A method, system, storage medium and equipment for optimizing a pressure swing adsorption coupled hydrogen production process based on a cross-platform dynamic decoupling algorithm

Abstract

This invention discloses an optimization method, system, storage medium, and equipment for a pressure swing adsorption (PSA) coupled hydrogen production process based on a cross-platform dynamic decoupling algorithm, belonging to the field of chemical system engineering and process simulation. It constructs an integrated master control unit, a steady-state process solver, a PSA dynamic model solver, an economic calculation module, and a response surface optimization module. In the PSA dynamic model solver, the physical property matrix, geometric parameters, and adsorption isotherm parameters are initialized; the state variable matrix covering the system state is initialized; a nested loop solution process between the PSA dynamic model solver and the steady-state process solver is entered; when a strict cyclic steady state is reached, the master control unit organizes and calculates the parameters; the economic calculation module performs economic calculations; finally, the response surface optimization module constructs a response surface proxy model, using preset indicators as collaborative optimization targets, and outputs the optimal operating parameter configuration. This invention achieves mechanism calculation and efficient collaborative optimization of the PSA coupled hydrogen production process.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of chemical systems engineering, numerical calculation of separation processes and process optimization. Specifically, it relates to an optimization method and system for a pressure swing adsorption coupled hydrogen production process, and more particularly to a high-fidelity dynamic mechanism modeling for heterogeneous multilayer fixed beds, a rigid decoupling solution of complex partial differential equations (PDEs) and its application in process digital twinning and techno-economic optimization. Background Technology

[0002] In the context of the current deep decarbonization of energy transition and upgrading of the chemical industry, the efficient purification and high-value recovery of hydrogen from low-grade industrial waste gases (such as syngas from waste gasification, blast furnace gas, and refinery dry gas) or reforming reaction products using mixed gas separation technology has significant engineering implications. To balance high recovery rates and high purity, industrial processes often couple enrichment units (such as membrane separation units, absorption units, or upstream reaction processes) with pressure swing adsorption (PSA) units used for purification, and recirculate the desorbed gas or part of the product gas generated by PSA, constructing a complex "purification-reflux" coupled hydrogen production process.

[0003] However, performing high-fidelity techno-economic evaluations and optimizing process parameters for such coupled reflux systems currently faces significant software platform barriers and computational mathematical bottlenecks. Most mainstream macroscopic chemical steady-state process simulation software lacks a solution model that accurately describes the dynamic mechanism of pressure swing adsorption (PSA), while specialized dynamic adsorption simulation software exhibits a high degree of independence, making efficient and high-frequency cross-platform data exchange difficult. This makes accurate joint iteration and steady-state calculations difficult for coupled PSA systems with reflux. Specifically, existing technologies face significant technical challenges in the following three aspects: (1) At the model establishment and solution level: Pressure swing adsorption (PSA) involves complex multiphase mass, heat and momentum transfer. Industrial-grade PSA typically uses multilayer adsorbents. When dealing with the interfaces between different adsorbents, traditional discretization methods are prone to sudden changes in physical parameters, which can cause a surge in the rigidity of local equations. At the same time, the steep pressure jump caused by instantaneous valve switching, if directly coupled with the momentum balance equation for iteration, can lead to frequent non-physical oscillations or even collapse of the partial differential equation solver. In addition, simulating a complete multi-tower, multi-step time-series process often requires the construction of an extremely large global state matrix, resulting in lengthy calculation times for each loop, making large-scale parameter optimization infeasible in terms of engineering time.

[0004] (2) In complex coupled processes, due to the strong nonlinear material reflux loop between the front-end process feed and the pressure swing adsorption unit (such as the full or partial reflux of desorbed gas to the upstream), a "tear stream" must be manually set for iterative calculation during co-simulation. At this time, the dynamic pressure swing adsorption model is very prone to the "ghost inventory effect" due to unreasonable initial bed conditions or guessed values. That is, a false mass source is released into the reflux loop during the numerical calculation process, resulting in non-conservation of system materials. This causes the mathematical iteration to eventually converge to a pseudo steady state that violates the mass conservation limit (such as the total hydrogen recovery rate of the system exceeding 100% in calculation), leading to the failure of process simulation.

