Simulation and machine learning based optimization method for directed allocation of hydrogen-containing gas by pressure swing adsorption

Abstract

A method for optimizing directional distribution of hydrogen-containing gas based on simulation and machine learning, comprising: establishing a dynamic simulation of a PSA process by using inlet data of the gas to be separated; selecting operation parameters to be optimized, setting the changeable range and change step of each operation parameter; adjusting the operation parameters in the changeable range by the change step to obtain simulation results; establishing a proxy model by using regression tree analysis and polynomial regression; training the proxy model by a machine learning algorithm according to the simulation results to analyze the relationship between the product indicators and the operation parameters; establishing a nonlinear mathematical model by taking the energy consumption and the production capacity of the PSA process as optimization objectives and the operation conditions of the PSA device as decision variables according to the recovery rate-operation parameter proxy model obtained by the polynomial regression; solving the nonlinear mathematical model to obtain the operation parameters under the optimal operation conditions that meet the constraint conditions; and outputting the operation parameters under the optimal operation conditions to complete the PSA gas distribution optimization.

Description

Technical Field

[0001] This invention relates to the field of pressure swing adsorption (PSA) technology, and more specifically to an optimization method for the directional distribution of hydrogen-containing gases via PSA based on simulation and machine learning. Background Technology

[0002] Pressure swing adsorption (PSA) is an important industrial technology for separating gas mixtures. It has wide applications in CO2 capture, CO enrichment and air separation, especially in the separation and purification of hydrogen in various industrial hydrogen-containing gases.

[0003] PSA (Polysorbent Acid Separation) uses solid adsorbents with selective adsorption properties, typically porous materials such as activated carbon and molecular sieves. These adsorbents possess a strong adsorption capacity for specific gas components. A mixed gas passes through an adsorption column filled with adsorbent, where specific gas molecules are adsorbed into the pore structure of the adsorbent. Different gas components have different adsorption capacities due to their varying affinities for the adsorbent. Gas adsorption occurs under pressure, and desorption occurs under pressure. Therefore, gases can be separated using adsorbents with different properties. The adsorption and desorption stages are controlled by changing the system pressure.

[0004] Typically, the pressure in the adsorption column is first increased to allow the adsorbent to selectively adsorb the target gas component. Then, the pressure is reduced to release the adsorbed target gas component. A common two-tower PSA process mainly consists of six steps: adsorption (AD), pressure equalization (DPE), venting (BD), purging (PG), pressure equalization (PPE), and pressurization (PR). The specific steps are described below:

[0005] 1. Adsorption (AD). The feed gas is continuously fed into the first adsorption tower under a certain pressure Pad. A portion of the high-purity H2 is sent to the second adsorption tower as purge gas, and the other portion is collected from the first adsorption tower as product. The adsorption time needs to be strictly controlled to ensure that the impurity adsorption capacity of the first adsorption tower does not reach saturation, thus preventing impurities from entering the pure hydrogen product;

[0006] 2. Depressurization and equalization (DPE). After the adsorption step, the first adsorption tower has a high pressure and the adsorption capacity reaches its maximum. To reduce the pressure in the first adsorption tower, it is connected to the second adsorption tower, allowing gas to flow from the top of the first tower to the second tower. This reduces the pressure in the first tower and increases the pressure in the second tower.

[0007] 3. Exhaust (BD). The pressure in the first adsorption tower is reduced to near atmospheric pressure by countercurrent flow, and the purge gas containing impurities is discharged from the bottom of the adsorption tower;

[0008] 4. Purging (PG). The first adsorption tower is purged with high-purity H2 obtained from the second adsorption tower undergoing step 1. Through purging, the remaining impurities attached to the first adsorption tower are further desorbed.

[0009] 5. Pressure equalization (PPE). Corresponding to pressure equalization, the first adsorption tower introduces the exhaust gas from the second adsorption tower in a countercurrent manner, thereby achieving initial pressurization;

[0010] 6. Pressurization (PR). The feed gas is fed into the first adsorption tower to further pressurize it. At the end of this step, the pressure in the adsorption tower reaches the adsorption pressure, and one adsorption cycle ends.

[0011] PSA technology is a periodic, unsteady-state separation process in which two adsorption towers alternately complete the six steps described above to separate hydrogen. High-concentration hydrogen is obtained at the top of the adsorption tower, while low-concentration desorbed gas is discharged from the bottom.

