A hydrogen network optimization method and system based on a gaussian process regression surrogate model

Abstract

The application discloses a hydrogen network optimization method and system based on a Gaussian process regression agent model, and belongs to the field of petroleum chemical process system engineering and technology.The method first constructs Gaussian process regression agent models of desulfurization, denitrogenation and dearomatics reaction kinetics for different hydrogenation devices, respectively, to accurately predict hydrogen consumption and product impurity content with operating temperature and pressure as input;then embeds the agent models of the devices into a hydrogen network optimization model to establish an integrated optimization problem of minimum total annual cost;finally, solves the optimization problem by using an improved differential evolution algorithm combined with a three-stage search strategy of global exploration-local development-convergence verification to obtain optimal operating parameters.The application shortens the optimization calculation time to about 90 seconds, solves the solving difficulty of traditional methods coupled with complex kinetics, realizes collaborative optimization of hydrogenation operation and hydrogen network operation while ensuring product quality, reduces hydrogen consumption and total annual cost of the system, and significantly improves the economic benefit and operation efficiency of the hydrogen system of a refinery.

Description

Technical Field

[0001] This invention belongs to the field of petrochemical process systems engineering and technology, and relates to a hydrogen network optimization method and system based on a Gaussian process regression surrogate model. Background Technology

[0002] Hydrogen is a critical resource for modern oil refineries. With the increasing weight and quality of crude oil being processed, and with increasingly stringent environmental regulations demanding cleaner fuels, refinery hydrogen consumption has risen significantly, making hydrogen cost the second largest operating cost after crude oil. Therefore, optimizing hydrogen systems to reduce hydrogen consumption is crucial for improving refinery economic efficiency.

[0003] Refinery hydrogen systems are typically modeled as a hydrogen network, including hydrogen sources and hydrogen sinks. Existing optimization methods mainly include pinch methods and mathematical programming. Pinch methods are intuitive but struggle to handle complex constraints; mathematical programming, while accurately describing the system, is complex to solve, especially when detailed reaction kinetics of the hydrogenation unit need to be coupled for accurate hydrogen consumption calculations, where the high nonlinearity of the model makes solving extremely difficult.

[0004] In existing technologies, although some studies have attempted to couple some hydrogenation kinetics to hydrogen network optimization, the following shortcomings exist: First, usually only the desulfurization reaction kinetics are considered, and multiple hydrogen consumption pathways such as denitrification and dearomatics are not fully considered, resulting in inaccurate hydrogen consumption prediction; second, the operation optimization of hydrogenation unit is separated from hydrogen network scheduling, making it impossible to achieve global optimization; third, due to the high complexity of the model, the solution time is as long as several days or even fails to converge, making it difficult to apply in engineering. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a hydrogen network optimization method and system based on a Gaussian process regression surrogate model. By introducing a high-precision surrogate model for hydrogenation reaction kinetics of different impurities, the difficulty of solving the optimization problem and the computation time are significantly reduced.

[0006] To achieve the above objectives, the present invention employs the following technical solution: A method for synchronizing and optimizing a hydrogen refueling unit and a hydrogen network based on a Gaussian process regression surrogate model includes the following steps: S1. For multiple hydrogenation units, Gaussian process regression surrogate models of reaction kinetics with different impurities are constructed respectively. The Gaussian process regression surrogate models of reaction kinetics take the operating temperature and operating pressure of the corresponding hydrogenation unit as input and the hydrogen consumption and impurity content in the product of the corresponding hydrogenation unit as output. The Gaussian process regression surrogate models of reaction kinetics with different impurities of the multiple hydrogenation units are trained using a dataset. S2, embed the Gaussian process regression surrogate model of the reaction kinetics of different impurities in the multiple hydrogenation devices after training into the hydrogen network, and establish an integrated optimization model with the goal of minimizing the total annual cost; S3, The improved differential evolution algorithm is used to solve the integrated optimization model to obtain the optimal operating temperature and pressure of each hydrogenation unit; The improved differential evolution algorithm solves the ensemble optimization model as follows: an initial population is generated through a hybrid initialization strategy; an adaptive parameter-controlled differential evolution algorithm is used to integrate multiple mutation strategies for iterative evolution; the predicted mean and uncertainty distribution of the ensemble optimization model are combined to obtain the unconfirmed subspace; within the unconfirmed subspace, a quasi-Newton method is used for directional iterative optimization to obtain the neighborhood of the local optimum; additional sampling is performed within the neighborhood of the local optimum, and the performance is predicted using a Gaussian process regression surrogate model; based on the predicted performance, elite candidate solutions are obtained; the elite candidate solutions are substituted into the Gaussian process regression surrogate model of reaction kinetics for hard constraint verification; solutions that violate the hard constraint verification are adjusted; based on the adjusted ensemble optimization model, the total annual cost is obtained; the total annual cost is compared with the actual total annual cost to determine the global optimum. S4. The global optimal solution is verified using the original mechanism model, and an optimized operation scheme is output after verification.

[0007] A further improvement of the present invention is that: Preferably, in S1, the data in the dataset is generated by a combination of uniform sampling, grid sampling, boundary focus sampling, corner sampling, and Latin hypercube sampling methods.

[0008] Preferably, the hydrogenation unit includes vacuum gas oil, coking diesel oil, straight-run diesel oil, catalytic cracking gasoline, and coal-to-oil. The impurities include sulfur, nitrogen, and aromatics.

[0009] Preferably, in S2, the kernel function of the Gaussian process regression surrogate model is a combination of the Matern kernel function and the white noise kernel function; the input of the Gaussian process regression surrogate model is the operating temperature and operating pressure, and the output is the total hydrogen consumption, sulfur content, nitrogen content and aromatic content in the product.