[0005] (3) Conventional methods often directly deduct the calorific value of by-products by linearly based on the market-market benchmark gas price when calculating the economic indicators of separation processes (such as hydrogen separation costs). This assumption completely ignores the fact that a large number of inert non-combustible components (such as CO2 and N2) in the by-products fluctuate drastically with changes in operating conditions, thus causing nonlinear degradation effects on the actual combustion temperature and Wobbe index of the exhaust gas. This one-sided economic accounting is very likely to lead to mathematical distortions where the separation cost is negative under certain operating conditions, thereby misleading the optimization direction of the global response surface and failing to obtain the optimal operating parameters that are truly feasible in engineering. Summary of the Invention

[0006] To overcome the difficulties faced by the existing technologies, the present invention aims to provide an optimization method, system, storage medium, and equipment for pressure swing adsorption (PDE) coupled hydrogen production process based on a cross-platform dynamic decoupling algorithm. This method represents a comprehensive innovation, from PDE mechanism equation discretization and multi-bed boundary reconstruction to full-process mass conservation control and double-layer nested steady-state iteration. It effectively improves computational efficiency while ensuring strict physical fidelity and introduces an objective economic discount benchmark to achieve multi-objective global optimization in an engineering sense. To achieve the above objectives, the present invention adopts the following technical solution: An optimization method for pressure swing adsorption coupled hydrogen production process based on a cross-platform dynamic decoupling algorithm includes the following steps: S1: Construct a cross-platform co-simulation architecture, which integrates a main control unit, a steady-state process solver, a pressure swing adsorption (PSA) dynamic model solver, an economic calculation module, and a response surface optimization module. For a PSA coupled process including a circulating reflux stream, a breakpoint on the reflux closed loop is set as a tear stream on the main control unit, and the predicted values ​​of the tear stream parameters are initialized. The feed stream and equipment parameters are input on the main control unit and transmitted to the steady-state process solver for steady-state process calculation. The parameters of the PSA unit inlet stream output by the steady-state process solver are obtained and transmitted to the PSA dynamic model solver. The stream parameters include the temperature, pressure, flow rate, and content of each gas phase component of the corresponding stream. The predicted values ​​of the tear stream parameters in step S1 are the initial predicted values ​​of the stream parameters set for the tear stream. S2: In the pressure swing adsorption dynamic model solver, based on the inlet flow parameters of the pressure swing adsorption unit, the physical property matrix, geometric parameters, and adsorption isotherm parameters are initialized; the adsorption bed is divided into discrete spatial grids, and a smooth transition function is used to perform nonlinear interpolation mapping of the physical property matrix and adsorption isotherm parameters of different adsorbent layers along the axial direction on the discrete spatial grid to obtain the spatial distribution of the physical property matrix and adsorption isotherm parameters after smooth transition; S3: In the pressure swing adsorption dynamic model solver, initialize the state variable matrix covering the system state: At the initial moment of the cycle, use the initial gas phase composition determined according to the start-up purging condition or the inlet flow parameters of the pressure swing adsorption unit to initialize the state variable matrix and obtain the initial state matrix of the pressure swing adsorption dynamic model solver. S4: Enter the nested loop solution process between the pressure swing adsorption dynamic model solver and the steady-state process solver, and the main control terminal performs variable transfer and judgment between the above modules; S5: Once a strict cyclic steady state is reached, the main control unit extracts the state variable matrix, product gas and desorbed gas stream parameters from the pressure swing adsorption dynamic model solver, and by-product stream parameters from the steady-state process solver for processing and calculation. When calculating the product gas flow rate, the amount of gas returning from the product gas main pipe to the adsorption tower for process consumption within the pressure swing adsorption cycle is deducted to obtain the net product gas yield and purity. The stream parameters of the net product gas, desorbed gas, and by-products after deduction are then processed into total stream parameters. S6: The main control unit transmits the total flow parameters, equipment parameters, and utility consumption parameters extracted from the steady-state process solver to the economic accounting module; finally, it calls the response surface optimization module to construct a response surface proxy model, uses preset indicators as the collaborative optimization target, and outputs the optimal operation parameter configuration.