[0012] In practice, the number of towers can be expanded and combined according to the separation scale and composition requirements. Traditional PSA systems are characterized by high product H2 purity, low operating costs, and ease of operation.

[0013] However, during the adsorption separation process, adsorbed impurities attached to the adsorption tower need to be purged into the desorbed gas by the product gas, resulting in hydrogen from the product gas also entering the desorbed gas and causing waste. Currently, the highest recovery rate of the PSA process is around 90%, with most falling between 60-80%. Therefore, the larger the scale of hydrogen gas processing in a PSA unit, the more H2 is lost, which is a significant limiting factor for hydrogen separation and enrichment in PSA. Summary of the Invention

[0014] The technical problem to be solved by the present invention is to address the above-mentioned defects in the prior art by providing an optimization method for the directional distribution of hydrogen-containing gas pressure swing adsorption based on simulation and machine learning. This method can transform PSA from a gas separation technology into a gas distribution technology, thereby reducing material waste and improving resource utilization.

[0015] According to the present invention, a method for optimizing the directional distribution of hydrogen-containing gas by pressure swing adsorption based on simulation and machine learning is provided, comprising:

[0016] A dynamic simulation of the PSA process is established using inlet data of the gases to be separated;

[0017] Select the operating parameters to be optimized, and set the range and step size of each operating parameter.

[0018] Within the variable range of each operating parameter, each operating parameter is adjusted one by one in units of their respective change step size to obtain simulation results;

[0019] Regression tree analysis and multinomial regression were used to establish surrogate models to demonstrate the relationship between PSA purity and recovery rate and operating parameters. Regression tree analysis was used to establish the purity-operating parameter surrogate model, while multinomial regression was used to establish the recovery rate-operating parameter surrogate model.

[0020] Based on the simulation results, a proxy model is trained using machine learning algorithms to analyze the relationship between product metrics and operational parameters.

[0021] For the recovery rate-operating parameter surrogate model obtained by multinomial regression, a nonlinear mathematical model is established with energy consumption and plant production capacity in the PSA process as optimization objectives and operating conditions of the PSA plant as decision variables.

[0022] Solve the nonlinear mathematical model to obtain the operating parameters under the optimal operating conditions that satisfy the constraints;

[0023] Output the operating parameters under optimal operating conditions to complete the PSA gas distribution optimization.

[0024] Preferably, the operating parameters to be optimized include adsorption pressure, adsorption time, and purge feed ratio.

[0025] Preferably, the formula for calculating the simulation results is:

[0026]

[0027]

[0028] Wherein, PUR represents the purity of the pure hydrogen product, REC represents the hydrogen recovery rate in the pure hydrogen product, the superscript feed represents the gas feed stream, the superscript ph represents the pure hydrogen product stream, y represents the hydrogen mole fraction, F represents the gas flow rate, and PE represents the cycle.

[0029] Preferably, the surrogate model obtained through regression tree analysis is represented by the following formula:

[0030] Y CT =f(X)

[0031] Where X represents a set of operational variables, Y CT The regression cost function of the surrogate model represents the fitted value predicted using the regression tree method.

[0032]

[0033]

[0034]

[0035] Among them, J CT This represents the cost function of the regression tree method, MSE represents the mean squared error, the subscripts left and right represent the left and right subsets respectively, the subscript node represents the corresponding node, and Y... node Y represents the predicted value of this node. d Let represent the actual value corresponding to the d-th training sample. For the learned model, R is used. 2 Indicating the effect of the fit:

[0036]

[0037] Preferably, the regression cost function obtained by polynomial regression is expressed as:

[0038]

[0039] Where, θ a θ ab θ abc θ con These represent the coefficients of the linear, quadratic, cubic, and constant terms of the polynomial, respectively.

[0040] Preferably, establishing a complete nonlinear mathematical model includes:

[0041] For k components, the feed flow rate is F feed The feed gas has a temperature and pressure of T. feed P feed The purity of component i is y i feed After multiple PSA separation processes with a cycle of PE, two separated gases are obtained; the high-purity hydrogen obtained is represented by the superscript ph, and the other hydrogen-containing gas is represented by the superscript sg; the total cycle time should be equal to the sum of the times of each stage.

