A multi-objective optimization method assisted by an online proxy model for green hydrogen coupled hydroprocessing process

Abstract

The application provides a kind of green hydrogen coupling hydrogenation process online agent model auxiliary multi-objective optimization method, comprising: for the green hydrogen-gray hydrogen mixed fluctuation of green hydrogen coupling hydrogenation process, the multi-objective operation parameter optimization problem of economy and carbon emission reduction is constructed, the sensitivity analysis of operation parameter is carried out, and the operation parameter to be optimized and its value range are obtained;First principle model is established for hydrogenation process and Latin hypercube sampling is carried out, and offline agent model is established by using Gaussian process regression;Under the condition of mixed hydrogen fluctuation, the initial online agent model is generated based on Gaussian process, the multi-objective optimization problem of economy and carbon emission reduction is solved by using fast non-dominated sorting optimization method, and the online agent model is periodically updated.The application considers investment cost and carbon emission reduction benefit, and realizes the balanced optimization of economic benefit and carbon emission reduction multi-objective under the condition of green hydrogen-gray hydrogen mixed frequent fluctuation.

Description

Technical Field

[0001] This invention relates to the field of chemical hydrogenation technology, specifically but not limited to an online proxy model-assisted multi-objective optimization method for green hydrogen coupled hydrogenation processes. Background Technology

[0002] With the global energy structure shifting towards low-carbon transformation, renewable energy-based hydrogen production technology is rapidly reshaping the hydrogen source structure of the chemical industry. Hydrogen is a key raw material for the synthesis of chemical products such as ammonia, ethylene glycol, and methanol. Coupled renewable energy power generation with water electrolysis hydrogen production technology, using green hydrogen to replace traditional gray hydrogen in chemical hydrogenation processes, can simultaneously solve the dual challenges of renewable energy consumption and decarbonization of chemical production.

[0003] However, the wind and solar energy, the primary sources of green hydrogen, exhibit significant volatility, leading to a marked spatiotemporal imbalance in green hydrogen supply. The stringent requirements for hydrogen purity, pressure, and flow rate in the hydrogenation process significantly conflict with the volatile nature of green hydrogen. Traditional solutions rely on large-scale hydrogen storage facilities for peak shaving, which significantly increases system costs. For example, the production of grey hydrogen in traditional chemical hydrogenation processes generates substantial amounts of CO2, and the high energy consumption of the distillation process also increases indirect carbon emissions. Therefore, some studies have utilized novel processes combining hydrogen generated from wind / solar power generation with subsequent chemical production. However, maintaining hydrogen volatility below 10% requires the construction of extensive energy and hydrogen storage facilities to suppress green hydrogen fluctuations, increasing green hydrogen production costs by more than 35%. Consequently, some researchers have attempted to construct hybrid hydrogen source systems that can effectively reduce the volatility of hydrogen used in chemical production without significantly increasing investment costs. However, existing research generally does not delve into the impact of green hydrogen-grey hydrogen ratio fluctuations on the optimization of chemical operating parameters, making it difficult for systems to simultaneously achieve optimal economic benefits and carbon reduction targets when dealing with frequent green hydrogen-grey hydrogen ratio fluctuations.

[0004] In view of this, a new method is needed to solve at least some of the above problems. Summary of the Invention

[0005] To address one or more problems in the existing technology, this invention proposes an online proxy model-assisted multi-objective optimization method for the green hydrogen coupled hydrogenation process, which achieves a balanced optimization of economic benefits and carbon emission reduction under the fluctuation of green hydrogen-gray hydrogen mixture.

