator 722 removes hydrogen gas from the processed feedstock 701, pressure may be reduced throughout the system with the highest pressures within the first reactor 712. By reducing pressure within the reactors 712, 716, 720, the reformation process and production of hydrogen are encouraged and supported.

[0134] In an example, the catalytic reformer controller 702 may optimize the reformer process based on a number of factors. One factor may include octane-barrels versus reactor temperature. The octane-barrels (in other words, the change in octane above a base value multiplied by the number of product barrels) vs reactor severity (reactor operating temperature) is a concave function with octane-barrels passing through a maximum “optimal” point. Thus, in an embodiment, the model may be trained to maximize or be utilized for maximizing that point on the curve. Similarly, the octane number of reformate versus reactor severity also follows a similar concave function with octane number passing through a maximum “optimal” point.

[0135] The machine learning model may receive gasoline blending pool data and use the gasoline blending pool data to generate an octane barrels target or octane number target for reformate to be produced. The gasoline blending pool data may show a need for a specific reformate properties to be produced by the catalytic reformer to meet blending pool requirements or customer specifications. For example, reformate having a high octane number may be used to offset the lower octane of gasoline produced through hydrocracking in order to meet customer specifications. In turn, this demand for reformate having a target property may be used by the machine learning model to generate target reactor’s operating parameters including operating temperatures of the catalytic reformer reactor. Further, the machine learning model may use a predicted state of the catalysts of the reactor to further modify the generated operating parameters resulting in greater reactor severity. The machine learning model may also publish a notification that based on reactor conditionsand the target property of reformate is not possible. The machine learning model may also publish what target property may be possible.

[0136] In an embodiment, one of the machine learning models may determine an amount of sweet naphtha to transport to the reformer in relation to an amount of sweet naphtha sent for blending. Such a model may be based on a total amount of sweet naphtha produced (for example, produced by a naphtha hydrotreater and / or hydrocracker) and an amount of sweet naphtha for a blending process in relation to an output of the reformer. A machine learning model may control for one or more target product properties of the reformate. Target product properties may include octane number, octane-barrels, benzene limits, Reid vapor pressure, end point (or boiling point of the reformate), and composition, especially concentrations of benzene, toluene, and xylene. A machine learning model may also identify adjustments to operational parameters to improve target properties and make recommendations to a user of the catalytic reformer. A machine learning model may recommend or direct upstream distillation operational parameter adjustments to increase the value of the products of the catalytic reformer.

[0137] The machine learning model may also generate or assign values for the feedstocks, products, and operational costs of the catalytic reformer system 700 based on business information. Business information may include the costs of operation, feedstock costs, and the internal and market prices of hydrogen gas, reformate, and other products from the catalytic reformer system 700. Feed stock costs may include the costs of distillation and hydrotreating the feedstocks in preparation for processing by the catalytic reformer. The values may be used by the machine learning model with an objective function to increase the overall value of the operation of the catalytic reformer system 700.

[0138] The machine learning model may also obtain values from an aromatic recovery unit to better increase and determine the value of the products produced by the catalytic reformer system 700. In some applications, the machine learning model may determine operational parameters to maximize reformate formation while minimizing cracked hydrocarbons as light ends. In other applications, the machine learning model may determine operational parameters to maximize yield while maintaining a lower target octane number. In yet other applications, the machine learning model may manage the catalytic reformer to maximize octane barrels of reformate. Alternatively, the machine learning model may determine operational parameters to maximize octane number while maintaining a lower yield target. The octane number or octane barrels to be produced by the catalytic reformer may be determined by an operator of the gasoline blending pool ormay be generated by the machine learning model based on the needs and requirements of the gasoline blending pool. In other words, gasoline blending pool requirements may be used to drive reactor severity in order to achieve a desired octane number or octane-barrels of the reformate. In some configurations, a second machine learning model managing the data of the gasoline blending pool may publish an octane number and quantity demand to the machine learning model over the catalytic reformer. The machine learning model over the catalytic reformer may generate a target octane number, a target yield, and may drive the operating parameters of the catalytic reformer to achieve the target octane number and the target quantity. In some applications, targets for reformer operation may be driven by reformate properties needed to satisfy a specification for a gasoline blend. For example, the machine learning model may provide target product properties to the catalytic reformer to achieve an octane number and yield, or octane-barrels needed to satisfy the finished gasoline properties for a gasoline tank or gasoline blending pool. This gasoline tank or blending pool may also be taking finished gasoline from other gasoline-producing units in the refinery.

[0139] The machine learning model over the catalytic reformer may also publish a request to an upstream process such as a distillation operation to adjust a cutoff of the feedstocks to change the composition of the feedstocks to increase the reformate yield and octane number of the catalytic reformer. The adjustments may be further adjusted based on a generated prediction of a state of the catalysts within the reactors of the catalytic reformer system 700.

[0140] In some applications, the machine learning model may reduce system pressure close to an operational constraint, which may include a compressor limit or a valve limit, in order to maximize reformate yield. The machine learning model may generate a target tolerance range or a single target point for the catalytic reformer to operate at to avoid exceeding the operational constraint. For example, the machine learning model may generate the target tolerance range or target point based on the catalytic reformer’s historical data demonstrating the catalytic reformer’s ability to respond to disturbances during steady state operation. For example, if the operational constraint is 55 psig / 39 meters of head, then the machine learning model may drive the pressure of a reactor within the catalytic reformer to a range of 55 psig / 39 meters of head to 57.75 psig / 40.6 meters of head or 5% of the operational constraint based on the catalytic reformer’s historical data demonstrating the catalytic reformer’s ability to respond to disturbances during steady state operation. Alternatively, the target tolerance range or a single target point may be user defined or arecommendation from the machine learning model may be implemented by a user. In some applications, the target may be within 5% of the operational constraint, while in others, the target may be within 1% of the operational constraint.

[0141] FIG. 7 is a schematic diagram of an enhanced aromatics recovery control system to enhance fluid production at a portion of a refinery, according to an embodiment of the disclosure. A section of the refinery corresponding to an aromatics recovery unit 1100 may receive reformate 730 from the catalytic reformer system 700 of FIG. 6. As shown, the reformate 730 may be fed into a reformate splitter 1102, where the reformate 730 may be separated into a light reformate 1104 comprising C5 and smaller hydrocarbons, a heavy reformate 1106 comprising C9 and heavier hydrocarbons, and a reformate cut rich in benzene, toluene, and xylene (the “ARU cut 1110”). The reformate splitter 1102 may be one or more distillation columns, strippers, separators, depentanizer, or other separation process that may be used to target the desired cut rich in benzene, toluene, and xylene. The properties and composition of the ARU cut 1110 may be determined by a sensor package 1108.

[0142] From the reformate splitter 1102, the ARU cut 1110 is a feedstock that is passed into the extractor column 1112 where a solvent 1114, such as sulfolane, is used to dissolve the benzene, toluene, and xylene from the ARU cut 1110. The rich solvent 1116 is then fed from the bottom of the extractor column 1112 to the stripper 1118. The properties and composition of the rich solvent 1116 may be measured, analyzed, and sampled by a sensor package 1120.

[0143] The paraffinic naphtha of the ARU cut 1110 rises through and exits the extractor column 1112. The paraffinic naphtha of the ARU cut 1110 may be mixed with solvent as a mixture 1122. The mixture 1122 is passed through the raffinate wash tower 1124 where the cleaned paraffinic naphtha is referred to as raffinate 1126. The properties and composition of the mixture 1122 and the raffinate 1126 may be measured, analyzed, and sampled respectively by a sensor package 1128 and a sensor package 1130. Washed solvent 1132 separated from the raffinate 1126 in the raffinate wash tower 1124 is passed into a stripper reboiler 1134.

[0144] The operating parameters of the extractor column 1112 include temperature, pressure, solvent to hydrocarbon ratio, solvent purity, and flow rates of solvent 1114, the ARU cut 1110, flow from the stripper reboiler, and stripper recycle 1136. Higher temperatures reduce viscosity of the solvent but also reduce the selectivity of the solventso more non-aromatic hydrocarbons may be dissolved in the solvent. Pressure maintains the solvent and feed in the liquid phase.

[0145] Solvent to hydrocarbon ratio is dependent on the flow rates of solvent 1114, the ARU cut 1110, flow from the stripper reboiler, and stripper recycle 1136 and may be altered to minimize solvent loss in the raffinate mixture 1122 as well as aromatic extraction efficiency. In some applications, the physical constraints of the extractor column 1112 may be reached where the flow rate of the ARU cut 1110 is too high which prevents the solvent from properly flowing downward through the extractor column 1112. This condition may be resolved by reducing the flow rate of the ARU cut 1110. A machine learning model may be used to determine increased or maximum flow rates to reduce or prevent exceeding operational constraints that may cause a loss in production or product quality. In some applications, the machine learning model may determine operating thresholds below the operational constraints based on the determined ability or confidence level of a specific aromatic recovery unit 1100 to avoid exceeding the operational constraints.

[0146] The stripper 1118 separates light paraffins and aromatics as stripper recycle 1136 from the rich solvent 1116 and feeds the stripper recycle 1136 back into the extractor column 1112. The solvent 1114 has a preference for lighter aromatics, and so as the stripper recycle 1136 dissolves in the solvent 1114, heavier aromatics are replaced and separate from the solvent. Consequently, some of the C9+ aromatics exit from the top of the extractor column 1112 as raffinate 1126. The stripper recycle 1136 may be measured, analyzed, and sampled by a sensor package 1138. The rich solvent 1140 exits the stripper 1118 and may be measured, analyzed, and sampled by a sensor package 1142. The rich solvent 1140 is then passed into the solvent recovery column 1148.

[0147] The stripper reboiler 1134 separates the washed solvent 1132 into solvent 1144 and hydrocarbons 1146 that may have been dissolved in the solvent 1144. The solvent 1144 may be passed into the solvent recovery column 1148. The hydrocarbons 1146 may be fed back into the extractor column 1112. The solvent 1144 may be measured, analyzed, and sampled by a sensor package 1150. The hydrocarbons 1146 may be measured, analyzed, and sampled by a sensor package 1152.

[0148] The solvent recovery column 1148 separates a benzene, toluene, xylene, and heavier aromatic mixture 1154 from the rich solvent 1140. The operating parameters of the solvent recovery column 1148 are temperature and stripping steam rate. The top temperature is controlled to separate the benzene, toluene, xylene, and heavier aromaticmixture 1154 from the solvent. Stripping steam is used to reduce the partial pressure of the aromatics and aids in aromatic separation from the solvent.

[0149] The benzene, toluene, xylene, and heavier aromatic mixture 1154 is sent to benzene, toluene, and xylene towers (the “BTX towers 1158”) where the benzene, toluene, xylene, and heavier aromatic mixture 1154 is separated into purified benzene 1160, toluene 1162, xylene 1164, and C9+ aromatics 1167. The benzene, toluene, xylene, and heavier aromatic mixture 1154 may be measured, analyzed, and sampled by a sensor package 1166. The purified benzene 1160, toluene 1162, and xylene 1164 may be measured, analyzed, and sampled by sensor packages 1168, 1170, 1172.

[0150] The resulting lean solvent 1114 is sent back to the extractor column 1112 to repeat the cycle. The lean solvent 1114 may be measured, analyzed, and sampled by a sensor package 1174. As the solvent 1114 degrades, the solvent 1114 may be sent to the solvent regenerator 1156. The regenerated solvent 1176 may be measured, analyzed, and sampled by a sensor package 1178. The solvent 1144 from the stripper reboiler 1134 may also be processed in the solvent recovery column 1148.

[0151] Operational data includes the operational parameters of each process of the aromatics recovery unit 1100 may be sent to an aromatics recovery controller 1180. The aromatics recovery controller 1180 controls each of the processes and may adjust the operational parameters of each process of the aromatics recovery unit 1100. Similar to previously described controllers, the aromatics recovery controller 1180 may obtain data associated with the equipment of the aromatics recovery unit. The aromatics recovery controller 1180 may also initiate capture of samples of fluids associated with the aromatics recovery unit through one of the sensor packages 1108, 1120, 1128, 1130, 1138, 1142, 1150, 1152, 1166, 1168, 1170, 1172, 1174, 1178. Each sensor package 1108, 1120, 1128, 1130, 1138, 1142, 1150, 1152, 1166, 1168, 1170, 1172, 1174, 1178 may include one or more temperature sensors; such as resistance temperature detectors, thermocouples, and infrared sensors; pressure sensors, such as hydrostatic pressure sensors and differential pressure sensors; flow rate sensors, such as coriolis flow meters, ultrasonic flow meters, and turbine flow meters for metering flow; density and composition analyzers, such as sulfur analyzers, refractometers, spectrometers, including near infrared spectrometry, and chromatographs; and sampling units for analyzing the composition of a sample through testing, distillation, or obtaining physical samples from the aromatics recovery unit 1100 to be analyzed in a lab or by the sample collection and analysis assembly 1182.

[0152] The sample collection and analysis assembly 1182, aromatics recovery controller 1180, or a machine learning model may receive data from the sensor packages 1108, 1120, 1128, 1130,1138, 1142, 1150, 1152, 1166, 1168, 1170, 1172, 1174, 1178 and analyze the data for each sample, process, and material moving through the aromatics recovery unit 1100. The data may be used by the sample collection and analysis assembly 1182, the aromatics recovery controller 1180, or the machine learning model to determine or predict one or more properties, component concentration, or composition of the material. The aromatics recovery controller 1180 may apply the data, properties, and / or spectra to one or more machine learning models of the local enhancement module 1184 and / or predictive controls module 1186 to produce an output indicative of adjustment to parameters and / or feed to improve the process or prevent larger deviations in operational parameters. The aromatics recovery controller 1180 may then utilize the output to adjust various parameters and / or feed associated with the aromatics recovery unit via the local enhancement module 1186.

[0153] A machine learning model may be trained with the historical data of the aromatics recovery unit 1100. The historical data may include feedstock data indicative of one or more of a property or composition the reformate and purchased feedstocks that may be fed into the aromatics recovery unit 1100. The historical data may also include operational data indicative of the operational parameters of each process of the aromatics recovery unit 1100, as well as, sensor data gathered from sensors disposed throughout the aromatic recovery unit 1100. Operational data may also include data related to incidents or when operational constraints of the aromatics recovery unit 1100 have been reached. Operational data may also include indicative of solvent flow and recycle rates, solvent condition, solvent regeneration parameters, and rates of new solvent being added to the aromatics recovery unit 1100. The historical data may also include product data indicative of one or more properties or composition of products of a process of the aromatics recovery unit 1100.

[0154] The machine learning model may be used to predict the effects of adjustments and implementation of adjustments to the extractor column 1112. In particular, the extractor column 1112 is a full liquid operation with recycle loops of solvent 1114 and stripper recycle 1136, which may result in significant delays in seeing the actual effects of changes to the operational parameters of the extractor column 1112. In particular, the delay in seeing the effect on operations may be longer than a change in feed quality. Thus, using a machinelearning model to predict the effects of adjustments to operational parameters may permit improved operation and consistency in product quality in spite of changing feed quality.

[0155] Constraints may exist for each process of the aromatic recovery unit 1100. For example, a raffinate flow or processing limit or a fractionation limit in one of the BTX towers 1158 may determine boundaries for upstream and downstream variables of the aromatic recovery unit 1100. Consequently, improvements to the operation of the aromatic recovery unit 1100 may be determined upstream by producing better reformate at the catalytic reformers. For example, a machine learning model may seek to optimize C6 to C8 aromatic production to improve the feed quality to the aromatic recovery unit 1100. The machine learning model may improve distillation of feedstocks and thus, feedstock quality being provided to the catalytic reformers and the aromatic recovery unit 1100.

[0156] Solvent loading, in an aromatics recovery unit, may determine how much aromatics can be absorbed in the solvent. Solvent loading may also be affected by solvent quality, feed quality, and tray fouling. Solvent quality may be affected by a result oxygen contamination, excessive reboiler temperatures, and regeneration system health. Sulfolane solvent may include an affinity for smaller molecules over larger molecules (for example, C6>C7>C8>C9). Further, in some embodiments, a feed composition analyzer may not be available. As such, the trained machine learning model of the aromatics recovery controller 1180 may maximize aromatics production and increase product purity as a result of feed composition prediction. The aromatics controller may determine the maximum feed target for the aromatics recovery unit that can meet product quality constraints.

[0157] A machine learning model is trained on historical data including operational data, product data of the aromatics recovery unit, and feedstock data indicative of feedstock fed into the aromatics recovery unit. The machine learning model may be applied to the aromatics recovery unit to assist in improving the processes, increasing product yield, improving a property or composition of a product. To do so, the machine learning model may request adjustments to the upstream processes including distillation units and catalytic reformer units to improve or change feedstocks fed into the aromatics recovery unit.

[0158] The machine learning model may operate by receiving one or more of operational data of an aromatics recovery unit indicative of operational parameters of the aromatics recovery unit and including sensor data from one or more sensors disposed to measure a parameter of the aromatics recovery unit, product data indicative of a property or component of a product from a process of the aromatics recovery unit, or feedstock data ofa reformate being fed into the aromatics recovery unit and using this data to generate a prediction of a process constraint based on the operating data.

[0159] Alternatively, the machine learning model may use the data to determine a change has occurred in a process of the aromatics recovery unit. The machine learning model may generate a prediction of the effect of the change on the operation of the aromatics recovery unit. The machine learning model may generate an adjustment to an operating parameter of the aromatics recovery unit based on the predicted process constraint and / or the predicted effect of the change. The machine learning model may generate an adjustment to an operating parameter of the aromatics recovery unit based on the predicted process constraint, predicted effect of the change, or based on the potential improvement of current operating parameters.

[0160] The machine learning model may publish one or more of the predicted process constraint, the predicted effect of the change, or the generated adjustment to a user. A user may instruct the machine learning model to implement the adjustment or the machine learning model may have authority to implement the adjustment. Alternatively, the user may request additional information including information that was used by the machine learning model to generate the predicted process constraint, the predicted effect of the change, or the generated adjustment.