[0042]

[0043] Within one cycle, the total flow rate entering the PSA unit should be equal to the total flow rate of the two streams exiting the PSA unit:

[0044]

[0045] Within one cycle, the flow rate of each component entering the PSA unit should be equal to the sum of the flow rates of the corresponding components in the two streams exiting the PSA unit:

[0046]

[0047] Syngas, the product syngas, is another gas besides pure hydrogen. Its total hydrogen flow rate is related to the total flow rate of the raw material hydrogen and the average hydrogen content and recovery rate as follows:

[0048]

[0049] The hydrogen-containing gas from the PSA outlet, used as a synthesis feedstock, should meet the following requirements:

[0050]

[0051]

[0052]

[0053] Where sf represents the coefficient of the ratio of CO, CO2 and H2 during the synthesis process;

[0054] The gas processing capacity per unit time is:

[0055]

[0056] Where PRO represents the gas handling capacity of the device; the energy consumption formula per unit time is:

[0057]

[0058] Where W represents the energy consumption per unit time.

[0059] Preferably, the optimized operating conditions are input into the purity-operating parameter proxy model to verify whether the pure hydrogen product meets the production requirements. Attached Figure Description

[0060] A more complete understanding of the invention and its accompanying advantages and features will be more readily apparent from the accompanying drawings and the following detailed description, wherein:

[0061] Figure 1 A flowchart illustrating a hydrogen-containing gas pressure swing adsorption directional distribution optimization method based on simulation and machine learning according to a preferred embodiment of the present invention is shown.

[0062] Figure 2 An example of the software interface and dynamic simulation process is illustrated.

[0063] It should be noted that the accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Note that the drawings illustrating structures may not be drawn to scale. Furthermore, in the drawings, identical or similar elements are labeled with the same or similar reference numerals. Detailed Implementation

[0064] To make the content of this invention clearer and easier to understand, the content of this invention will be described in detail below with reference to specific embodiments and accompanying drawings.

[0065] The inventors have advantageously discovered that if the PSA is used as a simple hydrogen purification device, both recovery rate and purity cannot be simultaneously achieved during the separation of hydrogen-containing gases using the PSA. This is because, during the desorption process, the PSA device requires a portion of the product hydrogen as a purge gas to regenerate the adsorption tower, thereby carrying impurities adsorbed on the adsorption tower into the purge gas. This working principle inevitably leads to unavoidable hydrogen waste. The purge gas from the PSA device mainly consists of CO, CO2, and hydrogen, i.e., syngas. This means that the components in the hydrogen-containing gas are all useful components. If the PSA is simply used as a hydrogen purification device, its separation effect and hydrogen utilization rate will be limited. However, if the PSA is used as a gas distribution method rather than a simple separation device, by adjusting the operating conditions, the feed gas can be simultaneously separated into two useful gases: one is high-purity hydrogen obtained from the top of the adsorption tower; the other is syngas obtained from the bottom of the adsorption tower. This syngas meets the needs of subsequent chemical product synthesis and has a specific hydrogen-to-carbon ratio. This allocation method starts from production needs, analyzes and calculates the gas composition in the raw material gas, and designs reasonable operating conditions so that all useful components in the raw material can enter the two outlet gases of the PSA unit and directly participate in production, avoiding the generation of purge gas and thus achieving zero waste of hydrogen-containing gas.

[0066] Therefore, this invention proposes a PSA gas distribution device based on process simulation and machine learning modeling optimization, which can distribute the feed gas into two hydrogen-containing gases for different production purposes according to actual needs. This invention is of great significance for reducing the cost of hydrogen-containing gas distribution processes in industry and improving the utilization rate of hydrogen-containing gases.

[0067] Specifically, Figure 1 A flowchart illustrating a hydrogen-containing gas pressure swing adsorption directional distribution optimization method based on simulation and machine learning according to a preferred embodiment of the present invention is shown.

[0068] like Figure 1 As shown, the hydrogen-containing gas pressure swing adsorption directional distribution optimization method based on simulation and machine learning according to a preferred embodiment of the present invention includes:

[0069] For hydrogen-containing syngas, this patent develops a PSA-based gas distribution technology and establishes a mathematical optimization method for gas distribution. The implementation flowchart of the technical solution is shown below. Figure 2 As shown. The specific technical solution is as follows:

[0070] 1. Using AspenAdsorption software, input the inlet data of the gas to be separated, and then use the inlet data of the gas to be separated to establish a dynamic simulation of the PSA process.

[0071] The inlet data for the gas to be separated includes the gas flow rate, the concentration of each component, and the temperature and pressure of the gas to be separated.