[0006] The technical solution to achieve the purpose of this invention is as follows:

[0007] A multi-objective optimization method assisted by an online surrogate model for a green-hydrogen coupled hydrogenation process includes: improving the hydrogenation process to transform fluctuations in green hydrogen flow rate into fluctuations in the ratio of green hydrogen to gray hydrogen flow rate; constructing a multi-objective operational parameter optimization problem for economics and carbon reduction; establishing an offline surrogate model using Gaussian process regression; generating an initial online surrogate model based on a Gaussian process under fluctuating mixed hydrogen conditions; solving the economic-carbon reduction multi-objective optimization problem using a fast non-dominated sorting optimization method; and periodically updating the online surrogate model to improve solution efficiency, thereby achieving a balanced optimization of economic benefits and carbon reduction under frequent fluctuations in green-hydrogen-gray hydrogen mixing. Specifically, the invention includes the following steps:

[0008] (1) The green hydrogen-gray hydrogen mixed hydrogenation process uses green hydrogen as the main hydrogen source. When green hydrogen is insufficient, gray hydrogen is used for dynamic compensation. When there is an excess, it is supplied to the chemical process through electro-thermal conversion, thereby transforming the fluctuation of green hydrogen flow rate into a dynamic change in the green hydrogen-gray hydrogen ratio.

[0009] (2) For the hydrogenation process under the fluctuation of green hydrogen-gray hydrogen mixture, a multi-objective operation parameter optimization problem of economics and carbon emission reduction was constructed. Sensitivity analysis was performed on the operation parameters, and the optimization decision variables and their value ranges were obtained.

[0010] The objective function for maximizing economic benefits and minimizing carbon emissions is established as follows:

[0011]

[0012] Where x represents the set of operands, i.e., x = [x1, x2, ..., xn]. n Economic benefits ∑E i Profit per unit time is calculated as the sum of the value of the main product and the value of the by-product produced per hour, minus the total cost. The total cost specifically includes the cost of various raw materials ∑C. j The cost of electricity consumed during the process, C pow And the total thermal energy cost C consumed in the process. heat Carbon emission optimization targets include direct carbon emissions (CE) from organic matter that does not enter the product or is not recycled. d Indirect carbon emissions CE i And carbon emissions from H2 feedstock

[0013] The purity of the produced product and the upper and lower limits of various operating parameters are used as constraints:

[0014]

[0015] Where ω i To ensure the purity of the produced product, ω i,limited Minimum purity limit for product i; xm,lowerbound and x m,upperbound These represent the lower and upper limits of the m-th operating parameter, respectively, and their values ​​need to be determined according to the specifications in the process manual and sensitivity analysis.

[0016] (3) Latin hypercube sampling was performed on the first principle model (FPM), and Gaussian process regression was used to establish an offline proxy model.

[0017] Dynamic Modeling (FPM) is a theoretical model built upon the fundamental laws of physics, chemistry, and mathematics. Starting from the most basic physical and chemical laws, it describes the properties and behavior of a system through derived formulas, without relying on empirical data or experimental fitting. FPM has significant advantages, including strong universality, high predictive power, and clear mechanisms. It can be applied to a wide range of conditions and scenarios and can reveal the internal working mechanisms and essential laws of a system. However, FPM also has some drawbacks, such as high computational complexity, especially in multiphase or multi-component systems requiring substantial computational resources. In the chemical engineering field, FPM typically involves fundamental equations of thermodynamics, kinetics, and transport phenomena, such as the first law of thermodynamics, the mass conservation equation, and kinetic equations, to describe processes such as fluid flow, heat and mass transfer, and chemical reactions. Commercial simulation software such as Aspen Plus is commonly used to construct FPMs to solve complex chemical process problems.

[0018] N sets of data containing the operating parameter x and the perturbation variable d are generated using Latin hypercube sampling. These data are then input into the FPM (Optical Process Module), and the optimization target value output by the FPM is used to train an offline surrogate model. The Gaussian process surrogate model is defined as follows:

[0019] y(x)=μ+∈(x),(x)~N(0,σ 2 )

[0020] Where μ represents the mean of the GP estimate of y(x); ∈(x) is the error term, which follows a mean of 0 and a variance of σ. 2 The error follows a normal distribution. The correlation between the errors of two randomly selected input variables x1 and x2 is calculated using a Gaussian exponential function, as shown in the following equation:

[0021]