[0161] The adjustments may include an adjustment to a feed rate of feedstocks including the reformate from an upstream catalytic reformer. In some applications, the machine learning model may generate a plurality of simulations of the operation of the aromatics recovery unit with each simulation having a different adjustment to an operational parameter of the aromatics recovery unit. Then the machine learning model may select a simulation of the plurality of simulations. The machine learning model may select the simulation with a highest production of one or more of benzene, toluene, or xylene while not exceeding an operational constraint of the aromatics recovery unit, or a simulation that operates a margin away from operational constraints of the processes of the aromatics recovery unit. Once selected, the adjustment to the parameter of the aromatics recovery unit may be based on the selected simulation.

[0162] The machine learning model may generate a prediction of solvent loading in an extractor column of the aromatics recovery unit based on the operating data and the product data. The prediction of solvent loading may be used to generate the adjustment to the operational parameter of the aromatics recovery unit. The adjustment may be an adjustment to a feed rate of solvent into an extractor column of the aromatics recovery unit or anadjustment to an operating parameter of a solvent recovery column of the aromatics recovery unit.

[0163] Alternatively, the machine learning model may generate a prediction of solvent quality based on the operating data and the product data. The prediction of solvent quality may be used to generate an adjustment to an operational parameter of the aromatics recovery unit. The adjustment may include an adjustment to the operational parameters of the solvent regenerator, the solvent recovery column, the extractor column, the stripper, or a request for fresh solvent.

[0164] The machine learning model may periodically, or when triggered by a change in the feedstock, products, or operational parameters of the aromatics recovery unit, generate a prediction of a process constraint based on the operating data. The prediction may change over time and be a prediction of fouling within equipment of a process of the aromatics recovery unit. When a process constraint is determined, the machine learning model may generate a margin away from the process constraint based on the historical data and a confidence interval that the process of the aromatics recovery unit will not exceed the process constraint during operation of the aromatics recovery unit. The margin may be used to prevent problems from arising based on the operation exceeding an operational constraint. Consequently, the machine learning model may use the margin to constrain the generated adjustment to prevent operation of the aromatics recovery unit from exceeding an operational constraint. In some application, the operational constraint may be an orifice size that causes a backup in flow because the flow rate is too high to move through the orifice.

[0165] The machine learning model may also use the generated process constraints to generate operating targets based on the margin or the constraint. The operating target may represent a maximum flow rate of feedstock into the aromatics recovery unit with a predetermine confidence level that the operational constraints of the aromatics recovery unit will not be exceeded. The generated adjustment may move operation of the process to the operating target.

[0166] FIG. 8 is a schematic diagram of hydrogen control system 1800 to enhance fluid production at a portion of a refinery, according to an embodiment of the disclosure. Similar to steam, several refinery operations may utilize hydrogen for various purposes. Further, a refinery may also produce some amount of hydrogen, either as a main product and / or as a by-product. Thus, hydrogen management may occur at the refinery level, rather than at a sub-level. Similar to previously described controllers, the hydrogen controller 1802 mayobtain data associated with the equipment that produces hydrogen and / or utilizes hydrogen, such as from a steam methane reformer 1808, a catalytic reformer 1809, a hydrolysis unit 1810, and external hydrogen sources 1812 (such data including an amount of hydrogen available at a selected times), and / or refinery equipment 1814 that utilizes hydrogen. The hydrogen controller 1802 may also initiate capture of samples of fluids associated with the devices and / or equipment that produce and / or utilize hydrogen. The sample collection and analysis assembly (not illustrated) may then analyze the samples and produce properties and / or a spectra for each sample. The hydrogen controller 1802 may apply the data, properties, and / or spectra to one or more machine learning models of the local enhancement module 1804 and / or predictive controls module 1806 to produce an output indicative of adjustment to parameters and / or feed. The hydrogen controller 1802 may then utilize the output to adjust various parameters and / or feed associated with the hydrogen production units and / or equipment or devices utilizing hydrogen via the local enhancement module 1806.

[0167] The hydrogen controller 1802 may prioritize hydrogen to units or portions of the refinery that are generating high value products and / or products or intermediaries that are utilized to produce high value products, as well as maximize purity of hydrogen, such as via methane reforming and / or hydrolysis. As such, at least one machine learning model for hydrogen management may maximize hydrogen production or acquisition for one or more different refinery operations, while providing sufficient hydrogen to other refinery operations that utilize hydrogen.

[0168] FIG. 9 is a schematic diagram of a feed control system 1900 to enhance fluid production at a portion of a refinery, according to an embodiment of the disclosure. Each refinery operation utilizes a feed stream or feedstock. Such a feed stream or feedstock may, in some examples, be adjusted, to optimize or enhance fluid production of a corresponding refinery operation. Similar to previously described controllers, the feed controller 1902 may obtain data associated with the equipment that produces a feed or intermediary, blends a feed or intermediary, and / or utilizes a feed and / or intermediary, such as from one or more blend tanks 1908, one or more in-line blend pipes 1910, one or more feedstock sources 1912, one or more intermediary sources 1914, one or more sample collection and analysis assemblies 1916, and / or refinery equipment 1918. The feed controller 1902 may also initiate capture of samples of fluids associated with the devices and / or equipment that produce, blend, and / or utilize feed. The sample collection and analysis assembly (not illustrated) may then analyze the samples and produce properties and / or a spectra for eachsample. The feed controller 1902 may apply the data, properties, and / or spectra to one or more machine learning models of the local enhancement module 1904 and / or predictive controls module 1906 to produce an output indicative of adjustment to parameters and / or feed. The feed controller 1902 may then utilize the output to adjust various parameters and / or blends of a feed or intermediary associated with the refinery equipment 1918 via the local enhancement module 1904. The trained machine learning model utilized for the feed controller 1902 may be trained to or be utilized to predict feed, blend, and / or intermediary properties and / or components that produce a maximum or greater than typical yield for a particular refinery operation.

[0169] FIG. 10 and FIG. 11 are simplified diagrams of control systems 2700 to enhance to enhance fluid production at refinery, according to an embodiment of the disclosure. As noted, control system 2700 may include an operation controller 2701. Further, the operation controller 2701 may connect to one or more sensors 2716A, 2716B, and up to 2716N, one or more devices 2718 A, 2718B, and up to 2718N (such as flow control devices and / or temperature control devices), one or more equipment 2720A, 2720B, and up to 2720N, one or more analyzers 2722A, 2722B, and up to 2722N, and one or more predictive controls 2714A, 2714B, and up to 2714N. The operation controller 2701 may include memory 2704 and one or more processors 2702. The memory 2704 may store instructions executable by one or more processors 2702. In an example, the memory 2704 may be a non-transitory machine-readable storage medium. As noted, the memory 2704 may store or include instructions executable by the processor 2702.

[0170] As used herein, “signal communication” refers to electric communication such as hardwiring two components together or wireless communication, as understood by those skilled in the art. For example, wireless communication may be Wi-Fi®, Bluetooth®, ZigBee, or other near-field communications. In addition, signal communication may include one or more intermediate controllers or relays disposed between elements in signal communication.

[0171] The memory 2704 may include or store sample and data collection and instructions 2706. Upon execution of such instructions, the operation controller 2701 may obtain samples associated with each equipment 2720A, 2720B, and up to 2720N. Further, the operation controller 2701 may obtain data from the one or more sensors 2716A, 2716B, and up to 2716N and / or one or more flow control devices 2718 A, 2718B, and up to 2718N. Upon collection of the samples, the operation controller 2701 may send the sample to one of the one or more analyzers 2722A, 2722B, and up to 2722N. The one of the one or moreanalyzers 2722A, 2722B, and up to 2722N may then analyze the sample and generate properties and / or a spectra.

[0172] The operation controller 2701 may connect to and receive data from the one or more predictive controls 2714A, 2714B, and up to 2714N. In an embodiment, the operation controller 2701 may receive the output from each trained machine learning model of each the predictive controls 2714A, 2714B, and up to 2714N. In an embodiment, the output may comprise a vector or, in other embodiments, a value indicative of a parameter adjustment.

[0173] The memory 2704 may include or store trained machine learning models 2708. The trained machine learning models 2708 may include at least one trained machine learning model to generate an output indicative of parameter and / or feed adjustment. The operation controller 2701 may apply the data, properties, spectra, and / or the output of each trained machine learning model from one or more predictive controls 2714A, 2714B, and up to 2714N to the trained machine learning models 2708 to generate an output indicative of parameter adjustments and / or feed adjustment.

[0174] The memory 2704 may include or store parameter adjustment instructions 2710. Upon generation of the output, the operation controller 2701 may adjust parameters associated with equipment at the refinery. Further the memory 2704 may include or store feed adjustment instructions 2712 to adjust feed based on the output.

[0175] In FIG. 11, predictive controls 2714 may connect to subsets of each of the components described in FIG. 10. For example, the predictive controls 2714 may connect to a subset of the sensors 2736A, 2736B, and up to 2736N, a subset of the devices 2738A, 2738B, and up to 2738N, a subset of the equipment 2740A, 2740B, and up to 2740N, and / or a subset of the analyzers 2742A, 2742B, and up to 2742N. The predictive controls 2714 may include a trained machine learning model 2728 and instructions stored in a memory 2726 and executable by a processor 2724.

[0176] The instructions may include sample and data collection instructions 2730, which when executed cause the predictive controls 2714 to collect various data points and / or properties. Based on application of the data received to the trained machine learning model 2728 and, in some embodiments, an output from the operation controller 2701, the predictive controls 2714 may supply or provide the output to the operation controller 2701.

[0177] FIG. 12 is a flow chart illustrating enhanced fluid production at a refinery, according to an embodiment of the disclosure. Unless otherwise specified, the actions of method 2800 may be completed within an operation controller and / or predictive controls. Specifically, method 2800 may be included in one or more programs, protocols, or instructions loadedinto the memory 2704 of operation controller 2701 and executed on the processor 2702 or one or more processors of the operation controller 2701 of FIG. 10 or FIG. 11. In other embodiments, method 2800 may be implemented in or included in components of FIGS. 1 A-24. The order in which the operations are described is not intended to be construed as a limitation, and any number of the described blocks may be combined in any order and / or in parallel to implement the methods.

[0178] At block 2802, the one or more predictive controls may each obtain data from corresponding sources. In such an example, each of the predictive controls may poll corresponding sensors, flow control devices, equipment, temperature control devices, and / or other data generating sources related to a corresponding refinery unit to obtain data therefrom. Further, the predictive controls may initiate sample collection for inputs or feedstock, as well as intermediaries and / or products produced by the corresponding source. Such a sample may be analyzed by one or more spectroscopic analyzers, which may subsequently produce properties and / or spectra of the samples.

[0179] At block 2804, once each sub-controller has obtained data, each sub-controller may apply that data, which may also include one or more different properties and / or spectra, to a corresponding machine leaning model stored therein. In another embodiment, the targeted product and / or targeted properties may be applied, in addition to the data described above, to the machine learning model of the predictive controls. The machine learning model may produce vectors, indicators, and / or other values indicative of one or more parameters that may cause the corresponding source to produce a targeted product. Each of the parameters output from the trained machine learning model may correspond to some aspect of the source. For example, the parameters may include temperature, amounts of other fluids used in the operation (for example, hydrogen or alkanes, among others), pressure, flow rate, residence time, feedstock used, and / or intermediaries used.

[0180] At block 2806, each of the predictive controls may determine whether the subparameters are different than currently set parameters. If the parameters are different, then, in some embodiments and at block 2808, the predictive controls may adjust one or more of the equipment, devices, or fluids. In such embodiments, each of the predictive controls may provide the adjustments to an equipment and device controller. In another embodiment, the predictive controls may provide the adjustments to the operation controller 2701 (FIG. 10 and FIG. 11) to perform such adjustments. In another embodiment, rather than or in addition to adjusting the equipment or devices, the predictive controls may provide the trained machine learning model output to the operation controller 2701.

[0181] At block 2810, the operation controller 2701 (FIG. 10 and FIG. 11) may determine updated parameters and / or fluid compositions and / or ratios based on application of the obtained data, as well as the outputs from each of the predictive controls, to a trained machine learning model. The operation controller 2701, in some embodiments, may first obtain data from the all, substantially all, a portion of the refinery, or from a corresponding operation. Data, as noted above, may include data from sensors, meters, equipment, and / or other devices, as well as properties and / or spectra obtained from samples taken from each unit within the refinery. Further, the operation controller 2701 may obtain the output of each model of each predictive controls. Once the operation controller 2701 obtains all relevant data, the operation controller 2701 may determine the updated parameters based on application of that data to a machine learning model.

[0182] At block 2812, the operation controller 2701 (FIG. 10 and FIG. 11) may determine whether the output of the model indicates updates to the parameters or whether the determined parameters are different than the current parameters. For example, the operation controller 2701 may compare the values of the updated parameters to the currently set parameters. If the parameters are different, then at block 2814, the operation controller 2701 may adjust the devices, equipment, or fluid within the refinery to the updated parameters.

[0183] In an embodiment, refinery operations may occur continuously or substantially continuously. As such, method 2800 may be an iterative and continuous process that occurs in real-time or near real-time. As target products change and / or other aspects of the refinery change, parameters may continue to be adjusted via method 2800.

[0184] FIG. 13 is a schematic diagram of a refinery 3500. Crude oil 3502 is initially processed by an atmospheric distillation tower 3504 into its constituent parts. Some of the lightest parts from the atmospheric distillation tower 3504 are a gas 3506 that is sent to gas processing 3508. Gas processing 3508 separates sour gas 3509 from the feeds thereto, which include gas 3506 and other gas 3507 resulting from different processes throughout the refinery. Gas processing 3508 diverts the sour gas 3509 to amine treating 3516. The remainder of the gas 3506 and other gas 3507 that is not sour gas is passed to the mercaptan treater 3510 that separates mercaptans from the fuel gas. The fuel gas may be finished as liquid petroleum gas (“LPG”) 3512. The separated butane 3514 may be kept as a finished product, sent to the gasoline blending pool, or sent to the butamer 3591 to be processed into isobutane 3593 as a finished product or further sent to an alkylation unit 3594.

[0185] Optionally, the atmospheric distillation tower 3504 may include or be connected to additional separation equipment 3505, such as a splitter, separator, debutanizer, or adeisopentanizer. A debutanizer and a deisopentanizer are distillation columns that may be used to further split lighter components, such as butane and isopentane, respectively, from the light naphtha 3536. In particular for a deisobutanizer, the cut off may be set between isobutane and butane as they have different boiling and condensation temperatures. The isobutane may be hydrotreated, passed through the mercaptan treater, or a gasoline desulfurization unit 3556 to remove sulfur and other contaminants. Once treated, the isobutane may be sent to alkylation 3594 or to the to the gasoline blending pool.

[0186] In particular for a deisopentanizer, the cutoff may be set between isopentane and n- pentane as they have different boiling and condensation temperatures. If the deisopentanizer is positioned to treat the light naphtha as part of atmospheric distillation 3504, the lighter components from the light naphtha may be hydrotreated or passed through a hydrotreater or the gasoline desulfurization unit 3556 for sulfur removal before further processing or being sent to the gasoline blending pool.

[0187] Amine treating 3516 separates the hydrogen sulfide gas 3520 from the refinery fuel 3518. The hydrogen sulfide gas 3520 and the hydrogen sulfide gas collected from processes throughout the refinery 3500 is sent to the sulfur plant 3522. The sulfur plant 3522 converts the hydrogen sulfide gas 3520 into sulfur 3530 as a finished product.

[0188] Additionally, sour water 3532 collected from processes throughout the refinery 3500 is sent to the sour water steam stripper 3528. The sour water steam stripper 3528 uses steam 3534 to remove hydrogen sulfide gas 3526 from the sour water 3532. The hydrogen sulfide gas 3526 is also sent to the sulfur plant 3522.

[0189] The atmospheric distillation tower 3504 separates light naphtha 3536 from the crude oil 3502. The light naphtha 3536 is sent to a hydrotreater 3538 that removes sulfur from the light naphtha 3536. The desulfurized light naphtha is then sent to isomerization 3540 to be processed into isomerate 3542. Isomerate 3542 are isomers of the light naphtha that have higher octane values. The isomerate 3542 may then be sent to the gasoline blending pool.

[0190] The atmospheric distillation tower 3504 separates heavy naphtha 3544 from the crude oil 3502. Optionally, the heavy naphtha 3544 may be sent to a splitter 3545. The splitter 3545 may include one or more splitter columns that separate heavier C7+ naphtha molecules from lighter components of the heavy naphtha 3544. The heavier C7+ naphtha may be sent to the hydrotreater 3554 to be desulfurized and produced as jet fuel 3558. The remaining lighter components of the heavy naphtha may be sent to a hydrotreater 3546 toremove sulfur from the heavy naphtha 3544 prior to being sent to the catalytic reformer 3548 to be converted into reformate 3550.

[0191] Alternatively, the heavy naphtha 3544 may be sent directly to the hydrotreater 3546 to remove sulfur from the heavy naphtha 3544 prior to being sent to the catalytic reformer 3548. From the hydrotreater 3546, the desulfurized heavy naphtha is sent to a catalytic reformer 3548 to be processed into reformate 3550. Reformate 3550 includes high-octane branched and cyclic hydrocarbons, such as benzene, toluene, xylene, and ethylbenzene. The reformate 3550 may then be sent to the gasoline blending pool.

[0192] Another product from the atmospheric distillation tower 3504 is jet fuel 3552. The jet fuel 3552 may include kerosene and other equivalent hydrocarbons. The jet fuel 3552 may be sent from the atmospheric distillation tower 3504 to a hydrotreater 3554 to remove contaminants from the jet fuel 3552, such as sulfur and mercaptans. Once desulfurized, the finished jet fuel 3558 may be sold or sent to the jet fuel blending pool to be blended with other products and additives and then made available for sale and distribution. The jet fuel 3552 may also be separated into kerosene (not shown).

[0193] Further, diesel 3560 is another product separated from crude oil 3502 by the atmospheric distillation tower 3504. The diesel 3560 is sent from the atmospheric distillation tower 3504 to a hydrotreater 3562 to remove sulfur. The desulfurized diesel 3564 may then be sold or sent to the diesel blending pool.