[0072] Specific software interfaces and dynamic simulation processes, for example, Figure 2 As shown.

[0073] The following operating parameters that can be changed during the PSA process and have a significant impact on PSA separation performance were selected: adsorption pressure (Pad), adsorption time (tad), and purge feed ratio (PFR), with their respective variable ranges and step sizes set. Different operating parameters have a significant impact on the hydrogen purity and recovery rate of the PSA unit. Specifically: when the adsorption pressure (Pad) increases, the adsorbent's adsorption capacity for the target gas increases, thus increasing hydrogen purity. However, high pressure also leads to more hydrogen being adsorbed, thus reducing the hydrogen recovery rate. Regarding adsorption time (tad), a longer adsorption time can increase the throughput per adsorption cycle and improve the recovery rate. However, excessively long adsorption times will cause the adsorbent to reach its adsorption load, reducing its adsorption capacity for impurities and thus reducing the purity of the hydrogen product. Increasing the purge feed ratio (PFR), i.e., increasing the amount of purge gas, allows impurities adsorbed on the adsorbent to be more fully desorbed, thereby improving the adsorbent's adsorption capacity and increasing the purity of the hydrogen product.

[0074] For example, Figure 1 As shown, based on the provided range and step size of variable X, k sets of X values ​​are set. Based on the selected range and step size, the sampling point locations to be sampled are set, laying the foundation for subsequent simulation and optimization.

[0075] However, a higher purge gas volume also means more hydrogen enters the desorbed gas, reducing the product hydrogen recovery rate. Therefore, it is necessary to select appropriate PSA operating conditions to achieve the separation requirements. Thus, by adjusting the individual operating parameters one by one in a certain step, the parameters can be adjusted in pairs within their variation range, thereby continuously adjusting the pressure swing adsorption separation effect and laying the foundation for optimizing the separation performance.

[0076] 2. The simulation software is controlled via a VBScript program to automate numerous simulations and obtain the simulation results. [X] a Pu a ] and [X a Re a ].

[0077] For example, by writing scripts in VBScript and inputting each set of operating parameters into the AspenAdsorption software based on a pre-defined sampling range, the AspenAdsorption software can be used to perform a rigorous dynamic simulation of the PSA process. This simulation method can obtain the changes in the flow rate, composition, and other states of the outlet stream during the PSA process over time, and thus determine the product purity and recovery rate of the PSA process within the cycle.

[0078] Specifically, the VBScript program controls the process simulation software, automatically changing the operating conditions sequentially according to the set range and step size, simulating, calculating, outputting and recording the corresponding simulation results, including the purity and recovery rate of hydrogen in PSA products, as well as the composition and flow rate of the desorbed gas.

[0079] Since the PSA process is dynamic, the simulation results change over time. Therefore, the purity and recovery rate of the hydrogen product within one cycle need to be obtained through integration. Detailed calculation equations are shown in Formulas 1 and 2. Where PUR represents the purity of the pure hydrogen product, REC represents the hydrogen recovery rate in the pure hydrogen product, the superscript "feed" indicates the gas feed stream, the superscript "ph" indicates the pure hydrogen product stream, y represents the hydrogen mole fraction, and F represents the gas flow rate. PE represents the cycle, and t represents time.

[0080]

[0081]

[0082] Furthermore, the PSA unit can only be considered to be operating stably if the inlet and outlet states remain essentially consistent over multiple consecutive cycles. Therefore, a Cyclic Steady State (CSS) test is required to verify the simulation results, and simulation results that achieve cyclic steady state within 10 cycles must be selected. The process simulation controlled by the VBScript program establishes a complete database of PSA unit product distribution ranges. By selecting data from the database, operating parameters that meet production targets are identified, thereby achieving precise and targeted distribution of hydrogen-containing gas based on the PSA.

[0083] In this way, the separation results of PSA devices under k different operating conditions can be obtained automatically, and a sufficient sample database can be established. These data lay the foundation for subsequent machine learning and mathematical optimization.

[0084] 3. Based on the obtained simulation results database, train the agent model using machine learning algorithms and analyze the relationship between product indicators and operating parameters.