[0022] Where R(∈(x1),∈(x2)) represents the correlation between the two error terms, and its value ranges from [0,1]. The larger the value, the stronger the correlation; x d θ represents the dimension of the input variable x. k Let l represent the scaling parameter used to control the sensitivity of the GP to the input in the k-th dimension. kThis represents the length scale parameter used to control the smoothness of the GP over data points in the k-th dimension. To optimize the GP surrogate model, the hyperparameter θ needs to be determined. k and l k Therefore, the method of maximizing the logarithmic marginal likelihood function is adopted, as shown in the following equation:

[0023]

[0024] Where, X = {x 1 ,x 2 ,…,x p} represents the input of all training data; f = {y(x 1 ),y(x 2 ),…,y(x p )} represents the output of all training data; p represents the number of training data; CM represents the P×P correlation matrix.

[0025] (4) Under mixed hydrogen fluctuation conditions, an initial online surrogate model is generated based on a Gaussian process. The fast non-dominated sorting optimization method is used to solve the multi-objective optimization problem of economic and carbon emission reduction. The solution efficiency is improved by periodically updating the online surrogate model.

[0026] When significant changes are detected in perturbation variables such as the green hydrogen to gray hydrogen ratio, feed flow rate, or product price, an initial population is constructed using the current perturbation variables and Gaussian distribution operating parameters. The NSGA-II algorithm is then used in the offline surrogate model to optimize the population to the Nth generation through crossover, mutation, and selection. off generation;

[0027] The NSGA-II algorithm is an improvement on the first-generation non-dominated sorting algorithm NSGA, which is based on non-dominated sorting and sharing. Its advantages lie in its fast non-dominated sorting method, the concept of crowding, and its elite preservation strategy. It exhibits good accuracy and speed in solving multi-objective problems and is less prone to getting trapped in local optimization. It uses simulated binary crossover and polynomial mutation to generate offspring, then employs fast non-dominated sorting to select individuals, preserving elite solutions in the process. In fast non-dominated sorting, individuals have two key parameters: the non-dominated level F and the crowding degree i. d When comparing different individuals, the non-dominated rank is first determined, with higher non-dominated ranks being better. When non-dominated ranks are the same, individuals with smaller crowding operators are better. If an individual z1 is superior to another individual z2 on all optimization objectives, then z1 dominates z2. All individuals not dominated by any other individual are assigned a non-dominated level F1. Then, individuals already assigned a non-dominated level are removed, and similarly, non-dominated levels F2, F3, ... are assigned. Finally, the crowding comparison operator for individuals is calculated, as shown in the following formula:

[0028]

[0029] Therefore, all individuals can be sorted according to their non-dominance level and crowding level.

[0030] The Nth off The operating parameters and the current perturbation variable are used as inputs to perform a real-world evaluation in FPM, obtaining target values ​​f1 and f2. The evaluated individuals are then stored in dataset S. If the maximum running time t is not reached... max The system enters an online optimization loop. When the number of individuals in dataset S exceeds the capacity of the online model, it removes the individual with the highest non-dominant level, trains the online agent model using the updated dataset, and regenerates the initial population for N. on The NSGA-II optimization method uses convergence (CI) and diversity (DI) metrics to screen undominated high-quality solutions for realistic FPM evaluation, continuously enriching the dataset S. An online surrogate model assists in the multi-objective optimization problem, as shown in the following equation:

[0031]

[0032] subject to X * ∈Trial Solutions

[0033] Among them, Z * F(X) is the ideal point formed by the optimal values ​​of all objectives in the current population (e.g., the minimum value of all objectives in a minimization problem). * X is the surrogate model's prediction of the target value; p It refers to an individual in the parent population, where ||·| represents the Euclidean distance; CI(X) * The predicted target value F(X) is used to measure the solution. * ) and ideal point Z * The closer the solution is to the ideal point, the smaller the CI value, indicating better convergence; DI(X) * ) is used to measure the new solution X * The degree of difference between the new solution and the most recent individual in the parent population is considered; a larger DI value indicates a greater difference between the new solution and the parent population, and better diversity. After reaching a time threshold, the system extracts the non-dominated optimal solution from the dataset S and outputs the actually usable optimization parameters through inverse normalization.