[0010] Preferably, in S2, the objective function of the integrated optimization model is:

[0011]

[0012] Where TAC represents the total annual cost, CNY·y -1 ; This indicates the cost of new hydrogen, CNY·y -1; Represents fuel gas revenue, CNY·y -1 ; For the electricity cost of the hydrogen refueling unit, CNY·y -1 ; For the fuel cost of the heating furnace, CNY·y -1 ; For high-pressure steam cost, CNY·y -1 ; For the value of low-pressure steam production, CNY·y -1 .

[0013] Preferably, the constraints of the integrated optimization model include reaction temperature constraints, reaction pressure constraints, product impurity content limits, and hydrogen flow rate constraints.

[0014] Preferably, in S3, during the process of identifying the subspace to be confirmed by the prediction mean and uncertainty distribution of the integrated optimization model, if 10 consecutive optimal solutions remain unchanged, the population is regenerated.

[0015] Preferably, in S3, during the directional iterative optimization process, the central difference method is used to calculate the pseudo gradient of the objective function with respect to the decision variables.

[0016] Preferably, in S4, the verification process through the original mechanism model includes cost verification, impurity content verification, hydrogen consumption composition verification, and comparative analysis.

[0017] A synchronous optimization system for a hydrogen refueling unit and a hydrogen network based on a Gaussian process regression surrogate model includes: The surrogate model module is used to construct Gaussian process regression surrogate models of reaction kinetics for different impurities for multiple hydrogenation units. The Gaussian process regression surrogate models of reaction kinetics take the operating temperature and operating pressure of the corresponding hydrogenation unit as input and the hydrogen consumption and impurity content in the product of the corresponding hydrogenation unit as output. The Gaussian process regression surrogate models of reaction kinetics for different impurities in the multiple hydrogenation units are trained using a dataset. An integrated modeling module is used to embed the Gaussian process regression surrogate model of the reaction kinetics of different impurities in the multiple hydrogenation units after training into the hydrogen network, and to establish an integrated optimization model with the goal of minimizing the total annual cost. An optimization solution module is used to solve the integrated optimization model using an improved differential evolution algorithm to obtain the optimal operating temperature and pressure for each hydrogenation unit; The improved differential evolution algorithm solves the ensemble optimization model as follows: an initial population is generated through a hybrid initialization strategy; an adaptive parameter-controlled differential evolution algorithm is used to integrate multiple mutation strategies for iterative evolution; the predicted mean and uncertainty distribution of the ensemble optimization model are combined to obtain the unconfirmed subspace; within the unconfirmed subspace, a quasi-Newton method is used for directional iterative optimization to obtain the neighborhood of the local optimum; additional sampling is performed within the neighborhood of the local optimum, and the performance is predicted using a Gaussian process regression surrogate model; based on the predicted performance, elite candidate solutions are obtained; the elite candidate solutions are substituted into the Gaussian process regression surrogate model of reaction kinetics for hard constraint verification; solutions that violate the hard constraint verification are adjusted; based on the adjusted ensemble optimization model, the total annual cost is obtained; the total annual cost is compared with the actual total annual cost to determine the global optimum. The verification output module is used to verify the global optimal solution through the original mechanism model, and output the optimized operation scheme after verification.

[0018] Compared with the prior art, the present invention has the following beneficial effects: This invention discloses a synchronous optimization method for hydrogenation units and hydrogen networks based on Gaussian process regression surrogate models. First, Gaussian process regression surrogate models are constructed for the desulfurization, denitrification, and dearomatics reaction kinetics of different hydrogenation units. These models use operating temperature and pressure as inputs to accurately predict hydrogen consumption and product impurity content for each unit, achieving a prediction accuracy R² of over 0.9999. Then, the surrogate models for each unit are embedded into a mixed integer nonlinear (MINLP) model for hydrogen network optimization, establishing an integrated optimization problem with the goal of minimizing the total annual cost (TAC). Finally, an improved differential evolution algorithm is used to solve the problem through a three-stage search strategy of "global exploration - local development - convergence verification." This invention reduces the optimization computation time from several days to approximately 90 seconds, solving the problem of difficulty in solving traditional methods due to coupled complex kinetics. While ensuring product quality, it achieves coordinated optimization of hydrogenation unit operation and hydrogen network operation, effectively reducing hydrogen consumption per unit and the total annual cost of the system, significantly improving the economic benefits and operational efficiency of the refinery's hydrogen system. The method of this invention also has the following advantages: (1) Significantly improved solution efficiency: By using a Gaussian process regression surrogate model for hydrogenation kinetics of different impurities to replace the computationally expensive original kinetic model, the solution time of complex MINLP problems is shortened from several days in traditional methods to less than 90 seconds.

[0019] (2) Comprehensive and accurate hydrogen consumption prediction: The surrogate model fully reflects the hydrogen consumption characteristics of the actual hydrogenation process and has high prediction accuracy (R²≥0.9999) by taking into account the coupled effects of various hydrogenation reaction kinetics such as molecular-level desulfurization, denitrification, and dearomatic removal.

[0020] (3) Global collaborative optimization: It realizes the synchronous collaborative optimization of the operating conditions of the hydrogenation unit and the operation of the hydrogen network of the whole plant, overcomes the limitations of step-by-step optimization, and in typical cases, the annual total cost can be reduced by about 12.8%, with significant economic benefits. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the optimized framework of the method of the present invention; Figure 2 Here is a flowchart of the differential evolution algorithm used in Example 1; Figure 3 This is a diagram illustrating a three-stage search strategy; Figure 4 To improve the accuracy of fitting the hydrogenation kinetics formula to the VGD Gaussian process regression surrogate model in Example 1; Figure 5 The accuracy plot of fitting the hydrogenation kinetics formula to the CD Gaussian process regression surrogate model in Example 1; Figure 6 The accuracy plot of fitting the hydrogenation kinetics formula to the SD Gaussian process regression surrogate model in Example 1; Figure 7 The accuracy diagram of fitting the hydrogenation kinetics formula to the K-Gaussian process regression surrogate model in Example 1; Figure 8 The accuracy diagram of fitting the hydrogenation kinetics formula to the CG Gaussian process regression surrogate model in Example 1; Figure 9 This is a diagram of the hydrogen network structure before optimization in Example 1; Figure 10 This is a diagram of the optimized hydrogen network structure from Example 1. Detailed Implementation

[0022] Hereinafter, the terms "first," "second," "third," and "fourth" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first," "second," "third," or "fourth" may explicitly or implicitly include one or more of that feature.