[0085] In the gas distribution process, the purity of the pure hydrogen product is a key performance indicator, typically requiring a specific purity level for production. However, the composition of the syngas (hydrogen-to-carbon ratio) must meet production demands, necessitating adjustments to the pure hydrogen recovery rate to a specific value. Therefore, to address the different requirements for purity and recovery rate, surrogate models were established using regression tree analysis and multinomial regression to demonstrate the relationship between PSA purity and recovery rate and operating parameters, applicable to the final gas distribution optimization. Regression tree analysis was used to establish the purity-operating parameter surrogate model, while multinomial regression was used for the recovery rate-operating parameter surrogate model.

[0086] In the gas distribution process, the purity of the pure hydrogen product is a key performance indicator, typically requiring a specific purity level for production. However, the composition of the syngas (hydrogen-to-carbon ratio) needs to meet production requirements, necessitating adjustments to the pure hydrogen recovery rate to a specific value. Therefore, to address the different requirements for purity and recovery rate, surrogate models were established using regression tree analysis and multinomial regression to demonstrate the relationship between PSA purity and recovery rate and operating parameters, applicable to the final gas distribution optimization. Regression tree analysis was used to establish the purity-operating parameter surrogate model, while multinomial regression was used for the recovery rate-operating parameter surrogate model. The surrogate model obtained through regression tree analysis can be represented by Equation 3:

[0087] Y CT =f(X) ss (3)

[0088] Where X represents a set of operational variables, Y CT This represents the fitted value predicted using the regression tree method. The regression cost function of this method can be expressed by Equation 4-6:

[0089]

[0090]

[0091]

[0092] Among them, J CT Y represents the cost function of the regression tree method, MSE represents the mean squared error, the subscripts left and right represent the left and right subsets respectively, and the subscript node represents the corresponding node. node Y represents the predicted value of this node. d Let R represent the actual value corresponding to the d-th training sample. For the learned model, R... 2 It is used to represent the effect of fitting, as shown in Equation 5:

[0093]

[0094] R 2 It is a number between 0 and 1, and the closer it is to 1, the better the regression effect. This represents the mean of the actual values ​​of the training samples. Generally, a value greater than 0.8 is considered to indicate a higher goodness of fit.

[0095] The equation obtained by polynomial regression can be expressed by Formula 8, where h(X) represents the corresponding fitted value obtained by polynomial regression, and θ a θ ab θ abc θ con Let each of the following terms represent the coefficients of the linear, quadratic, cubic, and constant terms of the polynomial:

[0096]

[0097] Similar to the regression tree method, MSE is used as the cost function in multinomial fitting, R 2 It is used to represent the goodness of fit of the model.

[0098] 4. The recovery rate-operating parameter surrogate model obtained from polynomial regression has a rigorous mathematical formula. Combining it with other mathematical models of the PSA process (Formulas 9-17) yields a nonlinear (NLP) mathematical model. For example, this nonlinear (NLP) mathematical model can be solved using GAMS software and the Baron solver to obtain the optimal operating conditions that satisfy the constraints, where the optimization objective is to minimize the energy consumption of the device per unit time. For example, ... Figure 1 The equation Re = f(x) is added to the constraints for optimization. This allows us to output the operating parameters under optimal operating conditions, thus completing the PSA gas distribution optimization.

[0099] Specifically, for the recovery rate-operating parameter surrogate model obtained by multinomial regression, energy consumption and plant production capacity in the PSA process are used as optimization objectives, and the operating conditions of the PSA plant are used as decision variables to establish a complete NLP (nonlinear) mathematical model.

[0100] Specifically, for example, for k components, the feed flow rate is F feed The feed gas has a temperature and pressure of T. feed P feed The purity of component i is y i feed After multiple PSA separation processes with a cycle length of PE, two separated gases are obtained; the high-purity hydrogen obtained is denoted by the superscript ph, and the other hydrogen-containing gas is denoted by the superscript sg. Since the PSA process is dynamic, the flow rate, composition, temperature, and pressure of each stream change over time.

[0101] The total cycle time should equal the sum of the times of each stage, as shown in Formula 9:

[0102]

[0103] Within one cycle, the total flow rate entering the PSA unit should be equal to the total flow rate of the two streams exiting the PSA unit, as shown in Formula 10:

[0104]

[0105] Accordingly, within one cycle, the flow rate of each component entering the PSA unit should be equal to the sum of the flow rates of the corresponding components in the two streams exiting the PSA unit, as shown in Formula 11:

[0106]

[0107] The product synthesis gas is another gas besides pure hydrogen. Its total hydrogen flow rate is related to the total feed hydrogen flow rate and the average hydrogen content and recovery rate, as shown in Formula 12.