[0194] The atmospheric distillation tower 3504 also separates atmospheric gas oil 3566 and atmospheric bottoms 3568 from the crude oil 3502. The atmospheric bottoms 3568 may be sent to vacuum distillation 3570 where the atmospheric bottoms 3568 may be further separated into light vacuum gas oil 3572 (“LVGO”), medium vacuum gas oil 3599 (“MVGO), heavy vacuum gas oil 3584 (“HVGO”), and vacuum residuum 3521. Vacuum distillation 3570 may also be configured to separate the atmospheric bottoms 3568 into more or less components.

[0195] The atmospheric gas oil 3566, the LVGO 3572, the MVGO 3599, and deasphalted oil 3501 from solvent deasphalting 3598 (“SDA”) may be sent to a hydrotreater 3574 to remove sulfur, then fed into a fluid catalytic cracker 3576. The fluid catalytic cracker 3576 processes the atmospheric gas oil 3566 and the light vacuum gas oil 3572 into naphtha 3578, jet fuel 3581, diesel 3582, fuel oil 3583, and butenes and pentenes 3592.

[0196] The butenes and pentenes 3592 from the fluid catalytic cracker 3576, as well as the isobutane 3593 from the butamer 3591, may be sent to an alkylation unit 3594 where thefeedstocks are processed into alkylate 3596. The alkylate 3596 may then be sold or sent to the gasoline blending pool. Alkylate 3596 has a high octane rating and low RVP.

[0197] The naphtha 3578 may be sent to a hydrotreater 3580 to further remove sulfur from the naphtha 3578. The desulfurized naphtha may then be sold or sent to the gasoline blending pool. The jet fuel 3581 may also be passed through a hydrotreater 3585 to remove sulfur and other contaminants and then may be sent to the jet fuel blending pool. The diesel 3582 may also be passed through a hydrotreater 3587 to remove sulfur and other contaminants and then sent to the diesel blending pool, and the fuel oil 3583 may be sold or stored for later sale and distribution. Lastly, the fuel oil 3583 may also be passed through a hydrotreater 3585 to remove sulfur and other contaminants and then sent to the fuel oil blending pool. In some embodiments, the fuel oil blending pool may be used to blend formulations of low sulfur fuel oil or ultra-low sulfur fuel oil.

[0198] The MVGO 3599, HVGO 3584, and the deasphalted oil 3501 may be sent to the hydrocracker 3586 to be processed into gasoline 3588, diesel 3590, and jet fuel 3595. The hydrocracker 3586 may also produce smaller hydrocarbons that may be sent to gas processing 3508. The hydrocracker 3586 process integrates desulfurization and other contaminant removal, so further hydrotreating is not necessary for its products. The gasoline 3588 may be sold or sent to the gasoline blending pool. The diesel 3590 may be sold or sent to the diesel blending pool. The jet fuel 3595 may be sold or sent to the jet fuel blending pool.

[0199] The vacuum residuum 3521 may be sent to solvent deasphalting 3598 (“SDA”) or used directly in asphalt 3519. In some applications, the vacuum residuum 3521 may be sent to the asphalt blending pool. The SDA 3598 may be used to extract lighter components from the vacuum residuum 3521 using solvents such as propane, butane, pentane, or a combination of these hydrocarbons to extract deasphalted oil 3501 from the vacuum residuum 3521. As discussed above, the deasphalted oil 3501 may be sent to the hydrocracker 3586 or the fluid catalytic cracker 3576 for refining into hydrocarbon products including naphtha, jet fuel, diesel, and fuel oil. Once the lighter components are removed, the remaining components may be referred to as pitch 3579. The pitch 3579 may also be used in the asphalt blending pool.

[0200] The vacuum residuum 3521 and pitch 3579 may also be sent to the coker 3597 to be processed into naphtha 3503 and processed through a gasoline desulfurization unit 3556 (“GDU”). The GDU 3556 may include one or more processes useful to remove contaminants from the naphtha 3503 in addition to gasoline desulfurization, includinghydrotreating processes, catalysts, particulate catches, clay treaters, salt driers, and mercaptan treaters, separators, and steam strippers. The naphtha 3503 may then be sent to the gasoline blending pool. The coker 3597 also processes the vacuum residuum 3521 and pitch 3579 into jet fuel 3511, diesel 3513, fuel oil 3515, and petroleum coke 3517. The jet fuel 3511 may be passed through a hydrotreater 3523 and then sent to the jet fuel blending pool. The diesel 3513 may be sent to a hydrotreater 3525 to remove sulfur and then sent to the diesel blending pool. The fuel oil 3515 may also be processed through hydrotreater 3527 and then sent to the fuel oil blending pool.

[0201] The hydrotreaters 3538, 3546, 3554, 3562, 3574, 3580, 3585, 3587, 3589, 3523, 3525, and 3527 refer broadly to desulfurization and contaminant removal processes generally, including hydrotreating processes, clay treaters, salt driers, mercaptan treaters, gasoline desulfurization units, separators, steam strippers, filters, and particulate catches.

[0202] Many of the connections, feedstocks, and outputs are not shown, and those of skill in the art recognize that different configurations are possible and processes may be added or replaced by other processes known in the art. For example, atmospheric distillation tower 3504 and vacuum distillation 3570 may each represent multiple units set up in parallel or series, and may separate their feedstocks into more or fewer crude oil components. Additional equipment may be added to each process to further refine and separate the products of each process. For example, products and components of the products may be passed through isomerization, reformation, and alkylation processes not shown in FIG. 13 to process the products and components to meet regulatory requirements and customer specifications.

[0203] FIG. 14 is a schematic diagram of a machine learning model 3800 according to one embodiment. As used herein, a “machine learning model” refers to a computer algorithm or model, such as a classification model, a regression model, a language model, an object detection model, a multi-modal model, or an artificial intelligence, that can be trained and tuned based on training input to approximate unknown functions. For example, a machine learning model may refer to a neural network or other machine learning algorithm or architecture that learns and approximates complex functions and generates outputs based on a plurality of inputs provided to the machine learning model. Further, a “machine learning model” may refer to one or multiple machine learning models that cooperatively generate one or more outputs based on corresponding inputs. For example, a machine learning model may refer to any system architecture having multiple discrete machine learning components that consider different kinds of information or inputs.

[0204] The machine learning model 3800 is connected to a historical database 3802. The historical database 3802 may include multiple separate databases of information including process data, incident information and analysis, business information, demand planning, historical forecast information, historical pricing and purchasing data, and process training information. In some embodiments, the historical database 3802 may be a copy of historical databases that are maintained and updated by the machine learning model 3800. The machine learning model 3800 is also connected to a training database 3803 that includes the process data and historical information that is used for training of the machine learning model 3800.

[0205] As discussed in relation to other figures of this disclosure, the machine learning model 3800 may be used to analyze and improve specific processes, process controllers, and process interactions. The machine learning model 3800 may also be used as a refinery controller.

[0206] In this embodiment, the machine learning model 3800 may receive feedstock sensor, sample, and process data 3804 from a feedstock process controller 3814 or directly from the sensors, sampling and testing systems, and labs obtaining data from the feedstock process. A feedstock process is any process preceding the targeted process under review by the machine learning model 3800. For example, the feedstock process may include one or more of atmospheric distillation, vacuum distillation, filtering, hydrotreater, mercaptan treater, splitter, stripper, and separator.

[0207] Further, the machine learning model 3800 may receive targeted process sensor, sample, and process data 3806 from a targeted process controller 3816 or directly from the sensors, sampling and testing systems, and labs obtaining data from the targeted process. The machine learning model 3800 may also receive subsequent process sensor, sample, and process data 3808 from a targeted process controller 3816 or directly from the sensors, sampling and testing systems, and labs obtaining data from the targeted process. The feedstock sensor, sample, and process data 3804, the targeted process sensor, sample, and process data 3806, and the subsequent process sensor, sample, and process data 3808 may include operating pressure and temperature data for the components of each process and sub-process. The feedstock sensor, sample, and process data 3804, the targeted process sensor, sample, and process data 3806, and the subsequent process sensor, sample, and process data 3808 may include material composition data describing the material moving through the process, such as feed rates, and quantities and percentages of contaminants, reactants, and the hydrocarbons moving through the process.

[0208] The machine learning model 3800 may also access business data 3810. Business data 3810 may include demand planning information, forecasting information for various products, current pricing information, product distribution information, and information regarding the current and anticipated regulatory landscape. Business data 3810 may also include information regarding the costs, availability, and location of storage, disposal, and in some cases carbon capture options of waste products from each process. Business data 3810 may also include historical and current costs of each process and the refinery products, as well as current and historical market pricing for each product. Business data 3810 may also include the inventory levels of various products that may be stored in the storage tanks 3575 of the various blending pools.

[0209] The machine learning model 3800 may also access lab data 3811 that is obtained from laboratory analysis of samples collected from processes within a refinery and products produced by those processes. Lab data 3811 may include chemical composition and component distribution data about each sample, as well as information about when and where the sample was taken, tests performed on the sample, and sensor data from the process at the time the sample was taken.

[0210] The machine learning model 3800 may also access weather data 3813 as the ambient temperature, pressure, and solar radiation may also affect the temperature and pressure of feedstocks, products, processes, and storage tanks within a refinery. The machine learning model may build algorithms to predict the effect of weather on process operating parameters, feedstocks, the resulting products from those processes, and the effect of weather on the storage of products within the blending pools.

[0211] The machine learning model 3800 may publish to and receive instructions from an engineering gateway 3812. The engineering gateway 3812 may act as a user interface for engineers to review process and machine learning model 3800 data, recommendations, warnings, and requests. A user may use the engineering gateway 3812 to assist the machine learning model 3800 in refining and tuning its algorithms to better predict and adjust each process in response to changes in ambient weather, demand seasonality, and composition and quality of feedstocks including the crude oil received for processing into refined hydrocarbons. The engineering gateway may also be used to facilitate active learning by the machine learning model 3800.

[0212] The machine learning model 3800 may access, receive data from, and provide instructions to a feedstock process controller 3814, a targeted process controller 3816, and a subsequent process controller 3818. In some embodiments, the machine learning model3800 may adjust an algorithm used by the targeted process controller 3816 based on changes identified from the adjusted algorithm. Once an adjusted algorithm has been selected by the machine learning model 3800, the machine learning model 3800 requests approval to implement the adjusted algorithm through the engineering gateway 3812. Upon receipt of approval, the machine learning model 3800 saves a copy of the approval information in the historical data 3802 and sends the approved adjusted algorithm to the targeted process controller 3816.

[0213] The machine learning model 3800 includes a data analysis module 3822. The data analysis module 3822 may be used by the machine learning model 3800 to review data and identify data that may be considered outliers and disregarded. Once the data set has been reviewed and amended to remove outlier data, the machine learning model 3800 may save the revised data set in either the training data 3803 or the historical data 3802 for retraining of the machine learning model 3800, for analysis and adjustment of a targeted process, or for later use.

[0214] The machine learning model 3800 includes an interpolation module 3824. The interpolation module 3824 may be used to analyze a data set and interpolate missing data from the data set. The interpolation module 3824 may compare the historical data 3802 with feedstock sensor, sample, and process data 3804, the targeted process sensor, sample, and process data 3806, and the subsequent process sensor, sample, and process data 3808 to identify process changes. The interpolation module 3824 may be able to identify and flag small deviations for further investigation. The interpolation module 3824 works with the communication module 3828 to publish the data regarding the flagged deviation for user review and guidance. For example, the interpolation module 3824 may be used to identify miscalibrated sensors or test data that may be inaccurate. The machine learning model 3800 may accomplish this by identifying and labeling data as outliers. Labeled data may be associated with a sensor or data and then reported to a user through the engineering gateway 3812. The machine learning model 3800 may use the interpolation module 3824 to identify components that may be wearing out and catalysts that may be deactivated or poisoned. The interpolation module 3824 may also communicate through the communication module 3828 with the engineering gateway 3812 to identify these potential process concerns to a user for further investigation.

[0215] The machine learning model 3800 includes a prediction module 3826. The prediction module 3826 may analyze the feedstock sensor, sample, and process data 3804, the targeted process sensor, sample, and process data 3806, and the subsequent processsensor, sample, and process data 3808 to predict changes in a process. For example, temperature excursions may be predicted in a process and the prediction module 3826 may communicate through the communication module 3828 with the engineering gateway 3812 to notify a user of a predicted temperature excursion and recommending process changes to avoid the temperature excursion. A user may accept the recommended process changes through the engineering gateway 3812 or provide different instructions for the machine learning model 3800 to implement. In predicting changes in a process, prediction module 3826 may recommend regeneration or replacement of a catalyst or repair or replacement of a component of a process as part of a maintenance procedure. The prediction module 3826 may provide a recommended maintenance window for the maintenance procedure to occur.

[0216] The machine learning model 3800 includes a communication module 3828. The communication module 3828 may translate recommendations, instructions, and data from other modules from machine code into a human language or convert data into graphs and other visual communication elements. The communication module 3828 may send or publish communications to the engineering gateway 3812. The communication module 3828 may also communicate with the various process controllers and other models used by the refinery.

[0217] In some embodiments, the machine learning model 3800 includes a confidence module 3830. The confidence module 3830 may review the analysis and recommendations of the different modules of the machine learning model 3800 and assign a confidence level to the analysis and recommendations. For example, the confidence module 3830 may produce SHAP values, or use LIME or anchors to analyze each algorithm that is adjusted or created by the machine learning model 3800. The results may be published by the communication module 3828 to the engineering gateway 3812 to assist a user in reviewing each algorithm and the changes recommended by the machine learning model 3800.

[0218] In some embodiments, the machine learning model 3800 includes a regulatory module 3832. The regulatory module 3832 may access the business data 3810 to assist in regulatory compliance. For example, the regulatory module 3832 may flag a process that may be predicted by the prediction module 3826 to move out of compliance. The regulatory module 3832 may issue a warning through the communication module 3828 to the engineering gateway 3812. The regulatory module 3832 may also identify windows of time when regulations are lessened and recommend changes to process parameters to reduce process costs. The regulatory module 3832 may make recommendations to gasoline blending pool controllers, diesel blending pool controllers, and jet fuel blending poolcontrollers to adjust blending parameters in line with upcoming regulatory changes. Further, the regulatory module 3832 may make recommendations for various process controllers to adjust their parameters to promote the production of blend components inline with demand that may accompany regulatory changes.

[0219] In some embodiments, the machine learning model 3800 includes a forecasting module 3834. The forecasting module 3834 may analyze business data 3810 and recommend that various process controllers adjust their parameters to promote the production of components that may be in greater demand. The forecasting module 3834 may also recommend maintenance windows for equipment producing products that may be in low demand during specific time frames. The forecasting module 3834 may forecast pricing for various components based on the historical data 3802 and the business data 3810 and publish the forecasted pricing via the communication module 3828 to the engineering gateway 3812.

[0220] In some embodiments, the machine learning model 3800 may include an authority module 3836 and additional modules 3838, such as a display module for converting data into graphical representations of a data set or a translation module for converting data between different languages, formats, or units. The authority module 3836 may track user instructions and approvals for various instructions and changes to be made by the machine learning model 3800 to the various controllers throughout the refinery. In some applications, the authority module 3836 may consider a recommendation from a module of the machine learning model 3800 and automatically authorize the machine learning model 3800 to issue an instruction to the targeted process controller 3816 to make a change to an associated process. In other applications, the authority module 3836 may direct the communication module 3828 to request approval through the engineering gateway 3812 before a recommendation may be implemented and instructions sent to the targeted process controller 3816.

[0221] The machine learning model 3800 provides better process control and faster changes to processes within a refinery. The machine learning model 3800 also provides better data and clarity to changes and predicted changes within the nonlinear processes found within a refinery.

[0222] FIG. 15 is a flow chart of a method, according to at least one embodiment of the present disclosure. The method includes analyzing a naphtha sample to provide naphtha sample properties, wherein the naphtha sample is taken from naphtha being supplied to a reformer at 4010 and predicting one or more naphtha properties associated with the naphthasample based on the naphtha sample properties and a first output from application of the naphtha sample properties to a machine learning model trained on historical data including operational parameters of the reformer, naphtha sample properties, and unit materials sample properties at 4020. The method further includes operating the reformer to produce unit materials, the unit materials having unit materials properties, and the unit materials including reformate and hydrogen at 4030 and providing operating parameters of the operating reformer to the machine learning model at 4040. The method then analyzes a unit material sample to provide unit material sample properties at 4050 and predicts one or more unit material properties associated with the unit material sample based on the unit material sample properties and a second output from application of the unit material sample properties to the machine learning model at 4060. The method then includes generating, by the machine learning model, an adjustment to a parameter of the reforming operation based on the naphtha sample properties and the unit material sample properties to optimize reformer production of unit materials having a property at or exceeding a target property for the unit materials at 4070. The machine learning model may perform all of the steps of the method of Fig. 15.

[0223] FIG. 16 is a flow chart of another method according to one embodiment, according to at least one embodiment of the present disclosure. The method includes receiving product data indicative of a benzene component concentration in a product of a catalytic reformer at 4110 and receiving feedstock data indicative of a feedstock for processing by the catalytic reformer at 4120. The method further includes generating, by a machine learning model, an adjustment to a parameter of an upstream distillation operation supplying the feedstock to the catalytic reformer based on the feedstock data and the product data to reduce the benzene component concentration in the product at 4130. The machine learning model is trained with distillation operation historical data indicative of operating parameters, operational constraints, the feedstock, and products of the distillation operation. The machine learning model is also trained with catalytic reformer historical data indicative of operating parameters, operational constraints, the feedstock, and products of the catalytic reformer. The machine learning model may perform all of the steps of the method of Fig. 16.

[0224] FIG. 17 is flow chart of yet another method, according to at least one embodiment of the present disclosure. The method includes receiving product data indicative of a property of a product of a catalytic reformer at 4210 and receiving feedstock data indicative of a feedstock for processing by the catalytic reformer at 4220. A machine learning modelthe generates an adjustment to a parameter of the catalytic reformer based on the feedstock data and the product data to improve the property of the product of the catalytic reformer at 4230. The machine learning model is trained with catalytic reformer historical data indicative of operating parameters, operational constraints, the feedstock, and products of the catalytic reformer. The machine learning model may perform all of the steps of the method of Fig. 17.