[0023] The synchronization method provided in this application can be applied to mobile phones, tablets, wearable devices, in-vehicle devices, augmented reality (AR) / virtual reality (VR) devices, laptops, and ultra-mobile personal computers. In this application, the specific type of terminal device is not limited to terminal devices such as mobile personal computers (UMPCs), netbooks, and personal digital assistants (PDAs).

[0024] It should be noted that the terms "first," "second," etc., used in the specification and drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] Step 1: Construction of Multi-Impurity Hydrogenation Kinetics Surrogate Models: For multiple hydrogenation units, including VGO (vacuum-fired gas oil), CD (coking diesel), SD (straight-run diesel), K (catalytic cracking gasoline), and CG (coal-to-oil), Gaussian process regression surrogate models were constructed for each unit to analyze the hydrogenation reaction kinetics of different impurities at the molecular level, focusing on desulfurization, denitrification, and aromatics removal. The Gaussian process regression surrogate model for each impurity's hydrogenation reaction kinetics uses the unit's operating temperature (T) and operating pressure (P) as input features to accurately predict the unit's total hydrogen consumption (TH), sulfur content (S_prod), nitrogen content (N_prod), and aromatics content (A_prod). Each surrogate model was trained based on data generated from a high-fidelity mechanistic model, ensuring that its prediction accuracy R² reached above 0.9999.

[0026] The coefficients (e.g., K) in the Gaussian process regression proxy model of the reaction kinetics of different impurities in the above-mentioned multiple hydrogenation units are related to temperature and pressure. The operating conditions, temperature and pressure of each hydrogenation unit are different, as are the properties of the feed (sugar content, nitrogen content and aromatic content). The coefficients of the same reaction in different units are also different. By constructing different models for different units with different operating conditions and feed properties, the accuracy of the final overall model will be higher.

[0027] The Gaussian process regression surrogate model is trained using a combination of the Matern kernel function and the white noise kernel function. The input operating temperature and pressure are standardized and preprocessed, and the output hydrogen consumption and impurity content are normalized to improve the model training efficiency and prediction stability.

[0028] Independent training datasets are used for the surrogate models of different hydrogenation devices. The training datasets are generated by uniformly sampling within the allowable operating temperature range [260℃, 400℃] and operating pressure range [2 MPa, 12 MPa] of the device through a high-fidelity mechanism model. The number of training samples for each device is no less than 1000 sets, which provides a theoretical basis and data support for the establishment of subsequent surrogate models.

[0029] Kinetic equation for the removal of molecular-level sulfides: (1) in, The content of different sulfur forms in the products of different hydrogenation units is expressed in ppm. The content of different sulfur forms in the feed of different hydrogenation units is expressed in ppm; the superscript SC indicates the sulfur form. The kinetic constants for desulfurization reactions of different sulfur types are represented; KS and KN represent the inhibition coefficients of feed sulfur content and nitrogen content on the desulfurization reaction, respectively. and The total sulfur content and total nitrogen content of the feed are expressed in ppm. This indicates the partial pressure of hydrogen in different hydrogenation units, in MPa. This represents the partial pressure dependence of different sulfur forms in different hydrogenation units; This indicates airspeed, h-1.

[0030] The kinetic equation for the hydrogenation denitrification reaction is: (2) in, This indicates the nitrogen content (ppm) of products from different hydrogenation units. This indicates the feed nitrogen content (ppm) for different hydrogenation units; Represents the reaction kinetic constants for denitrification reactions in different devices; This represents the hydrogen partial pressure dependence term for the denitrification reaction in different hydrogenation units.

[0031] Kinetic equation for aromatic hydrocarbon saturation: (3) (4) in, The conversion rate to aromatic saturation is expressed as %; and This represents the kinetic constants of the saturated forward and reverse reactions of aromatics in different hydrogenation units; The hydrogen partial pressure dependence term represents the aromatic saturation reaction in different hydrogenation units; M i This indicates the ratio of non-aromatic hydrocarbon content to aromatic hydrocarbon content in the feed. This indicates the aromatic content in the feedstock of different hydrogenation units.

[0032] Based on the above mechanistic model, Gaussian process regression surrogate models for the molecular-level desulfurization, denitrification, and dearomatics reaction kinetics were constructed for each of the five hydrogenation units: VGO, CD, SD, K, and CG. The kernel function used was a combination of the Matern kernel function and the WhiteKernel, and its expression is as follows: (5) in and This represents two distinct input vectors, each consisting of the outlet temperature and reaction pressure of the hydrogenation unit. The Euclidean distance between the input vectors; For the Kronecker delta function; For noise variance; The Matern kernel function has the following specific form: (6) In the formula, For length scale, v For smoothing parameters, For gamma function, for v The model employs a modified Bessel function of the second kind. During training, the input operating temperature and pressure are standardized, and the output hydrogen consumption and impurity content are normalized to improve training efficiency and prediction stability. Training data is generated using a structured hybrid sampling strategy: within an extended operating range (temperature ±5℃, pressure ±0.5MPa), at least 2000 samples are generated using a mixture of seven methods, including uniform sampling, grid sampling, boundary-focused sampling, corner sampling, and Latin hypercube sampling, with a focus on enhancing boundary region sampling to improve prediction accuracy at optimized boundaries. To prevent overfitting by the surrogate model, a three-dataset independent validation mechanism is used, with the difference in prediction accuracy between the training and test sets as the overfitting criterion. When the difference exceeds a threshold, automatic retraining with a simplified kernel function and enhanced regularization ensures the model maintains high accuracy while possessing good generalization ability. The prediction accuracy R² is consistently above 0.999, with R² in the boundary region not lower than 0.99.