[0007] Furthermore, the smooth transition function mentioned in step S2 is a Sigmoid-type logic function, and its formula is: ,in This is the transition steepness constant. It is the axial distance. This refers to the location of the interface.

[0008] Furthermore, step S4 specifically includes: (a) Inner layer operation step integral loop: The pressure swing adsorption dynamic model solver receives the inner layer tolerance (preset as the first inner layer tolerance for the first calculation) sent by the main control terminal, reads the loop timing table, and constructs the real-time boundary conditions of the partial differential equation system based on the geometric parameters of step S2, the physical property matrix after smooth transition, the spatial distribution of adsorption isotherm parameters, the initial state matrix of step S3, and the current timing step. The real-time pressure boundary is smoothed by continuous harmonic functions. Based on the set real-time boundary conditions and the physical property matrix, geometric parameters, and adsorption isotherm parameters, the partial differential equation system and its real-time boundary conditions are discretized onto a spatial grid using a first-order upwind difference scheme, and the convection term is determined. The algorithm proceeds in the forward direction, calling the variable step size numerical integration algorithm for rigid ordinary differential equations for adaptive numerical integration; after integration, the spatial physical field state at the end of the current time step is recorded as the initial value for the next time step; the instantaneous molar flow rate integral in the end spatial physical field state is accumulated into the virtual buffer tank model for use by other steps, and the total amount of product gas and desorbed gas before deducting internal consumption is calculated simultaneously; the predicted value of the tear stream parameters for a single complete cycle is output; this process is repeated until the relative residuals of all tear stream parameters in the two complete cycles before and after the pressure swing adsorption dynamic model solver meet the current inner layer tolerance, the predicted value of the tear stream parameters for the last single complete cycle is output, and the process proceeds to step (b). (b) Outer loop steady state determination and dynamic tolerance allocation: The first relative error calculation is to calculate the relative error of the parameters in the initial guess value of the tearing flow parameters in step S1 and the guess value of the tearing flow parameters in the last single complete cycle of the pressure swing adsorption dynamic model solver. The subsequent relative error calculation is to calculate the relative error of the parameters in the guess value of the tearing flow parameters output by the pressure swing adsorption dynamic model solver in the two outer loops before and after.

[0009] When the relative error of any parameter in the predicted tear flow parameters is greater than or equal to the outer layer error threshold, the outer layer loop enters a quasi-steady-state iteration period and issues a first inner layer tolerance to the inner layer loop. A first relaxation factor is assigned to parameters whose relative error in the predicted tear flow parameters is greater than the relaxation factor allocation threshold, and a second relaxation factor is assigned to parameters whose relative error is not greater than the relaxation factor allocation threshold, with the first relaxation factor being greater than the second relaxation factor. The predicted tear flow parameters are updated using the assigned relaxation factors and fed back to the steady-state process solver for calculation. After re-acquiring the pressure swing adsorption unit inlet flow parameters output by the steady-state process solver, the loop returns to step (a) for the inner layer loop. When the relative errors of all parameters in the predicted values ​​of the tearing flow parameters are less than the outer layer error threshold, the steady-state iteration period of the outer layer cycle is entered, and a second inner layer tolerance less than the first inner layer tolerance is issued to the inner layer cycle. The pressure swing adsorption unit inlet flow parameters output by the current steady-state process solver are locked as the constant feed flow parameters of the pressure swing adsorption dynamic model solver. Step (a) is continued until the inner layer cycle meets the second inner layer tolerance, reaches a strict cyclic steady state, and ends the inner and outer layer cycles. Furthermore, the partial differential equations include the overall mass continuity equation, the overall energy balance equation of the mixed bed, the transient heat conduction equation of the tower wall, and the mass transfer kinetic equation; and in numerical calculations, the local ideal gas equation of state is combined as an algebraic constraint condition for joint solution.