[0225] FIG. 18 is a flow chart of a method, according to at least one embodiment of the present disclosure. The method includes receiving product data indicative of a property of a product of a catalytic reformer at 4310 and receiving feedstock data indicative of a feedstock for processing by the catalytic reformer at 4320. The catalytic reformer includes a first reactor in series with a second reactor. The method further includes a machine learning model generating an adjustment to a parameter of the first reactor based on the feedstock data and the product data to improve the property of the product of the catalytic reformer while reducing yield loss of the product at 4330. The machine learning model is trained with catalytic reformer historical data indicative of operating parameters, operational constraints, the feedstock, and products of the catalytic reformer. The machine learning model generates an adjustment to a parameter of the second reactor based on the adjustment to the parameter of the first reactor, feedstock data, and the product data at 4340. The machine learning model may perform all of the steps of the method of Fig. 18.

[0226] FIG. 19 is a flow chart of another method according to one embodiment, according to at least one embodiment of the present disclosure. The method includes receiving product data indicative of a product component of a product of a catalytic reformer at 4410 and receiving feedstock data indicative of a feedstock for processing by the catalytic reformer at 4420. The method further includes receiving a target octane number for the reformate produced by the catalytic reformer at 4430 and generating, by a machine learning model, an adjustment to a parameter of the catalytic reformer based on the feedstock data and the product data to achieve the target octane number of the reformate while maximizing yield of the reformate at 4440. The machine learning model is trained with catalytic reformer historical data indicative of operating parameters, operational constraints, the feedstock, and products of the catalytic reformer. The machine learning model may perform all of the steps of the method of Fig. 19.

[0227] FIG. 20 is flow chart of yet another method, according to at least one embodiment of the present disclosure. The method includes receiving product data indicative of a property of a product of a catalytic reformer at 4510 and receiving feedstock data indicativeof a feedstock for processing by the catalytic reformer at 4520. The method further includes receiving a target octane number for the reformate produced by the catalytic reformer at 4530 and generating, by a machine learning model, an adjustment to a parameter of the catalytic reformer based on the feedstock data and the product data to achieve an enhanced yield of hydrogen gas while achieving the target octane number of the reformate at 4540. The machine learning model is trained with catalytic reformer historical data indicative of operating parameters, operational constraints, the feedstock, and products of the catalytic reformer. The machine learning model may perform all of the steps of the method of Fig. 20.

[0228] FIG. 21 is flow chart of a method, according to at least one embodiment of the present disclosure. The method includes receiving one or more of (a) operational data of an aromatics recovery unit indicative of operational parameters of the aromatic recovery unit and including sensor data from one or more sensors disposed to measure a parameter of the aromatics recovery unit, (b) product data indicative of a property or component of a product from a process of the aromatics recovery unit or (c) feedstock data of a reformate being fed into the aromatics recovery unit at 4610 and determining a change in the one or more of the operational data, the product data, or the feedstock data at 4620. The method further includes generating, by a machine learning model, a prediction of the effect of the change on the operation of the aromatic recovery unit at 4630 and generating, by a machine learning model, an adjustment to an operating parameter of the aromatics recovery unit based on the predicted effect of the change. The machine learning model is trained on historical data including operational data and product data of the aromatic recovery unit at 4640. Optionally, the method includes implementing the adjustment to the operating parameter of the aromatics recovery unit at 4650.

[0229] FIG. 22 is flow chart of another method, according to at least one embodiment of the present disclosure. The method includes receiving one or more of (a) operational data of an aromatics recovery unit indicative of operational parameters of the aromatic recovery unit and including sensor data from one or more sensors disposed to measure a parameter of the aromatics recovery unit, (b) product data indicative of a property or component of a product from a process of the aromatics recovery unit or (c) feedstock data of a reformate being fed into the aromatics recovery unit at 4710 and generating, by the machine learning model, a prediction of a process constraint based on the operating data at 4720. The method further includes generating, by a machine learning model, an adjustment to an operating parameter of the aromatics recovery unit based on the predicted process constraint at 4730.The machine learning model is trained on historical data including operational data and product data of the aromatic recovery unit. The method may optionally include publishing the predicted effect of the change to a user at 4740. The method may optionally include receiving an instruction to implement the adjustment to the operating parameter of the aromatics recovery unit at 4750. The method may optionally include implementing the adjustment to the operating parameter of the aromatics recovery unit at 4760.

[0230] FIG. 23 is flow chart of a method, according to at least one embodiment of the present disclosure. The method includes receiving one or more of (a) operational data of an aromatics recovery unit indicative of operational parameters of the aromatics recovery unit and including sensor data from one or more sensors disposed to measure a parameter of the aromatics recovery unit, (b) product data indicative of a property or component of a product from a process of the aromatics recovery unit or (c) feedstock data of a reformate being fed into the aromatics recovery unit at 4810. A machine learning model generates a prediction of solvent quality based on the operating data at 4820 and an adjustment to an operating parameter of the aromatics recovery unit based on the predicted solvent quality at 4830. In this embodiment, the machine learning model is trained on historical data including operational data and product data of the aromatics recovery unit. The method may optionally include implementing the adjustment to the operating parameter of the aromatics recovery unit at 4840.

[0231] FIG. 24 is flow chart of another method, according to at least one embodiment of the present disclosure. The method includes receiving one or more of (a) operational data of an aromatics recovery unit indicative of operational parameters of the aromatics recovery unit and including sensor data from one or more sensors disposed to measure a parameter of the aromatics recovery unit, (b) product data indicative of a property or component of a product from a process of the aromatics recovery unit or (c) feedstock data of a reformate being fed into the aromatics recovery unit at 4910. A machine learning model generates a prediction of solvent loading based on the operating data at 4920 and generates an adjustment to an operating parameter of the aromatics recovery unit to improve solvent loading based on the predicted solvent loading at 4930. In this embodiment, the machine learning model is trained on historical data including operational data and product data of the aromatics recovery unit. The method may optionally include implementing the adjustment to the operating parameter of the aromatics recovery unit at 4940.

[0232] In the drawings and specification, several embodiments of systems and methods to provide in-line mixing of hydrocarbon liquids have been disclosed, and although specificterms are employed, the terms are used in a descriptive sense only and not for purposes of limitation. Embodiments of systems and methods have been described in considerable detail with specific reference to the illustrated embodiments. However, it will be apparent that various modifications and changes may be made within the spirit and scope of the embodiments of systems and methods as described in the foregoing specification, and such modifications and changes are to be considered equivalents and part of this disclosure.

Claims

CLAIMSWHAT IS CLAIMED IS:

1. A method comprising: receiving product data indicative of a property of a product of a catalytic reformer; receiving feedstock data indicative of a feedstock for processing by the catalytic reformer; and generating, by a machine learning model, an adjustment to a parameter of the catalytic reformer based on the feedstock data and the product data to achieve a target property of the product of the catalytic reformer, wherein the machine learning model is trained with catalytic reformer historical data indicative of operating parameters, operational constraints, the feedstock, and products of the catalytic reformer.

2. The method of claim 1, further comprising generating, by the machine learning model, a prediction of coke formation within a reactor of the catalytic reformer, wherein generating, by the machine learning model, an adjustment to a parameter of the catalytic reformer is further based on the prediction of coke formation within a reactor of the catalytic reformer.

3. The method of any of claims 1 to 2, wherein the parameter of the catalytic reformer is one or more of a temperature within a reactor, a pressure within the reactor, or a feed rate of the feedstock into the reactor.

4. The method of claim 3, wherein the adjustment to the parameter of the catalytic reformer by the machine learning model is to reduce a pressure within the catalytic reformer to within five percent of an operational constraint in order to maximize reformate yield.

5. The method of claim 4, wherein the adjustment to the parameter of the catalytic reformer by the machine learning model is to reduce pressure within the catalytic reformer to within one percent of an operational constraint in order to maximize reformate yield.

6. The method of any of claims 1 to 5, wherein the machine learning model is programmed to maximize octane-barrels through control of reactor severity.

7. The method of claim 6, further comprising generating, by the machine learning model, a predictive concave function of octane-barrels versus reactor severity, wherein the property of the product of the catalytic reformer is octane-barrels, wherein generating, by the machine learning model, an adjustment to a parameter of the catalytic reformer to move the octane-barrels of the product of the catalytic reformer closer to a predicted maximum of the predictive concave function.

8. The method of any of claims 1 to 5, further comprising: generating, by the machine learning model, a value for the products of the catalytic reformer, a value for costs of operating the catalytic reformer, and a value for the feedstock fed to the catalytic reformer; and generating a recommendation, by the machine learning model, to improve the value of the catalytic reformer operation based on the value for the products of the catalytic reformer, a value for costs of operating the catalytic reformer, and a value for the feedstock fed to the catalytic reformer.

9. The method of claim 8, further comprising publishing the recommendation.

10. The method of claims 9, further comprising implementing the recommendation.

11. The method of claims 10, wherein implementing the recommendation includes adjusting an operational parameter of the catalytic reformer to improve the value of the catalytic reformer.

12. The method of any of claims 8 to 11, further comprising accessing business data, wherein the values are based on current business data.

13. The method of claim 12, wherein the business data includes a current market price for each of the products of the catalytic reformer.

14. The method of any of claims 8 to 13, wherein the recommendation is constrained by a limitation of the catalytic reformer.

15. The method of claim 14, wherein the limitation is a deactivation state of catalysts within the catalytic reformer.

16. The method of claim 14, wherein the limitation is a heat generation rate of the catalytic reformer.

17. The method of claim 14, wherein the limitation is a lower operational temperature limit of the catalytic reformer.

18. The method of any of claims 1 to 17, wherein the property of the product of the catalytic reformer is octane-barrels of reformate.

19. The method of any of claims 1 to 17, wherein the property of the product of the catalytic reformer is a target octane number of the reformate.

20. The method of any of claims 8 to 17, wherein generating, by the machine learning model, an adjustment to a parameter of the catalytic reformer is based on an objective function, the objective function operating to improve the value of products of the catalytic reformer.

21. The method of claim 20, wherein the objective function includes the value of the products of the catalytic reformer, the value of the operation of the catalytic reformer, and the value of the feedstock.

22. The method of claims 20 or 21, further comprising installing, by the machine learning model, the objective function on a sub-controller of the catalytic reformer.

23. The method of any of claims 1 to 22, wherein the catalytic reformer includes a continuous catalyst regeneration reactor and a catalyst regenerator, the method further comprising: predicting, by the machine learning model, an effect on catalysts of the continuous catalyst regeneration reactor based on the feedstock data; andgenerating, by the machine learning model, an adjustment to a parameter of the continuous catalyst regeneration reactor or the catalyst regenerator based on the predicted effect on catalysts of the continuous catalyst regeneration reactor and feedstock data.

24. The method of any of claims 1 to 22, further comprising: predicting, by the machine learning model, a deactivation state of catalysts of a reactor of the catalytic reformer; and generating, by the machine learning model, an adjustment to a parameter of the reactor based on the feedstock data and the predicted deactivation state of the catalysts of the reactor.

25. The method of any of claims 1 to 24, further comprising: receiving gasoline blending pool data at the machine learning model; and generating the target property of the product of the catalytic reformer based on the gasoline blending pool data.

26. The method of claim 25, wherein the gasoline blending pool data includes a customer specification for a blended formulation.

27. The method of any of claims 25 to 26, wherein the target property is a target octane- barrels of reformate.

28. The method of any of claims 25 to 27, wherein the target property is a target octane number of the reformate.

29. The method of any of claims 25 to 28, wherein the target property is measured or sampled proximate an exit of a reactor of the catalytic reformer.

30. The method of any of claims 25 to 28, wherein the target property is measured or sampled proximate an exit of a separator, a distillation unit, a stripper, or a stabilizer of the catalytic reformer.

31. The method of any of claims 25 to 30, wherein generating, by the machine learning model, the adjustment to a parameter of the catalytic reformer includes generating an adjustment to an operating temperature of a reactor of the catalytic reformer.

32. The method of claim 31, further comprising predicting, by the machine learning model, a deactivation state of catalysts of the reactor of the catalytic reformer, wherein generating an adjustment to an operating temperature of a reactor of the catalytic reformer is further based on the predicted deactivation state of catalysts of the reactor of the catalytic reformer.

33. The method of any of claims 25 to 32, further comprising implementing the generated adjustment to adjust the parameter of the catalytic reformer.

34. The method of any of claims 31 to 32, further comprising publishing a notification by the machine learning model that the target property of the product is not achievable based on constraints of the reactor of the catalytic reformer.

35. The method of any of claims 31 to 34, further comprising; generating, by the machine learning model, a possible property of the product closest to the target property of the product; and publishing a notification of a possible property of the product closest to the target property of the product.

36. The method of any of claims 31 to 35, further comprising generating, by the machine learning model, a determination to not implement the adjustment to a parameter of the catalytic reformer.

37. The method of any of claims 1 to 33, further comprising: receiving operational data of a downstream aromatics recovery unit; supplying the product of the catalytic reformer to the downstream aromatics recovery unit; wherein the adjustment to the parameter of the catalytic reformer is further based on the operational data of the downstream aromatics recovery unit.

38. The method of claim 37, further comprising predicting an effect of the adjustment to the parameter of the catalytic reformer on the downstream aromatics recovery unit.

39. The method of claim 38, further comprising generating an adjustment to an operational parameter of the downstream aromatics recovery unit based on the predicted effect of the adjustment to the parameter of the catalytic reformer on the downstream aromatics recovery unit.

40. The method of any of claims 37 to 39, further comprising implementing the adjustment to an operational parameter of the downstream aromatics recovery unit.

41. The method of any of claims 39 to 40, wherein the operational parameter of the downstream aromatics recovery unit is a feed rate of the product of the catalytic reformer into the downstream aromatics recovery unit.

42. The method of any of claims 1 to 41, wherein the machine learning model is programmed to drive the catalytic reformer to improve the property of the product of the catalytic reformer while seeking to improve yield of the reformate.

43. The method of claim 42, further comprising: generating a plurality of simulations, by the machine learning model, of the operation of the catalytic reformer based on the feedstock data, wherein each simulation has a different adjustment to an operational parameter; selecting, by the machine learning model, a simulation of the plurality of simulations, wherein the selected simulation has a highest simulated yield of reformate while improve the property of the product of the catalytic reformer.

44. The method of claim 43, wherein generating, by the machine learning model, the adjustment to the parameter of the catalytic reformer is based on the selected simulation.

45. A memory including instructions that causes a processor to perform the instructions including the method of any of claims 1 to 44.

46. A computing system including a processor and a memory including instructions that cause the processor to perform the instructions, wherein the instructions include the method of any of claims 1 to 41.

47. A system for enhancing reformate production for a dual reformer operation, the system comprising: a first reformer to receive a first feedstock and reform the first feedstock into first reformate, first hydrogen, and other first gasses; a second reformer to receive a second feedstock and reform the second feedstock into second reformate, second hydrogen, and other second gasses; a plurality of sensors to measure a parameter associated with one or more of the first reformer or the second reformer, the plurality of sensors positioned at one of (a) proximate the first reformer and the second reformer, (b) proximate the first reformer and within the second reformer, (c) proximate the second reformer and within the first reformer, or (d) within the first reformer and the second reformer; a plurality of refinery operation control devices each positioned (a) proximate and downstream or upstream of the first reformer and (b) proximate and downstream or upstream of the second reformer to control aspects of fluid flowing to or from the first reformer or the second reformer; a plurality of sample collection assemblies to collect samples of (a) a first fluid associated with the reformer and (b) a second fluid associated with the second reformer; a plurality of sample analysis assemblies to analyze collected samples to provide properties of the collected samples; and a reformer controller in signal communication with one or more of the first reformer or the second reformer, the plurality of sensors, the plurality of refinery operation control devices, and the plurality of one or more sample analysis assemblies, and storing a trained machine learning model, the reformer controller configured to: determine an output including predicted properties of the first feedstock, the second feedstock, and parameter settings of the plurality of refinery operation control devices, the first reformer, and the second reformer based on application of one or more of (i) data measured by the plurality of sensors, (ii) data corresponding to analysis from the plurality of one or more sample analysis assemblies, or (iii) a target product and corresponding target properties to the trained machine learning model, andadjust one or more of (a) a first amount of (a) the first feedstock or type of first feedstock and first parameters associated with the refinery operation control device and the first reformer and (b) a second amount of the second feedstock or type of second feedstock and second parameters associated with the refinery operation control device and the second reformer based on the output to enhance production of the first reformate, the second reformate, the first other gasses, and the second other gasses.

48. The system of claim 47, wherein the reformer controller is configured to: aggregate and analyze measured parameters from the plurality of sensors and the one or more collected properties of the one or more collected samples from the one or more sample analysis assemblies; compute optimized targets for the one or more products from the reforming operation using predictive modeling and historical data trends; dynamically adjust refinery operation control devices based on deviations from predefined quality specifications; provide real-time or near real-time feedback to operators for manual intervention when necessary; and ensure compliance with regulatory and performance constraints through automated enforcement mechanisms.

49. The system of claim 48, wherein the reformer controller is configured to dynamically update optimized targets based on real-time or near real-time data.

50. The system of claim 49, wherein the reformer controller is configured to determine the optimized targets using at least one of a linear programming model, a non-linear programming model, or a machine learning model, each tailored to dynamic refinery conditions.

51. The system of any of claims 48 to 49, wherein the reformer controller is further configured to adjust the one or more target properties for one refinery unit based on rundown results from the reforming operation to achieve an average octane-barrels of the products of the reforming operation.

52. The system of claim 51, wherein the system is configured to enhance the reforming operation in real-time or near real-time, with a latency of less than 5 seconds, based on the one or more measured parameters, the one or more collected properties of the one or more collected samples, and one or more achieved properties of the one or more products of the reforming operation.

53. The system of any of claims 48 to 52, wherein the reformer controller recommends pathways for achieving reforming operation product specifications, including adjusting one or more of fractionator cut points.

54. The system of any of claims 48 to 53, wherein the reformer controller monitors external factors impacting reforming operation performance.