[0033] Step 2, Integrated Optimization Model Establishment: The proxy models of each device trained in Step 1 are embedded into the hydrogen network MINLP model to establish a synchronous optimization system between the hydrogen refueling unit and the hydrogen network. This integrated model uses the minimization of TAC as the objective function, comprehensively considering hydrogen procurement costs, fuel gas recovery value, electricity costs, steam system costs, and heater fuel costs. The model also incorporates constraints, specifically including the temperature and pressure operating range of the devices, product quality indicators, hydrogen network flow balance, purity requirements, and equipment capacity limitations, forming a complex nonlinear constrained optimization problem.

[0034] The objective function is as follows: (7) (8) Where TAC represents the total annual cost, and CNY represents the total annual cost. y -1 ; This indicates the cost of new hydrogen, in CNY. y -1 ; Represents fuel gas revenue, CNY y -1 ; For the electricity cost of the hydrogen refueling unit, CNY y -1 ; For the fuel cost of the heating furnace, CNY y -1 ; For high-pressure steam cost, CNY y -1 ; For the value of low-pressure steam production, CNY y -1 .

[0035] In some embodiments of the present invention, the specific costs include the above-mentioned new hydrogen cost, fuel gas revenue, electricity cost of hydrogenation device, fuel cost of heating furnace, high-pressure steam cost and low-pressure steam output value, and the calculation formulas are as follows (4)-(14).

[0036] Specifically, the cost of new hydrogen refers to the cost of purchased hydrogen. New hydrogen is primarily used in the reaction process of the hydrogenation unit to remove impurities. The cost of new hydrogen can be calculated using the following formula: (9) in, Indicates the first i The flow rate of fresh hydrogen consumed by each hydrogenation unit is mol / s; This indicates the unit price of new hydrogen, in CNY. mol -1 .

[0037] Specifically, the revenue from fuel gas is the recovered value of hydrogen-rich fuel gas. The "low-value hydrogen" produced by the hydrogenation unit is sold as fuel. The revenue from fuel gas can be calculated using the following formula: (10) in, This indicates the price of hydrogen, in CNY. MJ -1 ; This indicates the flow rate of hydrogen source used as fuel gas, in Nm. 3 h -1 ; Indicates the enthalpy change of the reaction involving hydrogen, MJ Nm -3 ; The enthalpy change of the reaction that produces methane, MJ Nm -3 ; This indicates the purity of hydrogen from different hydrogen sources.

[0038] Specifically, the electricity cost for a hydrogenation unit depends on the specific operation of pumps and compressors in a refinery. Pumps and compressors are typically driven in two ways: by electric motors and by steam turbines, consuming electricity and high-pressure steam respectively. If the refinery uses electricity to drive the equipment, the electricity cost can be calculated using the following formula: (11) in: Indicates the first i The first hydrogenation unit j The power of each electrically driven pump, in kW; Indicates the first i The first hydrogenation unit k The power of the electric compressor, in kW; Electricity price, CNY kW -1 .

[0039] The power consumption of the pump can be calculated using the following formula: (12) in, Indicates passing through the first i The volumetric flow rate of the hydrogenation feed pump for each hydrogenation unit, in m³ / s. 3 h -1 ; and These represent the inlet and outlet pressures of the pump, respectively, in MPa; eta Indicates overall efficiency, %.

[0040] The compressor power is calculated as follows: (13) in, J represents the isobaric specific heat capacity of the compressed gas. mol -1 K -1 ; Indicates the first i The hydrogenation unit k The suction temperature of the electrically driven compressor is ℃; and They represent the first i The hydrogenation unit k The suction and discharge pressures of an electrically driven compressor, in MPa; γ represents the heat capacity ratio, Cp / Cv; Indicates the first i The hydrogenation unit k The flow rate of the compressed gas in an electrically driven compressor, in mol s -1 , Indicates the first i The hydrogenation unit k The compression efficiency of an electrically driven compressor.

[0041] Specifically, the fuel cost for the heating furnace is as follows: fuel gas is mainly used to heat the reaction feed to achieve a suitable reaction temperature. The fuel gas cost can be calculated using the following formula: (14) in, Indicates the first i The power of the heating furnace in the hydrogenation unit, in kW; This indicates the unit price of fuel gas, in CNY. MJ -1 The load on the heating furnace and the discharge temperature of the heating furnace have an approximately linear relationship, which can be expressed as: (15) in, Indicates the first i The outlet temperature of the heater in each hydrogenation unit, ℃; and To calculate the coefficients.

[0042] Specifically, the cost of high-pressure steam is: (16) in, This indicates the unit price of high-pressure steam, in CNY. t -1 ; Indicates the first i The first hydrogenation unit lThe amount of high-pressure steam consumed by each device, in tons. h -1 The calculation formula is as follows: (17) in, Indicates the first i The first hydrogenation unit l The power (kW) of the steam turbine used to drive the equipment is [value missing]. and MJ represents the enthalpy of high-pressure steam and low-pressure steam, respectively. t -1 ; Indicates the first i The first hydrogenation unit l The equipment uses the power of the steam turbine.

[0043] Specifically, the cost of low-pressure steam: (18) in, Indicates the first i The first hydrogenation unit l The amount of low-pressure steam consumed by each device, t h -1 ; This indicates the unit price of low-pressure steam, in CNY. t -1 .

[0044] The low-pressure steam flow rate is: (19) In some specific embodiments of the present invention, the constraints include reaction temperature constraints, reaction pressure constraints, product impurity content limits, and hydrogen flow rate constraints.