[0010] Furthermore, the continuous harmonic function of the real-time pressure boundary described in step S4(a) is: ,in and These are the initial and target pressures, respectively. For normalized dimensionless time.

[0011] Further, in step S4(b), the order of magnitude of the outer layer error threshold is set to 10. -4 Up to 10 -3 The first inner layer tolerance is greater than the second inner layer tolerance; the relaxation factor allocation threshold is less than 15%, the first relaxation factor is set to 0.60 to 0.99, and the second relaxation factor is set to 0.10 to 0.60; The estimated values ​​of the tear flow parameters, which are then updated using a relaxation factor weighting method, are calculated using the following formula: ,in, These are the guessed values ​​for the updated tearing flow parameters. The first relaxation factor or the second relaxation factor is allocated based on the relaxation factor allocation threshold. These are the estimated values ​​of the tearing flow parameters output from a single complete cycle of the pressure swing adsorption dynamic model solver. These are the estimated parameters for the tear stream from the previous outer cycle. After updating the estimated values ​​for the tear stream, a forced normalization step for the mole fractions of the gas phase components is also included.

[0012] Furthermore, in step S6, the formula is used. Calculate the cost of hydrogen separation SC ;in, The total annual cost is the sum of the annualized depreciation of the equipment, the annual cost of raw gas, and the annual operating expenses. For net product hydrogen production, The by-product calorific value revenue is calculated based on the benchmark gas price. The discount factor is used to characterize the deterioration of the calorific value of by-products.

[0013] A system for optimizing a pressure swing adsorption coupled hydrogen production process based on a cross-platform dynamic decoupling algorithm, applied to the above-mentioned method, the system comprising: The main control unit is configured to perform cross-platform data integration, parameter distribution, scheduling system timing calculations, and multi-objective collaborative optimization. The steady-state process solver is configured to receive the predicted values ​​of the tear stream, perform macroscopic steady-state process simulation calculations, and output the inlet stream parameters, by-product stream parameters, and utility consumption of the pressure swing adsorption unit. The dynamic model solver for pressure swing adsorption embeds a linear method spatial discrete grid and a rigid ordinary differential equation variable step size numerical integrator, configured to perform inner and outer double-layer nested loop dynamic solution to extract spatiotemporal physical field and flow characteristic information. The economic accounting module is configured to receive steady-state flow parameters and utility consumption, and calculate and output technical and economic evaluation indicators. The response surface optimization module is configured to build a proxy model based on multi-condition data and perform multi-objective collaborative optimization.

[0014] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0015] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described above.

[0016] Compared with the prior art, the present invention has the following advantages and beneficial technical effects: (1) Constructing a high-fidelity heterogeneous multilayer bed dynamic model: This invention establishes a rigorous non-isothermal dynamic pressure swing adsorption mathematical model and adopts a multiple isotherm strategy to accurately characterize the physical adsorption and chemical complexation characteristics of different adsorbents. At the same time, by constructing a dual-temperature coupled energy equation covering local thermodynamic equilibrium and tower wall thermal inertia, and innovatively using a smooth transition function to perform nonlinear interpolation mapping of physical property parameters on the spatial grid, the parameter abrupt changes at the interface of different adsorbents are eliminated, and a precise mathematical description of the internal heat and mass transfer phenomenon of heterogeneous multilayer bed pressure swing adsorption is realized.