[0034] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0035] 1. The present invention provides an online proxy model-assisted multi-objective optimization method for the green hydrogen coupled hydrogenation process. In this method, green hydrogen is used as the primary hydrogen source. When green hydrogen is insufficient, gray hydrogen is used for dynamic compensation. When green hydrogen is in excess, it is converted into thermal energy through electro-thermal conversion to power the hydrogenation process. Traditional methods either use gray hydrogen, which emits significant carbon, or employ a green hydrogen + hydrogen storage scheme, which has excessively high investment costs. The present invention balances investment costs with carbon reduction benefits and establishes a first-principles model for operational optimization, achieving an online balance between economic efficiency and carbon emissions.

[0036] 2. The online proxy model-assisted multi-objective optimization method for the green hydrogen coupled hydrogenation process of the present invention constructs a multi-objective operational parameter optimization problem of economy and carbon emission reduction based on the dynamic change of the green hydrogen to gray hydrogen flow ratio. However, the existing technology is mainly for operation optimization under fixed operating conditions. After large changes in operating conditions, it is usually necessary to re-optimize at a high time cost, which is difficult to adapt to the dynamic changes in the green hydrogen to gray hydrogen ratio caused by the fluctuation of renewable energy.

[0037] 3. The online proxy model-assisted multi-objective optimization method for the green hydrogen coupling hydrogenation process of the present invention constructs a proxy-assisted multi-objective optimization algorithm APB-NSGAII that integrates adaptive parameters, Pareto dual index sampling and NSGA-II. It effectively utilizes the previously established offline proxy model and updates the online proxy model for optimization, reduces the number of time-consuming FPM evaluations, and achieves the re-obtaining of the optimal solution with less optimization algorithm running time after the hydrogen ratio changes. Attached Figure Description

[0038] The accompanying drawings are provided to further illustrate the invention and, together with the description, serve to explain embodiments of the invention, but do not constitute a limitation thereof. In the drawings:

[0039] Figure 1 A flowchart illustrating the process of hydrogenating dimethyl oxalate to ethylene glycol using a green-gray hydrogen mixed hydrogen source according to an embodiment of the present invention is shown.

[0040] Figure 2 A flowchart of the online proxy model-assisted multi-objective optimization method for the green hydrogen coupling hydrogenation process of the present invention is shown.

[0041] Figure 3 A comparison chart of optimization results of different algorithms is shown for an embodiment of the present invention when green hydrogen is sufficient.

[0042] Figure 4 A comparison chart of optimization results of different algorithms is shown for a green hydrogen and gray hydrogen mixture according to an embodiment of the present invention.

[0043] Figure 5 A comparison chart of optimization results for a 3-month testing period according to an embodiment of the present invention is shown. Detailed Implementation

[0044] To further understand the present invention, preferred embodiments of the present invention are described below in conjunction with examples. However, it should be understood that these descriptions are only for further illustrating the features and advantages of the present invention, and not for limiting the scope of the claims of the present invention.

[0045] The description in this section pertains only to typical embodiments, and the present invention is not limited to the scope of the embodiments described. Combinations of different embodiments, substitution of some technical features in different embodiments, and substitution of similar or identical prior art with some technical features in the embodiments are also within the scope of the description and protection of the present invention.

[0046] This paper takes the hydrogenation of dimethyl oxalate (DMO) to ethylene glycol (EG) driven by mixed green hydrogen as an example to introduce an online surrogate model-assisted multi-objective optimization method for green hydrogen coupled hydrogenation process.