[0045] Specifically, the reaction temperature is constrained because after the feed oil and mixed hydrogen pass through the heater, they enter the reactor to react. The reaction effluent then exchanges heat with the feed oil to recover heat and cools the effluent to the target temperature. Therefore, the outlet temperature of the heater is the reaction temperature. This is expressed as follows: (20) in, Indicates the reaction temperature, in °C.

[0046] The model optimizes the desulfurization, denitrification, and aromatic saturation reaction depth, adjusting the reaction temperature and pressure values. However, the adjustment range should be within the allowable range of the device. The constraints on the reaction temperature are as follows: (twenty one) in, Indicates the lowest temperature at which the reaction occurs, expressed in °C. The highest temperature of the reaction is indicated in °C.

[0047] The specific pressure constraints for the reaction are as follows: There are multiple streams related to the reaction pressure, including feed oil, fresh hydrogen, and recycled hydrogen. Therefore, the minimum pressure of these three streams determines the overall reaction pressure. The feed oil, after being pressurized by the hydrogenation feed pump and heat-exchanged in the heat exchanger, mixes with the mixed hydrogen and then heat-exchanges with the reaction effluent before entering the heater and finally the reactor. This model, based on the pressure adjustment range of each hydrogenation unit, does not consider the compressor and only restricts the upper and lower limits of the pressure. (twenty two) in, The pressure of the reaction is expressed in MPa. The minimum pressure required for the reaction is expressed in MPa. This indicates the maximum pressure of the reaction, expressed in MPa.

[0048] Product impurity content limits: The refinery has set maximum impurity content limits for the products of each hydrogenation unit in accordance with environmental regulations, and these limits must be met. (twenty three) (twenty four) (25) In addition, the refinery also has corresponding regulations on the sulfur content of the feed for each hydrotreating unit: (26) in, This indicates the maximum sulfur impurity content of the product, expressed in ppm. This indicates the maximum nitrogen impurity content of the product, expressed in ppm. This indicates the content of aromatic impurities in the product, in %. This indicates the highest content of aromatic impurities in the product, in % %. This indicates the maximum content of sulfur impurities in the feed, expressed in ppm.

[0049] Hydrogen flow constraints require a balance in the total input flow of the hydrogen refueling unit, with each hydrogen source input device... i The total flow rate equals the device i Hydrogen consumption: (27) in, This represents the total hydrogen flow rate output from the hydrogen source, in Nm³. 3 h -1 ; This represents the total hydrogen flow rate into the hydrogen trap, in Nm³. 3 h-1 .

[0050] The maximum total output flow rate of the hydrogen source shall not exceed the total flow rate allocated to each hydrogen refueling unit. (28) in, This represents the maximum total hydrogen flow rate output from the hydrogen source, in Nm³. 3 h -1 .

[0051] The lower limit of hydrogen flow rate for the hydrogenation unit: the total hydrogen flow rate input to the hydrogenation unit shall not be less than the required hydrogen flow rate of the unit. (29) in, This indicates the minimum total hydrogen flow rate input to the hydrogen trap, in Nm³. 3 h -1 .

[0052] Step 3, Solving with the improved differential evolution algorithm: The improved differential evolution algorithm is used to solve the MINLP problem established above. By optimizing the operating temperature and pressure of each hydrogenation unit, as well as the flow matching between the hydrogen source and the hydrogen trap, the optimal combination of operating parameters and the best matching relationship that satisfy all process constraints are obtained.

[0053] The differential evolution algorithm is as follows: Figure 2 and Figure 3 As shown, it includes the following three stages: Step 301: Global exploration and potential subspace identification based on the GPR surrogate model (Gaussian Process Regression Surrogate Model).

[0054] Leveraging the continuously differentiable global response surface and uncertainty quantification capabilities provided by the GPR surrogate model, this algorithm rapidly predicts the performance and confidence intervals of arbitrary parameter combinations with minimal computational cost. The algorithm parameters are set as follows: initial population size 32, maximum iterations 140; a hybrid initialization strategy is used to generate the initial population, where 50% of individuals are generated through Latin hypercube sampling to ensure uniform spatial coverage, 25% are generated near parameter boundaries to enhance boundary exploration, and 25% are generated in the central region of the parameter space (30%-70% range) to focus on high-quality regions. The coefficient of variation F and crossover probability CR are adaptively adjusted with the iteration process: in the early evolutionary stage (first 40% of iterations), F is set to 0.8-1.2 to enhance global exploration, and in the later stage (last 40% of iterations), F decreases to 0.4-0.8 to strengthen local development; the baseline value of the crossover probability CR is 0.9, which is gradually increased to 1.1 in the later stages to accelerate convergence. The algorithm integrates three mutation strategies, randomly selected according to probability: rand / 1 (30%), best / 1 (30%), and current-to-best / 1 (40%). The algorithm also incorporates a stagnation detection mechanism. When it detects evolutionary stagnation (the optimal solution remains unchanged), it introduces diversity by regenerating the population to avoid premature convergence. By comprehensively analyzing the population convergence trajectory and the predicted mean and uncertainty distribution of the GPR surrogate models for each hydrogenation unit, one or more connected regions with significantly lower predicted total annual costs are identified, defined as the subspace to be confirmed. The term "significantly lower" is dynamically determined based on the fitness ranking of individuals in the population (top 10%–20%). During stagnation detection, the algorithm retains the top 20% of the best individuals, who may belong to different connected regions in the parameter space. Based on this, a local search is performed on the neighborhood centered on the current optimal solution with a radius of 2% of the parameter range as the primary subspace to be confirmed, for further development. When multiple connected regions are identified, the region with the lowest fitness is prioritized for further development, with other regions serving as alternatives. Throughout this process, when the algorithm detects evolutionary stagnation, it maintains diversity by retaining the top 20% of the best individuals and regenerating the remaining individuals, thus achieving a balance between global search and local development.