[0108]

[0109] For the synthesis of chemical products, the ratio of hydrogen to CO and CO2 in the feed gas often needs to satisfy a certain mathematical relationship. The hydrogen-containing gas from the PSA outlet, used as a synthesis feedstock, should satisfy formula 13-...

[0110] The constraints of Equation 15. Where sf represents the coefficient representing the ratio of CO, CO2, and H2 during the synthesis process:

[0111]

[0112]

[0113]

[0114] For the PSA process described in this paper, the gas throughput per unit time can be expressed by formula 16:

[0115]

[0116] Where PRO represents the gas processing capacity of the unit. Since the energy consumption of the PSA process is mainly due to the compression of the feed gas, the energy consumption per unit time is calculated as shown in Formula 17. W represents the energy consumption per unit time:

[0117]

[0118] Energy consumption in the PSA process is an important evaluation indicator; therefore, the optimization objective is to minimize energy consumption per unit time.

[0119] Therefore, the operating parameters can be output to initially complete the PSA gas distribution optimization.

[0120] 5. Finally, data can be substituted to validate the model. For example, the obtained optimization result X* can be input into the surrogate model.

[0121] This model optimizes energy consumption throughout the process, simultaneously distributing hydrogen-containing gas into pure hydrogen and syngas that meets the required proportions for chemical synthesis. This allows for the selection of PSA unit operating conditions that minimize energy consumption per unit time while achieving gas distribution. Thus, this invention transforms PSA from a gas separation technology into a gas distribution method.

[0122] For example, the PSA gas distribution method can produce two product gases: high-purity hydrogen at the top of the adsorption tower and the remaining syngas at the bottom. Since the composition of the remaining syngas is constrained in the mathematical model, the syngas obtained under the optimal operating conditions will necessarily meet the requirements. The purity of the product hydrogen is an important verification indicator, Pu, set to be greater than 99.9%. Here, the purity only needs to meet the requirement, not a specific value. Therefore, the optimized operating conditions are input into the previously obtained purity-operating parameter surrogate model, and the predicted results can be used to verify whether the pure hydrogen product meets the production requirements. After verification, the operating parameters are output to complete the PSA gas distribution optimization.

[0123] For a two-tower PSA unit Figure 2 This diagram illustrates an example of the software interface and dynamic simulation process. The process in this example is a typical two-tower PSA unit. Hollow arrows represent feedstock and product, FEED represents the gas feed, PPP represents the pure hydrogen product stream obtained at the top of the adsorption tower, and WASTE represents the syngas stream typically obtained at the bottom of the tower. T1 and T2 represent the two adsorption towers, TD1, TD2 and TW1, TW2 represent the gas buffer tanks at the top and bottom of the two towers, respectively, and TF, TW, and TP represent the gas buffer tanks for the feedstock and the two product gas streams, respectively. VF, VW, VP, and VPURGE represent valves in the process, and the periodic operation of the PSA process is controlled by the opening and closing status of these valves at different stages. Specific operating conditions are shown in the table below. Here, 1 represents a valve open, and 0 represents a valve closed.

[0124]

[0125] Compared with existing technologies and methods, this invention has the following innovative effects:

[0126] 1. The gas distribution technology based on PSA proposed in this invention differs from traditional PSA separation in principle. It no longer focuses solely on the pure hydrogen outlet of the PSA unit, but takes a holistic approach, aiming to improve the overall resource utilization rate of the process. By adjusting the production conditions of the PSA unit, the originally unusable desorbed gas is transformed into syngas with a specific composition that can be directly utilized.

[0127] 2. Based on PSA gas distribution technology, this invention establishes a complete mathematical model for gas distribution optimization. This model can rigorously optimize PSA gas distribution technology. Through this optimization method, not only can the precise distribution of hydrogen-containing gases be achieved, but also the operating conditions with the lowest energy consumption can be selected within all feasible ranges, further saving energy consumption while improving resource utilization.

[0128] 3. This invention combines process simulation, machine learning, and mathematical optimization, and applies them to the distribution of hydrogen-containing gases, which is a novel approach.

[0129] Furthermore, it should be noted that, unless otherwise specified, the terms "first," "second," "third," etc., in the specification are used only to distinguish the various components, elements, and steps in the specification, and are not used to indicate the logical or sequential relationships between the various components, elements, and steps.