[0047] First, a process for the hydrogenation of dimethyl oxalate (DMO) to ethylene glycol using a mixed hydrogen source of green and gray hydrogen was proposed. The hydrogenation process uses green hydrogen as the primary hydrogen source, supplemented by gray hydrogen when green hydrogen is insufficient, and supplied to the chemical process via electrothermal conversion when green hydrogen is in excess. This reduces energy and hydrogen storage costs by approximately 80% and transforms fluctuations in green hydrogen flow rate into dynamic changes in the ratio of green to gray hydrogen. The EG synthesis process was analyzed and simulated. The process flow diagram for the hydrogenation of dimethyl oxalate to ethylene glycol using a mixed green-gray hydrogen source is shown below. Figure 1 As shown.

[0048] The hydrogenation of DMO to EG is a typical reaction-separation-recycle process. First, the DMO feed is mixed with a liquid methanol stream, and then with a hydrogen feed and three other gas-phase recycle streams, forming a mixed feed stream. These mixed feedstocks are preheated at the reactor inlet via a preheater before entering the reactor to react. The heat from the preheater comes from heat exchange with the reactor outlet stream, thus improving energy efficiency and saving energy. In the reactor, the DMO hydrogenation reaction produces ethylene glycol, methanol, and ethanol. The main function of the methanol recovery tower is to collect reusable hydrogen and methanol vapor at the top. After flash evaporation, most of the gas is recycled back to the mixed feed stream, while some methanol becomes liquid, serving as the liquid product of the methanol recovery tower. The methanol and ethanol byproducts remaining at the bottom of the tower are separated in the gas stream of the dehydration tower. The liquid stream at the bottom of the dehydration tower mainly consists of EG and methyl glycolate (MG), which enters the EG product tower for further distillation. DMO and MG are drawn from the top of the tower and returned to the feed stream, while polyester-grade ethylene glycol is obtained at the bottom.

[0049] To maximize the economic benefits and minimize the total carbon emissions in the synthesis of EG from mixed green hydrogen and DMO, a multi-objective optimization model of economics and carbon emissions is established, with the purity of EG and CH4O and the upper and lower limits of various operating parameters as constraints.

[0050]

[0051] Here, x represents the set of 8 operational variables, i.e., x = [x1, x2, ..., x8]. Economic benefit is calculated based on profit per unit time, i.e., the value of the main product EG produced per hour, E. EG The value of by-product ET ET The value of the byproduct CH4O (E) CH4O The sum of these is then subtracted from the total cost. The total cost specifically includes the DMO raw material cost C. DMO H2 raw material costs The cost of electricity consumed in the process, C pow and the total thermal energy cost C consumed in the process. heat Carbon emission optimization targets include direct carbon emissions (CE) from organic matter that does not enter the product or is not recycled. d Indirect carbon emissions CE i And carbon emissions from H2 feedstock

[0052]

[0053] Where ω EG and ω CH4O The purity of the products EG and CH4O, respectively; x m、lowerbound and x m、upperbound These represent the lower and upper limits of the m-th operating parameter, respectively, and their values ​​need to be determined according to the specifications in the process manual and sensitivity analysis.

[0054] The calculations for each part of the optimization objective are shown in the following formula:

[0055]