55. The system of claim 54, further comprising a feedback mechanism that is configured to adjust one or more of refinery operation control devices, upon identifying external factors impacting reforming operation performance.

56. The system of any of claims 48 to 55, wherein the reformer controller provides predictive control insights identifying external changes impacting the reforming operation and proactively adjusting control settings.

57. The system of any of claims 47 to 56, wherein each of the plurality of sensors is positioned at one of (a) proximate to one of a plurality of refinery equipment or (b) within one of the plurality of refinery equipment.

58. A method for enhancing control of a reforming operation associated with a petroleum refining operation, the method comprising: supplying naphtha to a reformer associated with the petroleum refining operation, the naphtha having one or more naphtha properties; analyzing a naphtha sample via a first analyzer to provide naphtha sample properties; predicting one or more naphtha sample properties associated with the naphtha sample based on (A) the naphtha sample properties and (B) a first output from application of the naphtha sample properties to a first trained machine learning model;operating the reformer to produce one or more unit materials, the one or more unit materials having one or more unit materials properties, and the one or more unit materials comprising one or more of reformate, hydrogen, or reformer gas; analyzing the unit material sample via a second analyzer to provide unit material sample properties; predicting one or more unit material sample properties associated with the unit material sample based on (C) the unit material sample properties and (D) a second output from application of the unit material sample properties to a second trained machine learning model; controlling, during the reforming operation, based on the naphtha sample properties and the one or more unit material sample properties, one or more of:(a) one or more naphtha properties associated with the naphtha supplied to the reformer;(b) one or more unit product materials properties associated with the unit product materials;(c) operation of the reformer; or(d) operation of one or more upstream equipment or downstream equipment, so that the controlling, during the reforming operation, causes the reforming operation to produce one or more of:(i) one or more intermediate materials each having one or more properties within a range of one or more target properties of the one or more intermediate materials,(ii) one or more unit product materials each having one or more properties within a range of one or more target properties of the one or more unit product materials, or(iii) one or more downstream materials each having one or more properties within a range of one or more target properties of the one or more downstream materials.

59. A method compri sing : analyzing a naphtha sample to provide naphtha sample properties, wherein the naphtha sample is taken from naphtha being supplied to a reformer; predicting one or more naphtha properties associated with the naphtha sample based on the naphtha sample properties and a first output from application of the naphtha sampleproperties to a machine learning model trained on historical data including operational parameters of the reformer, naphtha sample properties, and unit materials sample properties; operating the reformer to produce unit materials, the unit materials having unit materials properties, and the unit materials including reformate and hydrogen; providing operating parameters of the operating reformer to the machine learning model; analyzing a unit material sample to provide unit material sample properties; predicting one or more unit material properties associated with the unit material sample based on the unit material sample properties and a second output from application of the unit material sample properties to the machine learning model; and generating, by the machine learning model, an adjustment to a parameter of the reforming operation based on the naphtha sample properties and the unit material sample properties to optimize reformer production of unit materials having a property at or exceeding a target property for the unit materials.

60. The method of claim 59, wherein the naphtha sample is analyzed with spectrometer to provide the naphtha sample properties.

61. The method of claims 59 or 60, wherein the machine learning model optimizes reformer production of unit materials by seeking to maximize octane-barrels, wherein the parameter is a reactor temperature.

62. The method of any of claims 59 to 61, wherein the parameter is reactor severity.

63. The method of any of claims 59 to 62, further comprising: generating, by the machine learning model, a value of naphtha sent to the reformer and a value of naphtha sent to other refinery processes; receiving, by the machine learning model, constraints of the other refinery processes for processing the naphtha and demand by the other refinery processes for the naphtha; and determining, by the machine learning model, a feed rate of naphtha to send to the reformer and the other refinery processes based on the value of naphtha sent to the reformer, the value of naphtha sent to other refinery processes, the constraints of the other refineryprocesses for processing the naphtha, and the demand by the other refinery processes for the naphtha.

64. The method of any of claims 59 to 62, further comprising determining a feed rate of naphtha from a naphtha source to the reformer based on a total amount of naphtha produced by the naphtha source less the amount of naphtha to be sent to a blending process.

65. The method of claims 63 or 64, further comprising adjusting a feed rate of naphtha to the reformer based on the determined feed rate of naphtha.

66. The method of any of claims 59 to 65, further comprising generating, by the machine learning model, an adjustment to a parameter of an upstream distillation operation based on the naphtha sample properties to optimize reformer production of unit materials.

67. The method of claim 66, further comprising adjusting the parameter of an upstream distillation operation based on the generated adjustment.

68. The method of any of claims 59 to 67, further comprising: receiving gasoline blending pool data at the machine learning model; and generating the target property of the product of the reformer based on the gasoline blending pool data.

69. The method of claim 68, wherein the gasoline blending pool data includes a customer specification for a blended formulation.

70. The method of any of claims 68 to 69, wherein the target property is a target octane- barrels of reformate.

71. The method of any of claims 68 to 70, wherein the target property is a target octane number of the reformate.

72. The method of any of claims 68 to 71, wherein the target property is measured or sampled proximate an exit of a reactor of the reformer.

73. The method of any of claims 68 to 71, wherein the target property is measured or sampled proximate an exit of a separator, a distillation unit, a stripper, or a stabilizer of the reformer.

74. The method of any of claims 68 to 73, wherein generating, by the machine learning model, the adjustment to a parameter of the reformer includes generating an adjustment to an operating temperature of a reactor of the reformer.

75. The method of claim 74, further comprising predicting, by the machine learning model, a deactivation state of catalysts of the reactor of the reformer, wherein generating an adjustment to an operating temperature of a reactor of the reformer is further based on the predicted deactivation state of catalysts of the reactor of the reformer.

76. The method of any of claims 68 to 75, further comprising implementing the generated adjustment to adjust the parameter of the reformer.

77. The method of any of claims 74 to 75, further comprising publishing a notification by the machine learning model that the target property of the product is not achievable based on constraints of the reactor of the reformer.

78. The method of any of claims 74 to 77, further comprising; generating, by the machine learning model, a possible property of the product closest to the target property of the product; and publishing a notification of a possible property of the product closest to the target property of the product.

79. The method of any of claims 74 to 78, further comprising generating, by the machine learning model, a determination to not implement the adjustment to a parameter of the reformer.

80. A memory including instructions that causes a processor to perform the instructions including the method of any of claims 59 to 79.

81. A computing system including a processor and a memory including instructions that cause the processor to perform the instructions, wherein the instructions include the method of any of claims 59 to 79.

82. A method comprising: receiving product data indicative of a benzene component concentration in a product of a catalytic reformer; receiving feedstock data indicative of a feedstock for processing by the catalytic reformer; and generating, by a machine learning model, an adjustment to a parameter of an upstream distillation operation supplying the feedstock to the catalytic reformer based on the feedstock data and the product data to adjust the benzene component concentration in the product, wherein the machine learning model is trained with distillation operation historical data indicative of operating parameters, operational constraints, the feedstock, and products of the distillation operation, wherein the machine learning model is trained with catalytic reformer historical data indicative of operating parameters, operational constraints, the feedstock, and products of the catalytic reformer.

83. The method of claim 82, further comprising generating, by the machine learning model, a predicted benzene component concentration in the product based on the adjustment to a parameter of an upstream distillation operation.

84. The method of any of claims 82 to 83, wherein the adjustment to a parameter of an upstream distillation operation supplying the feedstock to the catalytic reformer results in a reduction of C6 and smaller hydrocarbons in the feedstock.

85. The method of any of claims 82 to 84, further comprising generating, by the machine learning model, an adjustment to a parameter of the catalytic reformer based on the feedstock data and the product data to achieve a target benzene component concentration in the product.

86. The method of claim 85, further comprising generating, by the machine learning model, a predicted benzene component concentration in the product based on theadjustment to a parameter of an upstream distillation operation and the adjustment to a parameter of the catalytic reformer.

87. The method of any of claims 82 to 86, further comprising: generating, by the machine learning model, a value for each of the products of the catalytic reformer, a value for costs of operating the catalytic reformer, and a value for the feedstock fed to the catalytic reformer; and generating a recommendation, by the machine learning model, to improve the value of the catalytic reformer operation based on the value for each of the products of the catalytic reformer, a value for costs of operating the catalytic reformer, and a value for the feedstock fed to the catalytic reformer.

88. The method of claim 87, further comprising publishing the recommendation.

89. The method of claims 88, further comprising implementing the recommendation.

90. The method of claims 89, wherein implementing the recommendation includes adjusting an operational parameter of the catalytic reformer to improve the value of the catalytic reformer.

91. The method of claims 90, wherein adjusting the operational parameter of the catalytic reformer to improve the value of the catalytic reformer includes reducing operational pressure within the catalytic reformer to an operational limitation.

92. The method of any of claims 87 to 90, further comprising accessing business data, wherein the values are based on current business data.

93. The method of claim 92, wherein the business data includes a current market price for each of the products of the catalytic reformer.

94. The method of any of claims 87 to 93, wherein the recommendation is constrained by a limitation of the catalytic reformer.

95. The method of claim 94, wherein the limitation is a deactivation state of catalysts within the catalytic reformer.

96. The method of claim 94, wherein the limitation is a heat generation rate of the catalytic reformer.

97. The method of any of claims 87 to 96, wherein generating, by the machine learning model, an adjustment to a parameter of an upstream distillation operation is based on an objective function, the objective function operating to improve the value of products of the catalytic reformer.

98. The method of claim 97, wherein the objective function includes the value of the products of the catalytic reformer, the value of the operation of the catalytic reformer, and the value of the feedstock.

99. The method of claims 97 or 98, further comprising installing, by the machine learning model, the objective function on a sub-controller of the upstream distillation operation or the catalytic reformer.

100. The method of any of claims 82 to 99, wherein the feedstock data includes the operating parameters of the distillation operation, the method further comprising generating, by the machine learning model, a predicted composition of a feedstock based on the operating parameters of the distillation operation, wherein the feedstock data includes the predicted composition of the feedstock.

101. The method of any of claims 82 to 100, wherein the catalytic reformer includes a continuous catalyst regeneration reactor and a catalyst regenerator, the method further comprising: predicting, by the machine learning model, an effect on catalysts of the continuous catalyst regeneration reactor based on the feedstock data; and generating, by the machine learning model, an adjustment to a parameter of the continuous catalyst regeneration reactor or the catalyst regenerator based on the predicted effect on catalysts of the continuous catalyst regeneration reactor and feedstock data.

102. The method of any of claims 82 to 100, wherein the catalytic reformer includes a fixed bed reactor, the method further comprising: predicting, by the machine learning model, a deactivation state of catalysts of the fixed bed reactor; and generating, by the machine learning model, an adjustment to a parameter of the fixed bed reactor based on the predicted deactivation state of catalysts of the fixed bed reactor and feedstock data.

103. The method of any of claims 82 to 102, further comprising: receiving operational data of a downstream aromatics recovery unit; supplying the product of the catalytic reformer to the downstream aromatics recovery unit; wherein the adjustment to the parameter of the upstream distillation operation supplying the feedstock to the catalytic reformer is further based on the operational data of the downstream aromatics recovery unit.

104. The method of claim 103, further comprising predicting an effect of the adjustment to the parameter of the upstream distillation operation supplying the feedstock to the catalytic reformer on the downstream aromatics recovery unit.

105. The method of claim 104, further comprising generating an adjustment to an operational parameter of the downstream aromatics recovery unit based on the predicted effect of the adjustment to the parameter of the upstream distillation operation supplying the feedstock to the catalytic reformer on the downstream aromatics recovery unit.

106. The method of any of claims 103 to 105, further comprising implementing the adjustment to an operational parameter of the downstream aromatics recovery unit.

107. The method of any of claims 105 to 106, wherein the operational parameter of the downstream aromatics recovery unit is a feed rate of the product of the catalytic reformer into the downstream aromatics recovery unit.

108. The method of any of claims 82 to 107, further comprising: receiving gasoline blending pool data at the machine learning model; andgenerating a target property of the product of the catalytic reformer based on the gasoline blending pool data.

109. The method of claim 108, wherein the gasoline blending pool data includes a customer specification for a blended formulation.

110. The method of any of claims 108 to 109, wherein the target property is a target octane-barrels of reformate.

111. The method of any of claims 108 to 110, wherein the target property is a target octane number of the reformate.

112. The method of any of claims 108 to 111, wherein the target property is measured or sampled proximate an exit of a reactor of the catalytic reformer.

113. The method of any of claims 108 to 111, wherein the target property is measured or sampled proximate an exit of a separator, a distillation unit, a stripper, or a stabilizer of the catalytic reformer.

114. The method of any of claims 108 to 113, wherein generating, by the machine learning model, the adjustment to a parameter of the catalytic reformer includes generating an adjustment to an operating temperature of a reactor of the catalytic reformer.

115. The method of claim 114, further comprising predicting, by the machine learning model, a deactivation state of catalysts of the reactor of the catalytic reformer, wherein generating an adjustment to an operating temperature of a reactor of the catalytic reformer is further based on the predicted deactivation state of catalysts of the reactor of the catalytic reformer.

116. The method of any of claims 108 to 115, further comprising implementing the generated adjustment to adjust the parameter of the catalytic reformer.

117. The method of any of claims 114 to 115, further comprising publishing a notification by the machine learning model that the target property of the product is not achievable based on constraints of the reactor of the catalytic reformer.

118. The method of any of claims 114 to 117, further comprising; generating, by the machine learning model, a possible property of the product closest to the target property of the product; and publishing a notification of a possible property of the product closest to the target property of the product.

119. The method of any of claims 114 to 118, further comprising generating, by the machine learning model, a determination to not implement the adjustment to a parameter of the catalytic reformer.

120. A memory including instructions that causes a processor to perform the instructions including the method of any of claims 82 to 119.

121. A computing system including a processor and a memory including instructions that cause the processor to perform the instructions, wherein the instructions include the method of any of claims 82 to 119.

122. A method compri sing : receiving product data indicative of a property of a product of a catalytic reformer; receiving feedstock data indicative of a feedstock for processing by the catalytic reformer, wherein the catalytic reformer includes a first reactor in series with a second reactor; generating, by a machine learning model, an adjustment to a parameter of the first reactor based on the feedstock data and the product data to improve the property of the product of the catalytic reformer while reducing yield loss of the product, wherein the machine learning model is trained with catalytic reformer historical data indicative of operating parameters, operational constraints, the feedstock, and products of the catalytic reformer; andgenerating, by the machine learning model, an adjustment to a parameter of the second reactor based on the adjustment to the parameter of the first reactor, feedstock data, and the product data.

123. The method of claim 122, wherein the catalytic reformer includes a third reactor, the method further comprising generating, by the machine learning model, an adjustment to a parameter of the third reactor based on the adjustment to the parameter of the first reactor, the parameter of the second reactor, feedstock data, and the product data.

124. The method of claim 123, wherein the catalytic reformer includes a fourth reactor, the method further comprising generating, by the machine learning model, an adjustment to a parameter of the fourth reactor based on the adjustment to the parameter of the first reactor, the parameter of the second reactor, the parameter of the third reactor, feedstock data, and the product data.

125. The method of any of claims 122 to 124, further comprising generating, by the machine learning model, a prediction of coke formation within the first reactor and the second reactor of the catalytic reformer, wherein generating, by the machine learning model, an adjustment to a parameter of the catalytic reformer is further based on the prediction of coke formation within the first reactor and the second reactor of the catalytic reformer.

126. The method of any of claims 122 to 125, wherein the parameter of the first reactor is one or more of a temperature within the first reactor, a pressure within the first reactor, or a feed rate of the feedstock into the first reactor.

127. The method of any of claims 122 to 126, wherein the parameter of the second reactor is one or more of a temperature within the second reactor, a pressure within the second reactor, or a feed rate of the feedstock into the second reactor.

128. The method of any of claims 123 to 127, wherein the parameter of the third reactor is one or more of a temperature within the third reactor, a pressure within the third reactor, or a feed rate of the feedstock into the third reactor.

129. The method of any of claims 124 to 128, wherein the parameter of the fourth reactor is one or more of a temperature within the fourth reactor, a pressure within the fourth reactor, or a feed rate of the feedstock into the fourth reactor.

130. The method of any of claims 124 to 128, wherein the adjustment to the parameter of the first reactor by the machine learning model is to reduce a pressure within the catalytic reformer to within five percent of an operational constraint of the catalytic reformer in order to maximize reformate yield.

131. The method of claim 130, wherein the adjustment to the parameter of the first reactor by the machine learning model is to reduce pressure within the catalytic reformer to within one percent of an operational constraint of the catalytic reformer in order to maximize reformate yield.

132. The method of any of claims 122 to 131, wherein the property of the product of the catalytic reformer is octane-barrels of reformate.

133. The method of claim 132, wherein the machine learning model is programmed to maximize octane-barrels through control of reactor severity.

134. The method of claim 133, further comprising generating, by the machine learning model, a predictive concave function of octane-barrels versus reactor severity, wherein generating, by the machine learning model, an adjustment to a parameter of the first reactor moves the octane-barrels of the product of the catalytic reformer closer to a predicted maximum of the predictive concave function, wherein generating, by the machine learning model, an adjustment to a parameter of the second reactor moves the octane-barrels of the product of the catalytic reformer closer to a predicted maximum of the predictive concave function.

135. The method of any of claims 122 to 129, wherein the property of the product of the catalytic reformer is a target octane number of the reformate.

136. The method of any of claims 122 to 135, further comprising:generating, by the machine learning model, a value for products of the first reactor, a value for products of the second reactor, a value for costs of operating the first reactor, a value for costs of operating the second reactor, and a value for the feedstock fed to the catalytic reformer; and generating a recommendation, by the machine learning model, to improve the value of the catalytic reformer operation based on the value for the products of the first reactor, the value for the products of the second reactor, the value for costs of operating the first reactor, the value for costs of operating the second reactor, and the value for the feedstock fed to the catalytic reformer.

137. The method of claim 136, further comprising: generating, by the machine learning model, a value for products of a third reactor and a value for costs of operating the third reactor, wherein generating the recommendation, by the machine learning model, to improve the value of the catalytic reformer operation is further based on the value for the products of the third reactor and the value for costs of operating the third reactor.