[0055] Step 302, Targeted development and elite solution selection based on pseudo-gradient To improve the accuracy of the local surrogate model by increasing the sampling density within the identified subspace to be confirmed, the specific process is as follows: Utilizing the differentiability of the GPR model, a quasi-Newton method is used to perform directional iterative optimization within the subspace to be confirmed. During this directional iterative optimization process, the central difference method is employed to calculate the pseudo-gradient of the objective function with respect to the decision variables, providing a precise descent direction for the gradient descent-based local optimizer. The decision variables are the operating temperature T of each hydrogenation unit. out With operating pressure P r , forming a vector The pseudo gradient is calculated as follows: for the ...j Each decision variable, in the current solution x Positive and negative perturbations were applied at the locations respectively. in =10 -3 , u j and l j (where the upper and lower bounds of the variable are respectively), to obtain the perturbation point. x + and x - The fitness value is evaluated through the objective function. f ( x + )and f ( x - If the pseudo gradient is approximated as follows: (30) This pseudo-gradient indicates the local descent direction of the objective function at the current point. The local optimizer uses gradient descent, and its update rule is as follows: (31) Multiplied by Scaling the step size to match the range of variables, the learning rate =0.01. Perform 3 gradient updates, quickly converge to a local optimum, and then uniformly supplement 15 candidate solutions within a 2% radius of the optimal neighborhood (i.e., the newly generated points satisfy...). Based on the GPR prediction performance, a set of elite candidate solutions is selected for subsequent hard constraint verification and confirmation of the global optimal solution.

[0056] Step 303: Hard constraint verification and global optimal solution confirmation based on the mechanism model Elite candidate solutions are substituted into the dynamic model for hard constraint verification. For solutions that are close to or slightly violate the constraints, the corresponding elite candidate solutions are fine-tuned using an external penalty function method. For excessive aromatic hydrocarbon, sulfur, and nitrogen content, adjustments are made by decreasing the temperature or increasing the pressure, respectively, with a maximum of 15 adjustments to ensure that they fall entirely within the feasible region. After all feasible solutions are verified by the mechanistic model, the global optimal solution is determined by comparing it with the actual total annual cost.

[0057] By employing the three-stage search strategy described above, the algorithm effectively handles constraints and performs fine-grained optimization while ensuring global search capabilities, significantly improving optimization efficiency and result quality.

[0058] Step 4, Comprehensive Validation and Output of Optimization Results: Substitute the globally optimal operating parameters obtained from the surrogate model in Step 3 into the original mechanism model for complete posterior validation and comparative analysis. This includes: (1) Cost verification: Calculate the actual total annual cost under the operating conditions and compare it with the cost predicted by the proxy model in step 3. If the relative error is less than 0.5%, it is considered reliable. (2) Impurity content verification: Verify whether the sulfur, nitrogen and aromatic hydrocarbon content of the product strictly meets the process constraints, and whether the deviation from the predicted values ​​of the proxy model does not exceed 5 ppm (sulfur), 0.01 ppm (nitrogen) and 0.005 (aromatic hydrocarbon), respectively. (3) Hydrogen consumption composition verification: Desulfurization, denitrification, dearomatics and aromatic saturated hydrogen consumption under computer model, verifying the consistency with the component hydrogen consumption predicted by the proxy model; (4) Comparative analysis: The results are systematically compared with the original operating conditions in terms of operating parameters, hydrogen consumption, product quality, total annual cost and cost composition, to quantify the optimization effect.

[0059] If any of the above errors exceeds a preset threshold, the surrogate model is determined to have a significant deviation in the current optimal region, triggering the surrogate model retraining mechanism. The specific process is as follows: Supplementary training data is generated through intensive sampling within the optimal neighborhood; the surrogate model for the corresponding device is incrementally updated or completely retrained; and the process returns to step 3 for re-optimization. If all verification indicators meet the requirements, a complete optimization result report is output, including the optimized operating temperature and pressure of each device, hydrogen consumption composition and proportion, product sulfur / nitrogen / aromatic content, total annual cost composition, hydrogen source-hydrogen trap flow allocation scheme, and comprehensive comparative analysis charts with the original operating conditions, providing a complete decision-making basis for the refined operation of the refinery's hydrogen system.

[0060] A second aspect of the present invention discloses a system comprising: The proxy model module is used to store, load, and call Gaussian process regression proxy models for hydrogenation reaction kinetics of different impurities in various hydrogenation units. Specifically, for multiple hydrogenation units, Gaussian process regression proxy models for reaction kinetics of different impurities are constructed respectively. The Gaussian process regression proxy models for reaction kinetics of the corresponding hydrogenation unit are taken as inputs and the hydrogen consumption and impurity content in the products of the corresponding hydrogenation unit are taken as outputs. The Gaussian process regression proxy models for reaction kinetics of different impurities in the multiple hydrogenation units are trained using a dataset. ② The integrated modeling module is used to embed the Gaussian process regression surrogate model of the reaction kinetics of different impurities in the multiple hydrogenation units after training into the hydrogen network, and to establish an integrated optimization model with the goal of minimizing the total annual cost; ③ The optimization solution module, used by the improved differential evolution algorithm to solve the ensemble optimization model, is as follows: An initial population is generated through a hybrid initialization strategy; an adaptive parameter-controlled differential evolution algorithm is used to integrate multiple mutation strategies for iterative evolution; the predicted mean and uncertainty distribution of the ensemble optimization model are combined to obtain the subspace to be confirmed; within the subspace to be confirmed, a quasi-Newton method is used for directional iterative optimization to obtain the neighborhood of the local optimum; additional sampling is performed within the neighborhood of the local optimum, and performance is predicted using a Gaussian process regression surrogate model; based on the predicted performance, elite candidate solutions are obtained; the elite candidate solutions are substituted into the Gaussian process regression surrogate model of reaction kinetics for hard constraint verification; solutions that violate the hard constraint verification are adjusted; based on the adjusted ensemble optimization model, the total annual cost is obtained; the total annual cost is compared with the actual total annual cost to determine the global optimal solution. The verification output module is used to verify the global optimal solution through the original mechanism model, and output the optimized operation scheme after verification.