[0056] Among them, F wEG F wET F wCH4O F wDMO F wH2green F wH2gray P represents the mass flow rates of product EG, byproduct ET, byproduct CH4O, consumed DMO, consumed green hydrogen, and consumed gray hydrogen, respectively. EG P ET P CH4O P DMO P H2green P H2grayThese represent the unit prices of the product EG, byproduct ET, byproduct CH4O, DMO feedstock, green hydrogen feedstock, and gray hydrogen feedstock, respectively. pow E powg E heat E heatg These represent the total power consumed in the production process, the green electricity consumed, the total heat energy consumed, and the heat energy converted from green electricity, respectively. P pow P powg P heat P heatg These represent the grid power supply price, green electricity cost price, average heat energy price of steam pipeline network, and electricity-to-heat cost price, respectively. nc η represents the molar flow rate of various organic compounds that do not enter the product or byproduct. CE This indicates the proportion of organic compounds converted into CO2 per unit molar flow rate. CE pow and CE heat η represents the indirect carbon emissions corresponding to electricity consumption and the indirect carbon emissions corresponding to heat consumption, respectively. CEpow η CEpowg η CEheat η CEheatg These represent carbon emissions per unit of electricity supplied by the power grid, per unit of green electricity, per unit of steam pipeline network, and per unit of green electricity converted to heat, respectively. η P η COM η C η h These represent the operating efficiencies of the pump, compressor, distillation column, and heat exchanger, respectively. E P E P1 E P2 E P3 These represent the total power of the pump, P1 power, P2 power, and P3 power, respectively. E COM E COM1 E COM2 E COM3 These represent the total power of the compressor, the power of COM1, the power of COM2, and the power of COM3, respectively. C E C1 E C2 E C3 These represent the total heat consumption, C1 heat consumption, C2 heat consumption, and C3 heat consumption of the distillation column, respectively. H2green and C H2gray These represent the unit carbon emissions of green hydrogen and gray hydrogen, respectively. The default reference prices are: ethylene glycol 4200 yuan / ton, methanol 2500 yuan / ton, ethanol 6900 yuan / ton, DMO reference cost 3000 yuan / ton, gray hydrogen cost 20448 yuan / ton, grid power supply cost 0.725 yuan / kWh, and thermal energy cost 26.28 yuan / GJ.

[0057] For safety reasons, process design manuals often set overly conservative ranges for operating parameters. However, excessively large ranges can lead to a shortage of feasible solutions for initializing the population during optimization. Therefore, sensitivity analysis is used to select operating parameters and determine their ranges. This case study uses eight key operating variables involved in the process flow, including reactor temperature, pressure, pressure drop, the molar ratio of hydrogen to dimethyl oxalate (HDMR), the methanol molar recovery rate at the top of the methanol recovery tower, and the reflux ratios of the three towers. Subsequently, the value ranges of the eight operating variables are selected based on sensitivity analysis, and their typical values ​​are determined using a sequential method. The default values ​​and value ranges of the eight operating variables x and the five interference variables d = [d1, d2, d3, d4, d5] are shown in the table below:

[0058]

[0059] The FPM was normalized and Latin hypercube sampling was used to generate 2000 sets of data containing 8 operating parameters x and 5 perturbation variables d. These data were then input into the FPM, and an offline proxy model was built using the optimization objective value output by the FPM.

[0060] When a significant change is detected in perturbation variables such as the green hydrogen-to-gray hydrogen ratio, feed flow rate, or product price in d, the current perturbation variable d is used. c The initial population is constructed using the Gaussian distribution operation parameter x. In the offline agent model, the NSGA-II algorithm is used to optimize the population to the Nth generation through crossover, mutation, and selection. off Generation, the Nth off Using the operating parameters and current perturbation variables as inputs, a real-world evaluation is performed in FPM to obtain target values ​​f1 and f2, and the number of individuals evaluated (S) is recorded. n Store the dataset S; if the maximum running time t is not reached... max The system enters an online optimization loop when the number of individuals in dataset S reaches S. n Exceeding the online model capacity S a At that time, by removing the individual with the highest non-dominant level, the online agent model is trained using the updated dataset, and the initial population is regenerated for N. on The NSGA-II optimization uses convergence (CI) and diversity (DI) metrics to screen undominated high-quality solutions for FPM real evaluation, continuously enriching the dataset S. After reaching the time threshold, the system extracts the undominated optimal solution from the dataset S and outputs the actually usable optimization parameters through inverse normalization.

[0061] In the optimization comparison experiment, two typical cases were selected: excess green hydrogen and insufficient green hydrogen. When there was excess green hydrogen, the system converted the excess hydrogen energy into heat energy to reduce the carbon emissions caused by heat energy consumption. When there was insufficient hydrogen, the system introduced gray hydrogen, and the balance between economic benefits and carbon emissions was achieved by adjusting the operating parameters. The optimization results of three comparison algorithms—APB-NSGAII, NSGA-II, Pareto Global Efficient Optimization Algorithm (ParEGO), and surrogate model-assisted multi-objective particle swarm optimization (SAPSO)—were compared under the same 1-hour running condition. The experimental results are as follows: Figure 3 , Figure 4 As shown.