138. The method of claim 137, further comprising: generating, by the machine learning model, a value for products of a fourth reactor and a value for costs of operating the fourth reactor, wherein generating the recommendation, by the machine learning model, to improve the value of the catalytic reformer operation is further based on the value for the products of the fourth reactor and the value for costs of operating the fourth reactor.

139. The method of any of claims 136 to 138, further comprising publishing the recommendation.

140. The method of claims 139, further comprising implementing the recommendation.

141. The method of claims 140, wherein implementing the recommendation includes adjusting an operational parameter of the catalytic reformer to improve the value of the catalytic reformer.

142. The method of any of claims 136 to 141, further comprising accessing business data, wherein the values are based on current business data.

143. The method of claim 142, wherein the business data includes a current market price for each of the products of the catalytic reformer.

144. The method of any of claims 136 to 143, wherein the recommendation is constrained by a limitation of the catalytic reformer.

145. The method of claim 144, wherein the limitation is a deactivation state of catalysts within the catalytic reformer.

146. The method of claim 144, wherein the limitation is a heat generation rate of the catalytic reformer.

147. The method of claim 144, wherein the limitation is a lower operational temperature limit of the first reactor and a lower operational temperature limit of the second reactor.

148. The method of claim 147, wherein the lower operational temperature limit of the first reactor is different from the lower operational temperature limit of the second reactor.

149. The method of claim 148, further comprising generating, by the machine learning model, a prediction of the lower operational temperature limit of the first reactor and a prediction of the lower operational temperature limit of the second reactor based on the historical data.

150. The method of any of claims 147 to 149, wherein the limitation is a lower operational temperature limit of a third reactor.

151. The method of claim 150, wherein the limitation is a lower operational temperature limit of a fourth reactor.

152. The method of any of claims 136 to 146, wherein generating, by the machine learning model, an adjustment to a parameter of the catalytic reformer is based on an objective function, the objective function operating to improve the value of products of the catalytic reformer.

153. The method of claim 152, wherein the objective function includes the value for the products of the first reactor, the value for the products of the second reactor, the value for costs of operating the first reactor, the value for costs of operating the second reactor, and the value for the feedstock fed to the catalytic reformer.

154. The method of claim 153, wherein the objective function includes the value of the catalytic reformer operation is further based on the value for the products of a third reactor and the value for costs of operating the third reactor.

155. The method of claim 153, wherein the objective function includes the value of the catalytic reformer operation is further based on the value for the products of a fourth reactor and the value for costs of operating the fourth reactor.

156. The method of any of claims 152 to 155, further comprising installing, by the machine learning model, the objective function on a sub-controller of the catalytic reformer.

157. The method of any of claims 122 to 156, wherein the first reactor is a continuous catalyst regeneration reactor, the second reactor is a continuous catalyst regeneration reactor, and the catalytic reformer includes a catalyst regenerator, the method further comprising: predicting, by the machine learning model, an effect on catalysts of the first reactor based on the feedstock data; predicting, by the machine learning model, an effect on catalysts of the second reactor based on the feedstock data; and generating, by the machine learning model, an adjustment to a parameter of the first reactor, the second reactor, or the catalyst regenerator based on the predicted effect on catalysts of the first reactor, the predicted effect on catalysts of the second reactor, and feedstock data.

158. The method of claim 157, wherein a third reactor is a continuous catalyst regeneration reactor, the method further comprising: predicting, by the machine learning model, an effect on catalysts of the third reactor based on the feedstock data; and generating, by the machine learning model, an adjustment to a parameter of the third reactor or the catalyst regenerator based on the predicted effect on catalysts of the first reactor, the predicted effect on catalysts of the second reactor, the predicted effect on catalysts of the third reactor, and feedstock data.

159. The method of claim 158, wherein a fourth reactor is a continuous catalyst regeneration reactor, the method further comprising: predicting, by the machine learning model, an effect on catalysts of the fourth reactor based on the feedstock data; and generating, by the machine learning model, an adjustment to a parameter of the fourth reactor or the catalyst regenerator based on the predicted effect on catalysts of the first reactor, the predicted effect on catalysts of the second reactor, the predicted effect on catalysts of the third reactor, the predicted effect on catalysts of the fourth reactor, and feedstock data.

160. The method of any of claims 122 to 155, the method further comprising: predicting, by the machine learning model, a deactivation state of catalysts of the first reactor; predicting, by the machine learning model, a deactivation state of catalysts of the second reactor; and generating, by the machine learning model, an adjustment to a parameter of the first reactor or the second reactor based on the predicted deactivation state of catalysts of the first reactor, the predicted deactivation state of catalysts of the second reactor, and feedstock data.

161. The method of any of claims 122 to 160, further comprising: receiving operational data of a downstream aromatics recovery unit; supplying the product of the catalytic reformer to the downstream aromatics recovery unit;wherein the adjustment to the parameter of the first reactor is further based on the operational data of the downstream aromatics recovery unit, wherein the adjustment to the parameter of the second reactor is further based on the operational data of the downstream aromatics recovery unit.

162. The method of claim 161, further comprising predicting an effect of the adjustment to the parameter of the first reactor and the adjustment to the parameter of the second reactor on the downstream aromatics recovery unit.

163. The method of claim 162, further comprising generating an adjustment to an operational parameter of the downstream aromatics recovery unit based on the predicted effect of the adjustment to the parameter of the first reactor and the adjustment to the parameter of the second reactor on the downstream aromatics recovery unit.

164. The method of any of claims 161 to 163, further comprising implementing the adjustment to an operational parameter of the downstream aromatics recovery unit.

165. The method of any of claims 163 to 164, wherein the operational parameter of the downstream aromatics recovery unit is a feed rate of the product of the catalytic reformer into the downstream aromatics recovery unit.

166. The method of any of claims 122 to 165, further comprising: receiving gasoline blending pool data at the machine learning model; and generating a target property of the product of the catalytic reformer based on the gasoline blending pool data.

167. The method of claim 166, wherein the gasoline blending pool data includes a customer specification for a blended formulation.

168. The method of any of claims 166 to 167, wherein the target property is a target octane-barrels of reformate.

169. The method of any of claims 166 to 168, wherein the target property is a target octane number of the reformate.

170. The method of any of claims 166 to 169, wherein the target property is measured or sampled proximate an exit of a reactor of the catalytic reformer.

171. The method of any of claims 166 to 169, wherein the target property is measured or sampled proximate an exit of a separator, a distillation unit, a stripper, or a stabilizer of the catalytic reformer.

172. The method of any of claims 166 to 171, wherein generating, by the machine learning model, the adjustment to a parameter of the catalytic reformer includes generating an adjustment to an operating temperature of a reactor of the catalytic reformer.

173. The method of claim 172, further comprising predicting, by the machine learning model, a deactivation state of catalysts of the reactor of the catalytic reformer, wherein generating an adjustment to an operating temperature of a reactor of the catalytic reformer is further based on the predicted deactivation state of catalysts of the reactor of the catalytic reformer.

174. The method of any of claims 166 to 173, further comprising implementing the generated adjustment to adjust the parameter of the catalytic reformer.

175. The method of any of claims 172 to 173, further comprising publishing a notification by the machine learning model that the target property of the product is not achievable based on constraints of the reactor of the catalytic reformer.

176. The method of any of claims 172 to 175, further comprising; generating, by the machine learning model, a possible property of the product closest to the target property of the product; and publishing a notification of a possible property of the product closest to the target property of the product.

177. The method of any of claims 172 to 176, further comprising generating, by the machine learning model, a determination to not implement the adjustment to a parameter of the catalytic reformer.

178. A memory including instructions that causes a processor to perform the instructions including the method of any of claims 122 to 177.

179. A computing system including a processor and a memory including instructions that cause the processor to perform the instructions, wherein the instructions include the method of any of claims 122 to 177.

180. A method comprising: receiving product data indicative of a product component of a product of a catalytic reformer; receiving feedstock data indicative of a feedstock for processing by the catalytic reformer; receiving a target octane number for the reformate produced by the catalytic reformer; and generating, by a machine learning model, an adjustment to a parameter of the catalytic reformer based on the feedstock data and the product data to achieve the target octane number of the reformate, wherein the machine learning model is trained with catalytic reformer historical data indicative of operating parameters, operational constraints, the feedstock, and products of the catalytic reformer.

181. The method of claim 180, wherein the target octane number is received from a gasoline blending pool controller.

182. The method of any of claims 180 to 181, wherein the machine learning model is programmed to drive the catalytic reformer to achieve the target octane number of the reformate while seeking to improve yield of the reformate.

183. The method of claim 182, further comprising:generating a plurality of simulations, by the machine learning model, of the operation of the catalytic reformer based on the feedstock data, wherein each simulation has a different adjustment to an operational parameter; selecting, by the machine learning model, a simulation of the plurality of simulations, wherein the selected simulation has a highest simulated yield of reformate while achieving the target octane number of the reformate.

184. The method of claim 183, wherein generating, by the machine learning model, the adjustment to the parameter of the catalytic reformer is based on the selected simulation.

185. The method of any of claims 180 to 184, wherein the parameter of the catalytic reformer is a temperature of a reactor of the catalytic reactor.

186. The method of any of claims 180 to 185, wherein the machine learning model is programmed to maximize octane-barrels through control of reactor severity.

187. The method of claim 186, further comprising generating, by the machine learning model, a predictive concave function of octane-barrels versus reactor severity, wherein generating, by the machine learning model, the adjustment to a parameter of the catalytic reformer moves the octane number and yield of the product of the catalytic reformer closer to a predicted maximum of the predictive concave function.

188. The method of any of claims 180 to 187, wherein the parameter of the catalytic reformer is a pressure of a reactor of the catalytic reformer.

189. The method of claim 188, wherein the adjustment to the parameter of the catalytic reformer by the machine learning model is to reduce a pressure within the catalytic reformer to within five percent of an operational constraint of the catalytic reformer in order to maximize reformate yield.

190. The method of claim 189, wherein the adjustment to the parameter of the catalytic reformer by the machine learning model is to reduce pressure within the catalytic reformer to within one percent of an operational constraint of the catalytic reformer in order to maximize reformate yield.

191. The method of any of claims 180 to 190, wherein the catalytic reformer includes a first reactor in series with a second reactor, wherein generating the adjustment to a parameter of the catalytic reformer includes an adjustment to a parameter of the first reactor based on the feedstock data and the product data, the method further comprising generating, by the machine learning model, an adjustment to a parameter of the second reactor based on the adjustment to the parameter of the first reactor, feedstock data, and the product data.

192. The method of claim 191, wherein the catalytic reformer includes a third reactor, the method further comprising generating, by the machine learning model, an adjustment to a parameter of the third reactor based on the adjustment to the parameter of the first reactor, the parameter of the second reactor, feedstock data, and the product data.

193. The method of claim 192, wherein the catalytic reformer includes a fourth reactor, the method further comprising generating, by the machine learning model, an adjustment to a parameter of the fourth reactor based on the adjustment to the parameter of the first reactor, the parameter of the second reactor, the parameter of the third reactor, feedstock data, and the product data.

194. The method of any of claims 191 to 193, further comprising generating, by the machine learning model, a prediction of coke formation within the first reactor and the second reactor of the catalytic reformer, wherein generating, by the machine learning model, an adjustment to a parameter of the catalytic reformer is further based on the prediction of coke formation within the first reactor and the second reactor of the catalytic reformer.

195. The method of any of claims 191 to 194, wherein the parameter of the first reactor is one or more of a temperature within a reactor, a pressure within the reactor, or a feed rate of the feedstock into the reactor.

196. The method of any of claims 191 to 195, wherein the parameter of the second reactor is one or more of a temperature within a reactor, a pressure within the reactor, or a feed rate of the feedstock into the reactor.

197. The method of any of claims 192 to 196, wherein the parameter of the third reactor is one or more of a temperature within a reactor, a pressure within the reactor, or a feed rate of the feedstock into the reactor.

198. The method of any of claims 193 to 197, wherein the parameter of the fourth reactor is one or more of a temperature within a reactor, a pressure within the reactor, or a feed rate of the feedstock into the reactor.

199. The method of any of claims 191 to 198, further comprising: generating, by the machine learning model, a value for the products of the first reactor, a value for the products of the second reactor, a value for costs of operating the first reactor, a value for costs of operating the second reactor, and a value for the feedstock fed to the catalytic reformer; and generating a recommendation, by the machine learning model, to improve the value of the catalytic reformer operation based on the value for the products of the first reactor, the value for the products of the second reactor, the value for costs of operating the first reactor, the value for costs of operating the second reactor, and the value for the feedstock fed to the catalytic reformer.

200. The method of claim 199, further comprising: generating, by the machine learning model, a value for the products of a third reactor and a value for costs of operating the third reactor, wherein generating the recommendation, by the machine learning model, to improve the value of the catalytic reformer operation is further based on the value for the products of the third reactor and the value for costs of operating the third reactor.

201. The method of claim 200, further comprising: generating, by the machine learning model, a value for the products of a fourth reactor and a value for costs of operating the fourth reactor, wherein generating the recommendation, by the machine learning model, to improve the value of the catalytic reformer operation is further based on the value for the products of the fourth reactor and the value for costs of operating the fourth reactor.

202. The method of any of claims 199 to 201, further comprising publishing the recommendation.

203. The method of claims 202, further comprising implementing the recommendation.

204. The method of claims 203, wherein implementing the recommendation includes adjusting an operational parameter of the catalytic reformer to improve the value of the catalytic reformer.

205. The method of any of claims 199 to 204, further comprising accessing business data, wherein the values are based on current business data.

206. The method of claim 205, wherein the business data includes a current market price for each of the products of the catalytic reformer.

207. The method of any of claims 199 to 206, wherein the recommendation is constrained by a limitation of the catalytic reformer.

208. The method of claim 207, wherein the limitation is a deactivation state of catalysts within the catalytic reformer.

209. The method of claim 207, wherein the limitation is a heat generation rate of the catalytic reformer.

210. The method of claim 207, wherein the limitation is a lower operational temperature limit of the first reactor and a lower operational temperature limit of the second reactor.

211. The method of claim 210, wherein the lower operational temperature limit of the first reactor is different from the lower operational temperature limit of the second reactor.

212. The method of claim 211, further comprising generating, by the machine learning model, a prediction of the lower operational temperature limit of the first reactor and aprediction of the lower operational temperature limit of the second reactor based on the historical data.

213. The method of any of claims 210 to 212, wherein the limitation is a lower operational temperature limit of a third reactor.

214. The method of claim 213, wherein the limitation is a lower operational temperature limit of a fourth reactor.

215. The method of any of claims 199 to 214, wherein generating, by the machine learning model, an adjustment to a parameter of the catalytic reformer is based on an objective function, the objective function operating to improve the value of products of the catalytic reformer.

216. The method of claim 215, wherein the objective function includes the value for the products of the first reactor, the value for the products of the second reactor, the value for costs of operating the first reactor, the value for costs of operating the second reactor, and the value for the feedstock fed to the catalytic reformer.

217. The method of claim 216, wherein the objective function includes the value of the catalytic reformer operation is further based on the value for the products of a third reactor and the value for costs of operating the third reactor.

218. The method of claim 217, wherein the objective function includes the value of the catalytic reformer operation is further based on the value for the products of a fourth reactor and the value for costs of operating the fourth reactor.

219. The method of any of claims 215 to 218, further comprising installing, by the machine learning model, the objective function on a sub-controller of the catalytic reformer.

220. The method of any of claims 191 to 219, wherein the first reactor is a continuous catalyst regeneration reactor, the second reactor is a continuous catalyst regenerationreactor, and the catalytic reformer includes a catalyst regenerator, the method further comprising: predicting, by the machine learning model, an effect on catalysts of the first reactor based on the feedstock data; predicting, by the machine learning model, an effect on catalysts of the second reactor based on the feedstock data; and generating, by the machine learning model, an adjustment to a parameter of the first reactor, the second reactor, or the catalyst regenerator based on the predicted effect on catalysts of the first reactor, the predicted effect on catalysts of the second reactor, and feedstock data.

221. The method of claim 220, wherein a third reactor is a continuous catalyst regeneration reactor, the method further comprising: predicting, by the machine learning model, an effect on catalysts of the third reactor based on the feedstock data; and generating, by the machine learning model, an adjustment to a parameter of the third reactor or the catalyst regenerator based on the predicted effect on catalysts of the first reactor, the predicted effect on catalysts of the second reactor, the predicted effect on catalysts of the third reactor, and feedstock data.

222. The method of claim 221, wherein a fourth reactor is a continuous catalyst regeneration reactor, the method further comprising: predicting, by the machine learning model, an effect on catalysts of the fourth reactor based on the feedstock data; and generating, by the machine learning model, an adjustment to a parameter of the fourth reactor or the catalyst regenerator based on the predicted effect on catalysts of the first reactor, the predicted effect on catalysts of the second reactor, the predicted effect on catalysts of the third reactor, the predicted effect on catalysts of the fourth reactor, and feedstock data.

223. The method of any of claims 191 to 218, wherein the first reactor is a fixed bed reactor and the second reactor is a fixed bed reactor, the method further comprising: predicting, by the machine learning model, a deactivation state of catalysts of the first reactor;predicting, by the machine learning model, a deactivation state of catalysts of the second reactor; and generating, by the machine learning model, an adjustment to a parameter of the first reactor or the second reactor based on the predicted deactivation state of catalysts of the first reactor, the predicted deactivation state of catalysts of the second reactor, and feedstock data.

224. The method of any of claims 180 to 223, further comprising: receiving operational data of a downstream aromatics recovery unit; supplying the product of the catalytic reformer to the downstream aromatics recovery unit; wherein the adjustment to the parameter of the catalytic reformer is further based on the operational data of the downstream aromatics recovery unit.

225. The method of claim 224, further comprising predicting an effect of the adjustment to the parameter of the catalytic reformer on the downstream aromatics recovery unit.

226. The method of claim 225, further comprising generating an adjustment to an operational parameter of the downstream aromatics recovery unit based on the predicted effect of the adjustment to the parameter of the catalytic reformer on the downstream aromatics recovery unit.

227. The method of any of claims 224 to 226, further comprising implementing the adjustment to an operational parameter of the downstream aromatics recovery unit.