[0061] The following description, in conjunction with specific embodiments, provides further details.

[0062] Example 1 The hydrogen network of the hydrotreating units in a fuel oil refinery with a processing capacity of 8 million tons / year is optimized. The basic process of the refinery is as follows: after crude oil undergoes atmospheric and vacuum distillation, vacuum wax oil is refined in a 2.6 million tons / year wax oil hydrotreating unit (VGO) and then enters the catalytic cracking unit to produce catalytic gasoline and catalytic diesel. The latter two are then fed into a 1.4 million tons / year catalytic gasoline hydrotreating unit (CG) and an 800,000 tons / year catalytic diesel hydrotreating unit (CD), respectively. Straight-run jet fuel and diesel are refined in an 800,000 tons / year kerosene hydrotreating unit (K) and a 2.6 million tons / year diesel hydrotreating unit (SD), respectively, to finally obtain qualified products.

[0063] (1) Based on the kinetic formulas of hydrodesulfurization, denitrification, and dearomatics reactions, and combined with the specific feed sulfur content, nitrogen content, aromatic content, and feed flow rate of each unit, as shown in Table 1 below, and within the allowable operating temperature and pressure ranges of the units, as shown in Table 2 below, a training dataset was generated through a high-fidelity mechanism model. This dataset was used to train a Gaussian process regression surrogate model. Verification showed that the prediction accuracy R² of each surrogate model for hydrogen consumption and product impurity content all reached above 0.9999. Figures 4-8 As shown, it meets the requirements for engineering applications.

[0064] Table 1. Feed properties and operating conditions of each hydrotreating unit in the refinery in Example 1

[0065] Table 2. Refinery operation adjustment range and product quality requirements for Example 1 (2) The trained surrogate model is embedded into the hydrogen network optimization framework to establish a synchronous optimization model between the hydrogen refueling unit and the hydrogen network. This model aims to minimize the total annual cost, comprehensively considering multiple costs such as hydrogen procurement, compressor energy consumption, and steam consumption. Constraints cover multiple factors including the operating parameter ranges of each unit, product quality indicators, and the flow balance and purity requirements of the hydrogen network. Figure 1 As shown, this integrated optimization framework enables deep collaboration between hydrogen refueling unit operation optimization and hydrogen network operation scheduling.

[0066] (3) The improved differential evolution algorithm is used to solve the constructed optimization model. The algorithm flow is as follows: Figure 2 As shown in Table 3, the optimization process employs a three-stage search strategy, significantly improving convergence speed while ensuring global optimization capability. The changes in operating parameters of each hydrogenation unit before and after optimization are shown in Table 3, all of which have been adjusted compared to the original operating conditions. The corresponding hydrogen consumption comparison is shown in Table 4. Calculation results show that, under the premise of ensuring strict compliance with product quality standards for all units, the total hydrogen consumption of the system significantly decreased from 599.6 mol / s to 518.7 mol / s, a reduction of 13.5%. Through the coordinated optimization of hydrogenation unit operating parameters and the synchronous adjustment of the hydrogen network scheduling strategy, a rational redistribution of hydrogen resources was achieved. The hydrogen network structure before and after optimization is shown in Table 4. Figure 9 , Figure 10 As shown, the system transitioned from complete reliance on purchased hydrogen to an optimized mode that fully utilizes a mixture of internal hydrogen sources and purchased hydrogen. The purchased hydrogen flow rate decreased significantly from 599.6 mol / s to 423.0 mol / s, a reduction of 29.5%. Table 5 compares the total annual costs before and after optimization, showing a 24.2% reduction in annual steam costs and a significant reduction of 4.08 × 10⁻⁶ in the total annual system cost. 7 The total price was RMB, with an overall decrease of 12.8%.

[0067] Table 3. Comparison of operating parameters of each hydrogenation unit before and after optimization in Example 1

[0068] Table 4. Comparison of hydrogen consumption flow rates of each hydrogenation unit before and after optimization in Example 1.

[0069] Table 5. Comparison of annual cost composition before and after optimization in Example 1

[0070] Compared with traditional optimization methods, this invention achieves a breakthrough in computational efficiency. Traditional mixed-integer nonlinear programming methods typically require several days to solve such strongly nonlinear problems and are difficult to guarantee convergence. However, this invention, through the organic combination of a surrogate model and an improved optimization algorithm, reduces the optimization computation time from several days to 88 seconds, improving efficiency by more than three orders of magnitude. This provides a reliable technical means for achieving near real-time optimization of refinery hydrogen systems.

[0071] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0072] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0073] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