[0062] In the simulated online optimization experiment, data from three consecutive months was used, with the carbon tax set at 450 yuan / ton CO2. For ease of comparison, the change times of the five disturbance variables were set to the same time each day. The optimization results of APB-NSGAII, NSGA-II, and the fixed operating parameter strategy were compared. The experiment statistically analyzed the daily economic benefits including the carbon tax. To better compare different operating strategies, we used the economic benefits including the carbon tax under the fixed operating parameters as a benchmark, comparing the optimization results of the APB-NSGAII and NSGA-II algorithms as follows: Figure 5 As shown.

[0063] Optimization results show that, over a cumulative three-month period, compared to the fixed-parameter strategy, the economic benefits including carbon tax increased by RMB 2,291 and 829, respectively, while carbon emissions decreased by 20 and 418 tons of CO2. Compared to the optimization results of NSGA-II, the economic benefits including carbon tax increased by RMB 900 and 520, respectively, while carbon emissions decreased by 11 and 129 tons of CO2. These experimental results demonstrate that APB-NSGAII, compared to NSGA-II and the fixed-parameter strategy, can more effectively balance the goals of improving economic benefits and reducing carbon emissions in long-term operation, proving its superiority in the operational optimization of complex hydrogenation processes coupled with green hydrogen.

[0064] The description and application of the present invention herein are illustrative and not intended to limit the scope of the invention to the embodiments described above. The effects or advantages described in the specification may not be apparent in actual experimental cases due to uncertainties in specific conditions or other factors, and such descriptions are not intended to limit the scope of the invention. Variations and modifications to the embodiments disclosed herein are possible, and various substitutions and equivalents of the components in the embodiments are well known to those skilled in the art. It should be understood by those skilled in the art that the invention can be implemented in other forms, structures, arrangements, proportions, and with other components, materials, and parts without departing from the spirit or essential characteristics of the invention. Other variations and modifications can be made to the embodiments disclosed herein without departing from the scope and spirit of the invention.