228. The method of any of claims 226 to 227, wherein the operational parameter of the downstream aromatics recovery unit is a feed rate of the product of the catalytic reformer into the downstream aromatics recovery unit.

229. The method of any of claims 180 to 228, further comprising: receiving gasoline blending pool data at the machine learning model; and generating a target property of the product of the catalytic reformer based on the gasoline blending pool data.

230. The method of claim 229, wherein the gasoline blending pool data includes a customer specification for a blended formulation.

231. The method of any of claims 229 to 230, wherein the target property is a target octane-barrels of reformate.

232. The method of any of claims 229 to 231, wherein the target property is a target octane number of the reformate.

233. The method of any of claims 229 to 232, wherein the target property is measured or sampled proximate an exit of a reactor of the catalytic reformer.

234. The method of any of claims 229 to 232, wherein the target property is measured or sampled proximate an exit of a separator, a distillation unit, a stripper, or a stabilizer of the catalytic reformer.

235. The method of any of claims 229 to 234, wherein generating, by the machine learning model, the adjustment to a parameter of the catalytic reformer includes generating an adjustment to an operating temperature of a reactor of the catalytic reformer.

236. The method of claim 235, further comprising predicting, by the machine learning model, a deactivation state of catalysts of the reactor of the catalytic reformer, wherein generating an adjustment to an operating temperature of a reactor of the catalytic reformer is further based on the predicted deactivation state of catalysts of the reactor of the catalytic reformer.

237. The method of any of claims 229 to 236, further comprising implementing the generated adjustment to adjust the parameter of the catalytic reformer.

238. The method of any of claims 235 to 236, further comprising publishing a notification by the machine learning model that the target property of the product is not achievable based on constraints of the reactor of the catalytic reformer.

239. The method of any of claims 235 to 238, further comprising; generating, by the machine learning model, a possible property of the product closest to the target property of the product; and publishing a notification of a possible property of the product closest to the target property of the product.

240. The method of any of claims 235 to 239, further comprising generating, by the machine learning model, a determination to not implement the adjustment to a parameter of the catalytic reformer.

241. A memory including instructions that causes a processor to perform the instructions including the method of any of claims 180 to 240.

242. A computing system including a processor and a memory including instructions that cause the processor to perform the instructions, wherein the instructions include the method of any of claims 180 to 240.

243. A method compri sing : receiving product data indicative of a property of a product of a catalytic reformer; receiving feedstock data indicative of a feedstock for processing by the catalytic reformer; receiving a target octane number for the reformate produced by the catalytic reformer; and generating, by a machine learning model, an adjustment to a parameter of the catalytic reformer based on the feedstock data and the product data to achieve an enhanced yield of hydrogen gas while achieving the target octane number of the reformate, wherein the machine learning model is trained with catalytic reformer historical data indicative of operating parameters, operational constraints, the feedstock, and products of the catalytic reformer.

244. The method of claim 243, wherein the target octane number is received from a gasoline blending pool controller.

245. The method of any of claims 243 to 244, wherein the parameter of the catalytic reformer is a temperature of a reactor of the catalytic reactor.

246. The method of any of claims 243 to 245, wherein the parameter of the catalytic reformer is a pressure of a reactor of the catalytic reactor.

247. The method of any of claims 243 to 246, wherein the adjustment to the parameter of the catalytic reformer by the machine learning model is to reduce a pressure within the catalytic reformer to within five percent of an operational constraint of the catalytic reformer in order to maximize reformate yield.

248. The method of claim 247, wherein the adjustment to the parameter of the catalytic reformer by the machine learning model is to reduce pressure within the catalytic reformer to within one percent of an operational constraint of the catalytic reformer in order to maximize reformate yield.

249. The method of any of claims 243 to 248, wherein the catalytic reformer includes a first reactor in series with a second reactor, wherein generating the adjustment to a parameter of the catalytic reformer includes an adjustment to a parameter of the first reactor based on the feedstock data and the product data, the method further comprising generating, by the machine learning model, an adjustment to a parameter of the second reactor based on the adjustment to the parameter of the first reactor, feedstock data, and the product data.

250. The method of claim 249, wherein the catalytic reformer includes a third reactor, the method further comprising generating, by the machine learning model, an adjustment to a parameter of the third reactor based on the adjustment to the parameter of the first reactor, the parameter of the second reactor, feedstock data, and the product data.

251. The method of claim 250, wherein the catalytic reformer includes a fourth reactor, the method further comprising generating, by the machine learning model, an adjustment to a parameter of the fourth reactor based on the adjustment to the parameter of the first reactor, the parameter of the second reactor, the parameter of the third reactor, feedstock data, and the product data.

252. The method of any of claims 249 to 251, further comprising generating, by the machine learning model, a prediction of coke formation within the first reactor and the second reactor of the catalytic reformer, wherein generating, by the machine learning model, an adjustment to a parameter of the catalytic reformer is further based on the prediction of coke formation within the first reactor and the second reactor of the catalytic reformer.

253. The method of any of claims 249 to 252, wherein the parameter of the first reactor is one or more of a temperature within a reactor, a pressure within the reactor, or a feed rate of the feedstock into the reactor.

254. The method of any of claims 249 to 253, wherein the parameter of the second reactor is one or more of a temperature within a reactor, a pressure within the reactor, or a feed rate of the feedstock into the reactor.

255. The method of any of claims 250 to 254, wherein the parameter of the third reactor is one or more of a temperature within a reactor, a pressure within the reactor, or a feed rate of the feedstock into the reactor.

256. The method of any of claims 251 to 255, wherein the parameter of the fourth reactor is one or more of a temperature within a reactor, a pressure within the reactor, or a feed rate of the feedstock into the reactor.

257. The method of any of claims 249 to 256, further comprising: generating, by the machine learning model, a value for the products of the first reactor, a value for the products of the second reactor, a value for costs of operating the first reactor, a value for costs of operating the second reactor, and a value for the feedstock fed to the catalytic reformer; and generating a recommendation, by the machine learning model, to improve the value of the catalytic reformer operation based on the value for the products of the first reactor, the value for the products of the second reactor, the value for costs of operating the first reactor, the value for costs of operating the second reactor, and the value for the feedstock fed to the catalytic reformer.

258. The method of claim 257, further comprising: generating, by the machine learning model, a value for the products of a third reactor and a value for costs of operating the third reactor, wherein generating the recommendation, by the machine learning model, to improve the value of the catalytic reformer operation is further based on the value for the products of the third reactor and the value for costs of operating the third reactor.

259. The method of claim 258, further comprising: generating, by the machine learning model, a value for the products of a fourth reactor and a value for costs of operating the fourth reactor, wherein generating the recommendation, by the machine learning model, to improve the value of the catalytic reformer operation is further based on the value for the products of the fourth reactor and the value for costs of operating the fourth reactor.

260. The method of any of claims 257 to 259, further comprising publishing the recommendation.

261. The method of claims 260, further comprising implementing the recommendation.

262. The method of claims 261, wherein implementing the recommendation includes adjusting an operational parameter of the catalytic reformer to improve the value of the catalytic reformer.

263. The method of any of claims 257 to 262, further comprising accessing business data, wherein the values are based on current business data.

264. The method of claim 263, wherein the business data includes a current market price for each of the products of the catalytic reformer.

265. The method of any of claims 257 to 264, wherein the recommendation is constrained by a limitation of the catalytic reformer.

266. The method of claim 265, wherein the limitation is a deactivation state of catalysts within the catalytic reformer.

267. The method of claim 265, wherein the limitation is a heat generation rate of the catalytic reformer.

268. The method of claim 265, wherein the limitation is a lower operational temperature limit of the first reactor and a lower operational temperature limit of the second reactor.

269. The method of claim 268, wherein the lower operational temperature limit of the first reactor is different from the lower operational temperature limit of the second reactor.

270. The method of claim 269, further comprising generating, by the machine learning model, a prediction of the lower operational temperature limit of the first reactor and a prediction of the lower operational temperature limit of the second reactor based on the historical data.

271. The method of any of claims 268 to 270, wherein the limitation is a lower operational temperature limit of a third reactor.

272. The method of claim 271, wherein the limitation is a lower operational temperature limit of a fourth reactor.

273. The method of any of claims 257 to 272, wherein generating, by the machine learning model, an adjustment to a parameter of the catalytic reformer is based on an objective function, the objective function operating to improve the value of products of the catalytic reformer.

274. The method of claim 273, wherein the objective function includes the value for the products of the first reactor, the value for the products of the second reactor, the value for costs of operating the first reactor, the value for costs of operating the second reactor, and the value for the feedstock fed to the catalytic reformer.

275. The method of claim 274, wherein the objective function includes the value of the catalytic reformer operation is further based on the value for the products of a third reactor and the value for costs of operating the third reactor.

276. The method of claim 275, wherein the objective function includes the value of the catalytic reformer operation is further based on the value for the products of a fourth reactor and the value for costs of operating the fourth reactor.

277. The method of any of claims 249 to 276, wherein the first reactor is a continuous catalyst regeneration reactor, the second reactor is a continuous catalyst regeneration reactor, and the catalytic reformer includes a catalyst regenerator, the method further comprising: predicting, by the machine learning model, an effect on catalysts of the first reactor based on the feedstock data; predicting, by the machine learning model, an effect on catalysts of the second reactor based on the feedstock data; and generating, by the machine learning model, an adjustment to a parameter of the first reactor, the second reactor, or the catalyst regenerator based on the predicted effect on catalysts of the first reactor, the predicted effect on catalysts of the second reactor, and feedstock data.

278. The method of claim 277, wherein a third reactor is a continuous catalyst regeneration reactor, the method further comprising: predicting, by the machine learning model, an effect on catalysts of the third reactor based on the feedstock data; and generating, by the machine learning model, an adjustment to a parameter of the third reactor or the catalyst regenerator based on the predicted effect on catalysts of the first reactor, the predicted effect on catalysts of the second reactor, the predicted effect on catalysts of the third reactor, and feedstock data.

279. The method of claim 278, wherein a fourth reactor is a continuous catalyst regeneration reactor, the method further comprising: predicting, by the machine learning model, an effect on catalysts of the fourth reactor based on the feedstock data; and generating, by the machine learning model, an adjustment to a parameter of the fourth reactor or the catalyst regenerator based on the predicted effect on catalysts of the first reactor, the predicted effect on catalysts of the second reactor, the predicted effect oncatalysts of the third reactor, the predicted effect on catalysts of the fourth reactor, and feedstock data.

280. The method of any of claims 249 to 279, wherein the first reactor is a fixed bed reactor and the second reactor is a fixed bed reactor, the method further comprising: predicting, by the machine learning model, a deactivation state of catalysts of the first reactor; predicting, by the machine learning model, a deactivation state of catalysts of the second reactor; and generating, by the machine learning model, an adjustment to a parameter of the first reactor or the second reactor based on the predicted deactivation state of catalysts of the first reactor, the predicted deactivation state of catalysts of the second reactor, and feedstock data.

281. The method of any of claims 243 to 280, further comprising: generating a plurality of simulations, by the machine learning model, of the operation of the catalytic reformer based on the feedstock data, wherein each simulation has a different adjustment to an operational parameter; selecting, by the machine learning model, a simulation of the plurality of simulations, wherein the selected simulation has a highest simulated yield of hydrogen while achieving the target octane number of the reformate.

282. The method of claim 281, wherein generating, by the machine learning model, the adjustment to the parameter of the catalytic reformer is based on the selected simulation.

283. The method of any of claims 243 to 282, further comprising: receiving operational data of a downstream aromatics recovery unit; supplying the product of the catalytic reformer to the downstream aromatics recovery unit; wherein the adjustment to the parameter of the catalytic reformer is further based on the operational data of the downstream aromatics recovery unit.

284. The method of claim 283, further comprising predicting an effect of the adjustment to the parameter of the catalytic reformer on the downstream aromatics recovery unit.

285. The method of claim 284, further comprising generating an adjustment to an operational parameter of the downstream aromatics recovery unit based on the predicted effect of the adjustment to the parameter of the catalytic reformer on the downstream aromatics recovery unit.

286. The method of any of claims 283 to 285, further comprising implementing the adjustment to an operational parameter of the downstream aromatics recovery unit.

287. The method of any of claims 285 to 286, wherein the operational parameter of the downstream aromatics recovery unit is a feed rate of the product of the catalytic reformer into the downstream aromatics recovery unit.

288. The method of any of claims 243 to 287, further comprising: receiving gasoline blending pool data at the machine learning model; and generating the target property of the product of the catalytic reformer based on the gasoline blending pool data.

289. The method of claim 288, wherein the gasoline blending pool data includes a customer specification for a blended formulation.

290. The method of any of claims 288 to 289, wherein the target property is a target octane-barrels of reformate.

291. The method of any of claims 288 to 290, wherein the target property is a target octane number of the reformate.

292. The method of any of claims 288 to 291, wherein the target property is measured or sampled proximate an exit of a reactor of the catalytic reformer.

293. The method of any of claims 288 to 291, wherein the target property is measured or sampled proximate an exit of a separator, a distillation unit, a stripper, or a stabilizer of the catalytic reformer.

294. The method of any of claims 288 to 293, wherein generating, by the machine learning model, the adjustment to a parameter of the catalytic reformer includes generating an adjustment to an operating temperature of a reactor of the catalytic reformer.

295. The method of claim 294, further comprising predicting, by the machine learning model, a deactivation state of catalysts of the reactor of the catalytic reformer, wherein generating an adjustment to an operating temperature of a reactor of the catalytic reformer is further based on the predicted deactivation state of catalysts of the reactor of the catalytic reformer.

296. The method of any of claims 288 to 295, further comprising implementing the generated adjustment to adjust the parameter of the catalytic reformer.

297. The method of any of claims 294 to 295, further comprising publishing a notification by the machine learning model that the target property of the product is not achievable based on constraints of the reactor of the catalytic reformer.

298. The method of any of claims 294 to 297, further comprising; generating, by the machine learning model, a possible property of the product closest to the target property of the product; and publishing a notification of a possible property of the product closest to the target property of the product.

299. The method of any of claims 294 to 298, further comprising generating, by the machine learning model, a determination to not implement the adjustment to a parameter of the catalytic reformer.

300. A memory including instructions that causes a processor to perform the instructions including the method of any of claims 243 to 299.

301. A computing system including a processor and a memory including instructions that cause the processor to perform the instructions, wherein the instructions include the method of any of claims 243 to 299.

302. A method compri sing : receiving one or more of (a) operational data of an aromatics recovery unit indicative of operational parameters of the aromatics recovery unit and including sensor data from one or more sensors disposed to measure a parameter of the aromatics recovery unit, (b) product data indicative of a property or component of a product from a process of the aromatics recovery unit or (c) feedstock data of a reformate being fed into the aromatics recovery unit; determining a change in the one or more of the operational data, the product data, or the feedstock data; generating, by a machine learning model, a prediction of an effect of the change on the operation of the aromatics recovery unit; and generating, by a machine learning model, an adjustment to an operating parameter of the aromatics recovery unit based on the predicted effect of the change, wherein the machine learning model is trained on historical data including operational data and product data of the aromatics recovery unit.

303. The method of claim 302, further comprising implementing the adjustment to the operating parameter of the aromatics recovery unit.

304. The method of any of claims 302 to 303, further comprising publishing the predicted effect of the change to a user.

305. The method of any of claims 302 to 304, further comprising receiving an instruction to implement the adjustment to the operating parameter of the aromatics recovery unit.

306. The method of any of claims 302 to 305, wherein the machine learning model is trained on historical data including feedstock data indicative of feedstock fed into the aromatics recovery unit.

307. The method of any of claims 302 to 306, wherein the adjustment is an adjustment to a feed rate of the reformate.

308. The method of any of claims 302 to 307, further comprising:generating a plurality of simulations, by the machine learning model, of the operation of the aromatics recovery unit based on the change in the operational data, wherein each simulation has a different adjustment to an operational parameter of the aromatics recovery unit; selecting, by the machine learning model, a simulation of the plurality of simulations, wherein the selected simulation has a highest production of one or more of benzene, toluene, or xylene while not exceeding an operational constraint of the aromatics recovery unit.

309. The method of claim 308, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit is based on the selected simulation.

310. The method of any of claims 302 to 309, further comprising generating, by the machine learning model, a prediction of solvent loading in an extractor column of the aromatics recovery unit based on the operating data and the product data, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit is based on the predicted solvent loading.

311. The method of any of claims 302 to 310, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to a feed rate of solvent into an extractor column of the aromatics recovery unit.

312. The method of any of claims 302 to 311, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to an operating parameter of a solvent recovery column of the aromatics recovery unit.

313. The method of any of claims 302 to 312, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to a flow rate of stripper recycle from a stripper of the aromatics recovery unit into an extractor column of the aromatics recovery unit.

314. The method of any of claims 302 to 313, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to an operating parameter of a solvent recovery column of the aromatics recovery unit.

315. The method of claim 314, further comprising generating, by the machine learning model, a prediction of solvent quality based on the operating data and the product data, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit is based on the predicted solvent quality.

316. The method of claim 315, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit is a request for fresh solvent to be added to the aromatics recovery unit.

317. The method of any of claims 302 to 316, further comprising generating, by the machine learning model, a prediction of a process constraint based on the operating data.

318. The method of claim 317, wherein the prediction of the process constraint is generated on a regular periodic basis.

319. The method of claim 317, wherein the prediction of the process constraint is generated when the change in the one or more of the operational data, the product data, or the feedstock data is determined.

320. The method of any of claims 302 to 319, further comprising generating, by the machine learning model, a prediction of fouling within a process of the aromatics recovery unit, based on the operational data.

321. The method of claim 320, further comprising generating, by the machine learning model, a prediction of a process constraint based on the prediction of fouling within a process of the aromatics recovery unit.

322. The method of claim 321, wherein the prediction of the process constraint is generated on a regular periodic basis.

323. The method of claims 321, wherein the prediction of the process constraint is generated when the change in the one or more of the operational data, the product data, or the feedstock data is determined.