[0130] It is understood that although the present invention has been disclosed above with reference to preferred embodiments, these embodiments are not intended to limit the present invention. For any person skilled in the art, many possible variations and modifications can be made to the technical solutions of the present invention based on the disclosed technical content, or equivalent embodiments with equivalent changes, without departing from the scope of the present invention. Therefore, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An optimization method for the directional distribution of hydrogen-containing gas by pressure swing adsorption based on simulation and machine learning, characterized in that... include: A dynamic simulation of the PSA process is established using inlet data of the gases to be separated; Select the operating parameters to be optimized, and set the range and step size of each operating parameter. Within the variable range of each operating parameter, each operating parameter is adjusted one by one in units of their respective change step size to obtain simulation results; Regression tree analysis and multinomial regression were used to establish surrogate models to demonstrate the relationship between PSA purity and recovery rate and operating parameters. Regression tree analysis was used to establish the purity-operating parameter surrogate model, while multinomial regression was used to establish the recovery rate-operating parameter surrogate model. Based on the simulation results, a proxy model is trained using machine learning algorithms to analyze the relationship between product metrics and operational parameters. For the recovery rate-operating parameter surrogate model obtained by multinomial regression, a nonlinear mathematical model is established with energy consumption and plant production capacity in the PSA process as optimization objectives and operating conditions of the PSA plant as decision variables. Solve the nonlinear mathematical model to obtain the operating parameters under the optimal operating conditions that satisfy the constraints; Output the operating parameters under optimal operating conditions to complete the PSA gas distribution optimization.

2. The method according to claim 1, characterized in that, The operating parameters to be optimized include adsorption pressure, adsorption time, and purge feed ratio.

3. The method according to claim 1 or 2, characterized in that, The formula for calculating the simulation results is as follows: Wherein, PUR represents the purity of the pure hydrogen product, REC represents the hydrogen recovery rate in the pure hydrogen product, the superscript feed represents the gas feed stream, the superscript ph represents the pure hydrogen product stream, y represents the hydrogen mole fraction, F represents the gas flow rate, and PE represents the cycle.

4. The method according to claim 1 or 2, characterized in that, The surrogate model obtained through regression tree analysis is represented by the following formula: Y CT =f(X) Where X represents a set of operational variables, Y CT The regression cost function of the surrogate model represents the fitted value predicted using the regression tree method. Among them, J CT This represents the cost function of the regression tree method, MSE represents the mean squared error, the subscripts left and right represent the left and right subsets respectively, the subscript node represents the corresponding node, and Y... node Y represents the predicted value of this node. d Let represent the actual value corresponding to the d-th training sample. For the learned model, R is used. 2 Indicating the effect of the fit:

5. The method according to claim 1 or 2, characterized in that, The regression cost function obtained by polynomial regression is expressed as: Where, θ a θ ab θ abc θ con These represent the coefficients of the linear, quadratic, cubic, and constant terms of the polynomial, respectively.

6. The method according to claim 1 or 2, characterized in that, Establishing a complete nonlinear mathematical model includes: For k components, the feed flow rate is F feed The feed gas has a temperature and pressure of T. feed P feed The purity of component i is y i feed After multiple PSA separation processes with a cycle of PE, two separated gases are obtained; the high-purity hydrogen obtained is represented by the superscript ph, and the other hydrogen-containing gas is represented by the superscript sg; the total cycle time should be equal to the sum of the times of each stage. Within one cycle, the total flow rate entering the PSA unit should be equal to the total flow rate of the two streams exiting the PSA unit: Within one cycle, the flow rate of each component entering the PSA unit should be equal to the sum of the flow rates of the corresponding components in the two streams exiting the PSA unit: Syngas, the product syngas, is another gas besides pure hydrogen. Its total hydrogen flow rate is related to the total flow rate of the raw material hydrogen and the average hydrogen content and recovery rate as follows: The hydrogen-containing gas from the PSA outlet, used as a synthesis feedstock, should meet the following requirements: Where sf represents the coefficient of the ratio of CO, CO2 and H2 during the synthesis process; The gas processing capacity per unit time is: Where PRO represents the gas handling capacity of the device; the energy consumption formula per unit time is: Where W represents the energy consumption per unit time.

7. The method according to claim 1 or 2, characterized in that... Also includes: The optimized operating conditions are input into the purity-operating parameter proxy model to verify whether the pure hydrogen product meets the production requirements.

Images