Claims

1. A multi-objective optimization method assisted by an online surrogate model for a green hydrogen coupled hydrogenation process, characterized in that, Includes the following steps: S1. Obtain the dynamic change of the flow ratio of green hydrogen to gray hydrogen during the green hydrogen coupled hydrogenation process; S2. Based on the dynamic change of the green hydrogen and gray hydrogen flow ratio, construct a multi-objective operation parameter optimization problem for economic and carbon emission reduction, and perform sensitivity analysis on the operation parameters to obtain the operation parameters to be optimized and their value range. S3. Establish a first-principles model (FPM) for the green hydrogen coupled hydrogenation process, and perform Latin hypercube sampling on the FPM. Based on the sampled data, establish an offline surrogate model using Gaussian process regression. S3 specifically includes: S3-1. A first-principles model (FPM) for the green hydrogen coupled hydrogenation process was established based on the kinetic and thermodynamic mechanisms of the hydrogenation process. Latin hypercube sampling was then used to generate N sets of data containing operating parameters from the model. and disturbance variables Data; S3-2. Input the N sets of data into the first-principles model (FPM), and use the optimization objective function value output by the first-principles model (FPM) to train a GP-based offline agent model. The agent model is as follows: ， in, Indicates GP pair The estimated mean; The error term follows a pattern with a mean of 0 and a variance of . The normal distribution; S3-3, Two input variables randomly selected from the sample and The and The correlation between error terms represents the dependency relationship between function values ​​corresponding to different input points. The Gaussian exponential function is chosen as the kernel function, and its calculation formula is as follows: ， in, This indicates the correlation between two error terms, and its value range is [0, 1]. Indicates input variables Dimensions This represents the scaling parameter used to control the sensitivity of the GP to the input in the k-th dimension. This represents the length scale parameter used to control the smoothness of GP for data points in the k-th dimension; S3-4. Determining hyperparameters by maximizing the logarithmic marginal likelihood function. and The log-marginal likelihood function is maximized as follows: ， in, This represents the input of all training data; This represents the output of all training data; It is the set of all hyperparameters; Indicates the given input data and hyperparameters Output The conditional probability density function; Indicates the amount of training data; Represents the correlation matrix of P×P; T denotes transpose; S4. Construct a surrogate-assisted multi-objective optimization algorithm APB-NSGAII that integrates adaptive parameters, Pareto dual-index sampling, and NSGA-II. This algorithm first generates an initial online surrogate model based on the Gaussian process (GP) with adaptive parameters. Then, it uses a fast non-dominated sorting optimization method to solve the multi-objective operational parameter optimization problem of economics and carbon emission reduction, obtains the optimal operational parameters, and periodically updates the online surrogate model. Multi-objective optimization and model updates alternate until the optimization runtime is exhausted. Specifically, this includes: S4-1. When changes are detected in the green hydrogen to gray hydrogen ratio, feed flow rate, or product price disturbance variables, an initial population is constructed using the current disturbance variable and the operating parameters of the Gaussian distribution. S4-2, Execute the NSGA-II algorithm in the offline agent model for optimization iterations up to the [number missing]. The optimized operating parameters are obtained and input into the first-principles model (FPM) for evaluation to obtain the target value. , The evaluated individuals are then stored in dataset S; the APB-NSGAII algorithm then enters an online optimization loop. S4-3. In the online optimization loop, when the number of individuals in dataset S exceeds the individual capacity of the online agent model, the individual with the lowest non-dominant level is removed to update the dataset. Then, the online agent model is trained and updated, and the initial population is generated using the online agent model for NSGA-II optimization up to the [missing information]. generation; S4-4, from the... From the optimal solutions of the time series, high-quality solutions that are not dominated by convergence and diversity metrics are selected and input into the first-principles model (FPM) for evaluation. The evaluated solutions are then input into the dataset S. If the optimization algorithm's running time does not reach the maximum running time... If the online optimization loop continues, then continue until the desired result is reached. Then the optimal solution in dataset S is output; the multi-objective optimization problem used to filter high-quality solutions that are not dominated by convergence and diversity metrics is as follows: ， ， in, It is the ideal point formed by the optimal values ​​of all objectives in the current population. The surrogate model of the solution predicts the target value. These are individuals from the parent population, and ∥⋅∥ represents the Euclidean distance. The predicted target value used to measure the solution With the ideal point The degree of closeness Used to measure new solutions The degree of difference between the system and the nearest individual in the parent population; after reaching the time threshold, the system extracts the non-dominated optimal solution from the dataset S and outputs the actually usable optimization parameters through inverse normalization.

2. The online proxy model-assisted multi-objective optimization method for the green hydrogen coupled hydrogenation process according to claim 1, characterized in that, The green hydrogen coupled hydrogenation process described in S1 uses green hydrogen as the main hydrogen source. When green hydrogen is insufficient, gray hydrogen is used for dynamic compensation. When green hydrogen is in excess, it is converted into heat energy through electro-thermal conversion to supply the hydrogenation process.

3. The online proxy model-assisted multi-objective optimization method for the green hydrogen coupled hydrogenation process according to claim 1, characterized in that, The objective function for the multi-objective operational parameter optimization problem of economics and carbon emission reduction described in S2 is: ， in, This indicates maximizing economic benefits. This indicates minimizing carbon emissions. Represents the set of operands. This represents the economic benefit of the i-th product. Let j represent the cost of the j-th raw material. This represents the cost of electricity consumed. This represents the total cost of the heat energy consumed. This refers to direct carbon emissions from organic matter that does not enter the product or is not recycled. Indicates indirect carbon emissions. This indicates the carbon emissions resulting from the use of hydrogen as a feedstock; The constraints of the objective function are: ， in, To ensure the purity of the produced products, For products Minimum purity limits; and Representing the first The lower and upper limits of each operation parameter.

Images