324. The method of any of claims 317, 318, 319, 321, 322, or 323, further comprising generating, by the machine learning model, a margin away from the process constraint based on the historical data and a confidence interval that the process of the aromatics recovery unit will not exceed the process constraint during operation of the aromatics recovery unit.

325. The method of claim 324, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit is constrained by the margin away from the process constraint.

326. The method of any of claims 317, 318, 319, 321, 322, 323, 324, or 325, further comprising generating, by the machine learning model, an operating target for a process of the aromatics recovery unit based on the process constraint.

327. The method of claim 326, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment moves operation of the process to the operating target.

328. The method of any of claims 302 to 327, further comprising generating, by the machine learning model, a target feed rate of feedstock into the aromatics recovery unit, based on the operational data.

329. The method of claim 328, further comprising generating, by the machine learning model, a prediction of a maximum feed rate of feedstock into the aromatics recovery unit, wherein the target feed rate of feedstock into the aromatics recovery unit is based on the predicted maximum feed rate of feedstock.

330. The method of claim 329, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes anadjustment of feed rate of feedstock into the aromatics recovery unit to the target feed rate of feedstock into the aromatics recovery unit.

331. A memory including instructions that causes a processor to perform the instructions including the method of any of claims 302 to 330.

332. A computing system including a processor and a memory including instructions that cause the processor to perform the instructions, wherein the instructions include the method of any of claims 302 to 330.

333. A system for enhancing aromatics recovery for an aromatics recovery operation, the system comprising: an aromatics recovery unit to receive a feedstock and capture aromatics within the feedstock; a plurality of sensors to measure a parameter associated with the aromatics recovery unit and each positioned at one of (a) proximate the aromatics recovery unit or (b) within the aromatics recovery unit; a plurality of refinery operation control devices each positioned proximate and downstream or upstream of the aromatics recovery unit and to control aspects of fluid flowing to or from the aromatics recovery unit; one or more sample collection assemblies to collect samples of the fluid associated with the aromatics recovery unit; one or more sample analysis assemblies to analyze each collected sample to provide properties of collected samples; and an aromatics recovery controller in signal communication with the aromatics recovery unit, the plurality of sensors, the plurality of refinery operation control devices, and the one or more sample analysis assemblies, and storing a trained machine learning model, the aromatics recovery controller configured to: determine an output including predicted properties of the feedstock, parameter settings of the plurality of refinery operation control devices and the aromatics recovery unit based on application of one or more of (i) data measured by the plurality of sensors, (ii) data corresponding to analysis from the one or more sample analysis assemblies, or (iii) a target product and corresponding target properties to the trained machine learning model, andadjust one or more of an amount of feedstock or type of feedstock and parameters associated with the plurality of refinery operation control devices and aromatics recovery unit based on the output to enhance capture of aromatics.

334. A method compri sing : receiving one or more of (a) operational data of an aromatics recovery unit indicative of operational parameters of the aromatics recovery unit and including sensor data from one or more sensors disposed to measure a parameter of the aromatics recovery unit, (b) product data indicative of a property or component of a product from a process of the aromatics recovery unit or (c) feedstock data of a reformate being fed into the aromatics recovery unit; generating, by a machine learning model, a prediction of a process constraint based on the operating data; and generating, by the machine learning model, an adjustment to an operating parameter of the aromatics recovery unit based on the predicted process constraint, wherein the machine learning model is trained on historical data including operational data and product data of the aromatics recovery unit.

335. The method of claim 334, further comprising implementing the adjustment to the operating parameter of the aromatics recovery unit.

336. The method of any of claims 334 to 335, further comprising publishing the generated adjustment to the operating parameter of the aromatics recovery unit to a user.

337. The method of any of claims 334 to 336, further comprising receiving an instruction to implement the adjustment to the operating parameter of the aromatics recovery unit.

338. The method of any of claims 334 to 337, wherein the machine learning model is trained on historical data including feedstock data indicative of feedstock fed into the aromatics recovery unit.

339. The method of any of claims 334 to 338, wherein the adjustment is an adjustment to a feed rate of the reformate.

340. The method of any of claims 334 to 339, further comprising: generating a plurality of simulations, by the machine learning model, of the operation of the aromatics recovery unit based on the predicted process constraint in the operational data, wherein each simulation has a different adjustment to an operational parameter of the aromatics recovery unit; selecting, by the machine learning model, a simulation of the plurality of simulations, wherein the selected simulation has a highest production of one or more of benzene, toluene, or xylene while not exceeding an operational constraint of the aromatics recovery unit.

341. The method of claim 340, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit is based on the selected simulation.

342. The method of any of claims 334 to 341, further comprising generating, by the machine learning model, a prediction of solvent loading in an extractor column of the aromatics recovery unit based on the operating data and the product data, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit is based on the predicted solvent loading.

343. The method of any of claims 334 to 342, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to a feed rate of solvent into an extractor column of the aromatics recovery unit.

344. The method of any of claims 334 to 343, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to an operating parameter of a solvent recovery column of the aromatics recovery unit.

345. The method of any of claims 334 to 344, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to a flow rate of stripper recycle from a stripper of the aromatics recovery unit into an extractor column of the aromatics recovery unit.

346. The method of any of claims 334 to 345, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to an operating parameter of a solvent recovery column of the aromatics recovery unit.

347. The method of claim 346, further comprising generating, by the machine learning model, a prediction of solvent quality based on the operating data and the product data, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit is based on the predicted solvent quality.

348. The method of claim 347, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit is a request for fresh solvent to be added to the aromatics recovery unit.

349. The method of any of claims 334 to 348, wherein the prediction of the process constraint is generated on a regular periodic basis.

350. The method of any of claims 334 to 349, wherein the prediction of the process constraint is generated when a change in the one or more of the operational data, the product data, or the feedstock data is determined.

351. The method of any of claims 334 to 350, further comprising generating, by the machine learning model, a prediction of fouling within a process of the aromatics recovery unit, based on the operational data.

352. The method of claim 351, wherein generating, by the machine learning model, the prediction of the process constraint is based on the prediction of fouling within a process of the aromatics recovery unit.

353. The method of any of claims 334 to 352, further comprising generating, by the machine learning model, a margin away from the process constraint based on the historical data and a confidence interval that the process of the aromatics recovery unit will not exceed the process constraint during operation of the aromatics recovery unit.

354. The method of claim 353, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit is constrained by the margin away from the process constraint.

355. The method of any of claims 334 to 354, further comprising generating, by the machine learning model, an operating target for a process of the aromatics recovery unit based on the process constraint.

356. The method of claim 355, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment moves operation of the process to the operating target.

357. The method of any of claims 334 to 356, further comprising generating, by the machine learning model, a target feed rate of feedstock into the aromatics recovery unit, based on the operational data.

358. The method of claim 357, further comprising generating, by the machine learning model, a prediction of a maximum feed rate of feedstock into the aromatics recovery unit, wherein the target feed rate of feedstock into the aromatics recovery unit is based on the predicted maximum feed rate of feedstock.

359. The method of claim 358, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment of feed rate of feedstock into the aromatics recovery unit to the target feed rate of feedstock into the aromatics recovery unit.

360. A memory including instructions that causes a processor to perform the instructions including the method of any of claims 334 to 359.

361. A computing system including a processor and a memory including instructions that cause the processor to perform the instructions, wherein the instructions include the method of any of claims 334 to 359.

362. A system for enhancing aromatics recovery for an aromatics recovery operation, the system comprising: an aromatics recovery unit to receive a feedstock and capture aromatics within the feedstock; a plurality of sensors to measure a parameter associated with the aromatics recovery unit and each positioned at one of (a) proximate the aromatics recovery unit or (b) within the aromatics recovery unit; a plurality of refinery operation control devices each positioned proximate and downstream or upstream of the aromatics recovery unit and to control aspects of fluid flowing to or from the aromatics recovery unit; one or more sample collection assemblies to collect samples of the fluid associated with the aromatics recovery unit; one or more sample analysis assemblies to analyze each collected sample to provide properties of collected samples; and an aromatics recovery controller in signal communication with the aromatics recovery unit, the plurality of sensors, the plurality of refinery operation control devices, and the one or more sample analysis assemblies, and storing a trained machine learning model, the aromatics recovery controller configured to: determine an output including predicted properties of the feedstock, parameter settings of the plurality of refinery operation control devices, and the aromatics recovery unit based on application of one or more of (i) data measured by the plurality of sensors, (ii) data corresponding to analysis from the one or more sample analysis assemblies, or (iii) a target product and corresponding target properties to the trained machine learning model, and adjust one or more of an amount of feedstock or type of feedstock and parameters associated with the plurality of refinery operation control devices and aromatics recovery unit based on the output to enhance capture of aromatics.

363. A method compri sing : receiving one or more of (a) operational data of an aromatics recovery unit indicative of operational parameters of the aromatics recovery unit and including sensor data from one or more sensors disposed to measure a parameter of the aromatics recovery unit, (b) product data indicative of a property or component of a product from a process ofthe aromatics recovery unit or (c) feedstock data of a reformate being fed into the aromatics recovery unit; generating, by a machine learning model, a prediction of solvent quality based on the operating data; and generating, by the machine learning model, an adjustment to an operating parameter of the aromatics recovery unit based on the predicted solvent quality, wherein the machine learning model is trained on historical data including operational data and product data of the aromatics recovery unit.

364. The method of claim 363, further comprising implementing the adjustment to the operating parameter of the aromatics recovery unit.

365. The method of any of claims 363 to 364, further comprising publishing the generated adjustment to the operating parameter of the aromatics recovery unit to a user.

366. The method of any of claims 363 to 365, further comprising receiving an instruction to implement the adjustment to the operating parameter of the aromatics recovery unit.

367. The method of any of claims 363 to 366, wherein the machine learning model is trained on historical data including feedstock data indicative of feedstock fed into the aromatics recovery unit.

368. The method of any of claims 363 to 367, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to a feed rate of the feedstock.

369. The method of any of claims 363 to 368, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to an operational parameter of a solvent regenerator.

370. The method of any of claims 363 to 369, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to an operational parameter of a solvent recovery column.

371. The method of any of claims 363 to 370, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to an operational parameter of a extractor column.

372. The method of claim 371, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to a feed rate of solvent into an extractor column of the aromatics recovery unit.

373. The method of any of claims 363 to 372, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to a flow rate of stripper recycle from a stripper of the aromatics recovery unit into a extractor column of the aromatics recovery unit.

374. The method of any of claims 363 to 373, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to an operational parameter of a stripper.

375. The method of any of claims 363 to 374, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes a request for fresh solvent.

376. The method of any of claims 363 to 375, further comprising: generating a plurality of simulations, by the machine learning model, of the operation of the aromatics recovery unit based on the predicted solvent quality, wherein each simulation has a different adjustment to an operational parameter of the aromatics recovery unit; selecting, by the machine learning model, a simulation of the plurality of simulations, wherein the selected simulation has a highest production of one or more of benzene, toluene, or xylene while not exceeding an operational constraint of the aromatics recovery unit.

377. The method of claim 376, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit is based on the selected simulation.

378. The method of any of claims 363 to 377, further comprising generating, by the machine learning model, a prediction of solvent loading in an extractor column of the aromatics recovery unit based on the operating data and the product data, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit is based on the predicted solvent loading.

379. The method of any of claims 363 to 378, further comprising generating, by the machine learning model, a prediction of a process constraint based on the operating data.

380. The method of claim 379, wherein the prediction of the process constraint is generated on a regular periodic basis.

381. The method of claim 380, wherein the prediction of the process constraint is generated when a change in the one or more of the operational data, the product data, or the feedstock data is determined.

382. The method of any of claims 363 to 381, further comprising generating, by the machine learning model, a prediction of fouling within a process of the aromatics recovery unit, based on the operational data.

383. The method of claim 382, further comprising generating, by the machine learning model, a prediction of a process constraint based on the prediction of fouling within a process of the aromatics recovery unit.

384. The method of claim 383, wherein the prediction of the process constraint is generated on a regular periodic basis.

385. The method of claims 383, wherein the prediction of the process constraint is generated when a change in the one or more of the operational data, the product data, or the feedstock data is determined.

386. The method of any of claims 379 to 385, further comprising generating, by the machine learning model, a margin away from the process constraint based on the historical data and a confidence interval that the process of the aromatics recovery unit will not exceed the process constraint during operation of the aromatics recovery unit.

387. The method of claim 386, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit is constrained by the margin away from the process constraint.

388. The method of any of claims 379 to 387, further comprising generating, by the machine learning model, an operating target for a process of the aromatics recovery unit based on the process constraint.

389. The method of claim 388, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment moves operation of the process to the operating target.

390. The method of any of claims 379 to 389, further comprising generating, by the machine learning model, a target feed rate of feedstock into the aromatics recovery unit, based on the operational data.

391. The method of claim 390, further comprising generating, by the machine learning model, a prediction of a maximum feed rate of feedstock into the aromatics recovery unit, wherein the target feed rate of feedstock into the aromatics recovery unit is based on the predicted maximum feed rate of feedstock.

392. The method of claim 391, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment of feed rate of feedstock into the aromatics recovery unit to the target feed rate of feedstock into the aromatics recovery unit.

393. A memory including instructions that causes a processor to perform the instructions including the method of any of claims 363 to 392.

394. A computing system including a processor and a memory including instructions that cause the processor to perform the instructions, wherein the instructions include the method of any of claims 363 to 392.

395. A method comprising: receiving one or more of (a) operational data of an aromatics recovery unit indicative of operational parameters of the aromatics recovery unit and including sensor data from one or more sensors disposed to measure a parameter of the aromatics recovery unit, (b) product data indicative of a property or component of a product from a process of the aromatics recovery unit or (c) feedstock data of a reformate being fed into the aromatics recovery unit; generating, by a machine learning model, a prediction of solvent loading based on the operating data; and generating, by the machine learning model, an adjustment to an operating parameter of the aromatics recovery unit to improve solvent loading based on the predicted solvent loading, wherein the machine learning model is trained on historical data including operational data and product data of the aromatics recovery unit.

396. The method of claim 395, further comprising implementing the adjustment to the operating parameter of the aromatics recovery unit.

397. The method of any of claims 395 to 396, further comprising publishing the generated adjustment to the operating parameter of the aromatics recovery unit to a user.

398. The method of any of claims 395 to 397, further comprising receiving an instruction to implement the adjustment to the operating parameter of the aromatics recovery unit.

399. The method of any of claims 395 to 398, wherein the machine learning model is trained on historical data including feedstock data indicative of feedstock fed into the aromatics recovery unit.

400. The method of any of claims 395 to 399, further comprising generating, by a machine learning model, a prediction of solvent quality based on the operating data,wherein generating, by the machine learning model, the adjustment to the operating parameter of the aromatics recovery unit to improve solvent loading is further based on the predicted solvent quality.

401. The method of any of claims 395 to 400, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to a feed rate of the feedstock.

402. The method of any of claims 395 to 401, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to an operational parameter of a solvent regenerator.

403. The method of any of claims 395 to 402, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to an operational parameter of a solvent recovery column.

404. The method of any of claims 395 to 403, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to an operational parameter of a extractor column.

405. The method of claim 404, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to a feed rate of solvent into an extractor column of the aromatics recovery unit.

406. The method of any of claims 395 to 405, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to a flow rate of stripper recycle from a stripper of the aromatics recovery unit into a extractor column of the aromatics recovery unit.

407. The method of any of claims 395 to 406, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment to an operational parameter of a stripper.

408. The method of any of claims 395 to 407, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes a request for fresh solvent.

409. The method of any of claims 395 to 408, further comprising: generating a plurality of simulations, by the machine learning model, of the operation of the aromatics recovery unit based on the predicted solvent loading, wherein each simulation has a different adjustment to an operational parameter of the aromatics recovery unit; selecting, by the machine learning model, a simulation of the plurality of simulations, wherein the selected simulation has a highest production of one or more of benzene, toluene, or xylene while not exceeding an operational constraint of the aromatics recovery unit.

410. The method of claim 409, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit is based on the selected simulation.

411. The method of any of claims 395 to 410, further comprising generating, by the machine learning model, a prediction of a process constraint based on the operating data.

412. The method of claim 411, wherein the prediction of the process constraint is generated on a regular periodic basis.

413. The method of claim 412, wherein the prediction of the process constraint is generated when a change in the one or more of the operational data, the product data, or the feedstock data is determined.

414. The method of any of claims 395 to 413, further comprising generating, by the machine learning model, a prediction of fouling within a process of the aromatics recovery unit, based on the operational data.

415. The method of claim 414, further comprising generating, by the machine learning model, a prediction of a process constraint based on the prediction of fouling within a process of the aromatics recovery unit.

416. The method of claim 415, wherein the prediction of the process constraint is generated on a regular periodic basis.

417. The method of claims 415, wherein the prediction of the process constraint is generated when a change in the one or more of the operational data, the product data, or the feedstock data is determined.

418. The method of any of claims 411 to 417, further comprising generating, by the machine learning model, a margin away from the process constraint based on the historical data and a confidence interval that the process of the aromatics recovery unit will not exceed the process constraint during operation of the aromatics recovery unit.

419. The method of claim 418, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit is constrained by the margin away from the process constraint.

420. The method of any of claims 411 to 419, further comprising generating, by the machine learning model, an operating target for a process of the aromatics recovery unit based on the process constraint.

421. The method of claim 420, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment moves operation of the process to the operating target.

422. The method of any of claims 411 to 421, further comprising generating, by the machine learning model, a target feed rate of feedstock into the aromatics recovery unit, based on the operational data.

423. The method of claim 422, further comprising generating, by the machine learning model, a prediction of a maximum feed rate of feedstock into the aromatics recovery unit,wherein the target feed rate of feedstock into the aromatics recovery unit is based on the predicted maximum feed rate of feedstock.

424. The method of claim 423, wherein generating, by the machine learning model, the adjustment to the parameter of the aromatics recovery unit, the adjustment includes an adjustment of feed rate of feedstock into the aromatics recovery unit to the target feed rate of feedstock into the aromatics recovery unit.

425. A memory including instructions that causes a processor to perform the instructions including the method of any of claims 395 to 424.

426. A computing system including a processor and a memory including instructions that cause the processor to perform the instructions, wherein the instructions include the method of any of claims 395 to 424.

Images