[0075] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for synchronous optimization of a hydrogen refueling unit and a hydrogen network based on a Gaussian process regression surrogate model, characterized in that, Includes the following steps: S1. For multiple hydrogenation units, Gaussian process regression proxy models of reaction kinetics with different impurities are constructed respectively. The Gaussian process regression proxy models of reaction kinetics take the operating temperature and operating pressure of the corresponding hydrogenation unit as inputs and the hydrogen consumption and impurity content in the product of the corresponding hydrogenation unit as outputs. The Gaussian process regression surrogate model for the reaction kinetics of different impurities in the multiple hydrogenation units was trained using the dataset. S2, embed the Gaussian process regression surrogate model of the reaction kinetics of different impurities in the multiple hydrogenation devices after training into the hydrogen network, and establish an integrated optimization model with the goal of minimizing the total annual cost; S3, The improved differential evolution algorithm is used to solve the integrated optimization model to obtain the optimal operating temperature and pressure of each hydrogenation unit; The improved differential evolution algorithm solves the ensemble optimization model as follows: an initial population is generated through a hybrid initialization strategy; an adaptive parameter-controlled differential evolution algorithm is used to integrate multiple mutation strategies for iterative evolution; the predicted mean and uncertainty distribution of the ensemble optimization model are combined to obtain the unconfirmed subspace; within the unconfirmed subspace, a quasi-Newton method is used for directional iterative optimization to obtain the neighborhood of the local optimum; additional sampling is performed within the neighborhood of the local optimum, and the performance is predicted using a Gaussian process regression surrogate model; based on the predicted performance, elite candidate solutions are obtained; the elite candidate solutions are substituted into the Gaussian process regression surrogate model of reaction kinetics for hard constraint verification; solutions that violate the hard constraint verification are adjusted; based on the adjusted ensemble optimization model, the total annual cost is obtained; the total annual cost is compared with the actual total annual cost to determine the global optimum. S4. The global optimal solution is verified using the original mechanism model, and an optimized operation scheme is output after verification.

2. The method for synchronous optimization of hydrogenation device and hydrogen network based on Gaussian process regression surrogate model according to claim 1, characterized in that, In S1, the data in the dataset is generated by a combination of uniform sampling, grid sampling, boundary focus sampling, corner sampling, and Latin hypercube sampling methods.

3. The method for synchronous optimization of hydrogenation device and hydrogen network based on Gaussian process regression surrogate model according to claim 1, characterized in that, The hydrogenation unit includes vacuum gas oil, coking diesel, straight-run diesel, catalytic cracking gasoline, and coal-to-oil. The impurities include sulfur, nitrogen, and aromatics.

4. The method for synchronous optimization of hydrogenation device and hydrogen network based on Gaussian process regression surrogate model according to claim 1, characterized in that, In S2, the kernel function of the Gaussian process regression surrogate model is a combination of the Matern kernel function and the white noise kernel function; the input of the Gaussian process regression surrogate model is the operating temperature and operating pressure, and the output is the total hydrogen consumption, sulfur content, nitrogen content and aromatic content in the products.

5. The method for synchronous optimization of hydrogenation device and hydrogen network based on Gaussian process regression surrogate model according to claim 1, characterized in that, In S2, the objective function of the integrated optimization model is: Where TAC represents the total annual cost, CNY·y -1 ; This indicates the cost of new hydrogen, CNY·y -1 ; Represents fuel gas revenue, CNY·y -1 ; For the electricity cost of the hydrogen refueling unit, CNY·y -1 ; For the fuel cost of the heating furnace, CNY·y -1 ; For high-pressure steam cost, CNY·y -1 ; For the value of low-pressure steam production, CNY·y -1 .

6. The method for synchronous optimization of hydrogenation device and hydrogen network based on Gaussian process regression surrogate model according to claim 5, characterized in that, The constraints of the integrated optimization model include reaction temperature constraints, reaction pressure constraints, product impurity content limits, and hydrogen flow rate constraints.

7. The method for synchronous optimization of hydrogenation device and hydrogen network based on Gaussian process regression surrogate model according to claim 1, characterized in that, In S3, during the process of identifying the subspace to be confirmed by the prediction mean and uncertainty distribution of the integrated optimization model, if 10 consecutive optimal solutions remain unchanged, the population is regenerated.

8. The method for synchronous optimization of hydrogenation device and hydrogen network based on Gaussian process regression surrogate model according to claim 1, characterized in that, In S3, during the directional iterative optimization process, the central difference method is used to calculate the pseudo gradient of the objective function with respect to the decision variables.

9. The method for synchronous optimization of hydrogenation device and hydrogen network based on Gaussian process regression surrogate model according to claim 1, characterized in that, In S4, the verification process using the original mechanism model includes cost verification, impurity content verification, hydrogen consumption composition verification, and comparative analysis.

10. A synchronous optimization system for a hydrogen refueling device and a hydrogen network based on a Gaussian process regression surrogate model, characterized in that, include: The proxy model module is used to construct Gaussian process regression proxy models of reaction kinetics for multiple hydrogenation units with different impurities. The Gaussian process regression proxy models of reaction kinetics take the operating temperature and operating pressure of the corresponding hydrogenation unit as inputs and the hydrogen consumption and impurity content in the products of the corresponding hydrogenation unit as outputs. The Gaussian process regression surrogate model for the reaction kinetics of different impurities in the multiple hydrogenation units was trained using the dataset. An integrated modeling module is used to embed the Gaussian process regression surrogate model of the reaction kinetics of different impurities in the multiple hydrogenation units after training into the hydrogen network, and to establish an integrated optimization model with the goal of minimizing the total annual cost. An optimization solution module is used to solve the integrated optimization model using an improved differential evolution algorithm to obtain the optimal operating temperature and pressure for each hydrogenation unit; The improved differential evolution algorithm solves the ensemble optimization model as follows: an initial population is generated through a hybrid initialization strategy; an adaptive parameter-controlled differential evolution algorithm is used to integrate multiple mutation strategies for iterative evolution; the predicted mean and uncertainty distribution of the ensemble optimization model are combined to obtain the unconfirmed subspace; within the unconfirmed subspace, a quasi-Newton method is used for directional iterative optimization to obtain the neighborhood of the local optimum; additional sampling is performed within the neighborhood of the local optimum, and the performance is predicted using a Gaussian process regression surrogate model; based on the predicted performance, elite candidate solutions are obtained; the elite candidate solutions are substituted into the Gaussian process regression surrogate model of reaction kinetics for hard constraint verification; solutions that violate the hard constraint verification are adjusted; based on the adjusted ensemble optimization model, the total annual cost is obtained; the total annual cost is compared with the actual total annual cost to determine the global optimum. The verification output module is used to verify the global optimal solution through the original mechanism model, and output the optimized operation scheme after verification.

Images