[0017] (2) Significantly improves solution efficiency and numerical stability: This invention introduces a continuous harmonic cosine function to smooth the rigidity of the pressure field and uses a virtual buffer tank to decouple the massive and intertwined multi-tower coupling into a single-tower periodic iteration with strict closed-loop mass conservation. This algorithm, combined with a variable step size numerical integration strategy, significantly reduces the simulation calculation time for a single complete physical cycle to the order of seconds (achieving a simulation speedup of nearly 100 times), solving the problems of numerical divergence and excessively long calculation time in complex pressure swing adsorption processes.

[0018] (3) Global Steady-State Determination and Rapid Convergence: In the outer-layer steady-state decoupling determination, this invention strictly adopts the maximum relative error of all parameters of the torn stream as the convergence criterion, fundamentally eliminating the risk of premature convergence due to the achievement of a single parameter. Simultaneously, it innovatively assigns different relaxation factors to different parameters based on the relaxation factor allocation threshold, scientifically weighting the guessed values ​​of the torn stream parameters, and supplementing this with forced normalization of the gas phase components. This mechanism can accelerate convergence for trace impurities that are difficult to converge, and maintain numerical stability under the impact of stream parameter fluctuations, ensuring that the system converges to the physically true global cyclic steady state within a very small number of iterations.

[0019] (4) Realize cross-platform joint simulation and digital twin: This invention breaks through the technical barrier of the lack of dynamic mechanism model of pressure swing adsorption in the current mainstream business process simulation software, and constructs a cross-platform joint simulation architecture driven by the main control end, realizing a panoramic closed-loop digital twin of pressure swing adsorption coupling process from micro-partial differential mechanism solution to macro-full process economic evaluation.

[0020] (5) Establishing an objective economic evaluation system and cost optimization: In view of the cost distortion caused by the abnormally high calorific value of waste gas under low recovery rate conditions in conventional economic models, this invention introduces an industrial discount factor that reflects the characteristics of low-pressure desorbed gas and high inert components into the techno-economic accounting. This method objectively defines the economic boundary of the separated system and eliminates mathematical distortions.

[0021] (6) Achieving automated batch multi-objective collaborative optimization: This invention utilizes a response surface surrogate model (based on Box-Behnken experimental design), abandoning the inefficient traditional mode that relies on engineers manually inputting and modifying parameters one by one. With maximizing hydrogen recovery rate and minimizing hydrogen separation cost as the collaborative optimization objectives, and under the premise of ensuring physical constraints and purity requirements, it realizes the systematic automatic traversal of the key parameter space of the coupled process, greatly improving optimization efficiency and ease of engineering implementation. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of a cross-platform co-simulation architecture driven by the main control unit. Figure 2Flowchart of membrane separation-pressure swing adsorption coupled hydrogen extraction process Figure 3 Pressure evolution curves of the bed in a 6-tower, 24-step pressure swing adsorption process. Figure 4 The convergence process of the relative error of the tearing flow parameters during the quasi-steady-state iteration period. Figure 5 The convergence history of all stream parameters of the exhaust gas at the top of the tower during the final 30 cycles in the steady-state iteration period is as follows: (a) CO concentration, (b) CO2 concentration, (c) CH4 concentration, (d) N2 concentration, (e) H2 concentration, (f) temperature, (g) pressure, (h) flow rate. Detailed Implementation

[0023] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. Furthermore, to clearly illustrate the cross-platform dynamic decoupling algorithm and architecture of the present invention, specific software models (such as MATLAB, UniSim Design, etc.) and devices are used as examples in the embodiments. However, the scope of protection of the present invention is not limited to specific software environments; all chemical process simulation software systems fall within the scope of protection of the present invention.

[0024] Example 1: Global Optimization of Low-Grade Gasification Syngas Film Separation-Pressure Swing Adsorption Coupled Hydrogen Production Process Based on the system described in this invention, this embodiment runs on an electronic computer device equipped with a 13th Gen Intel(R) Core(TM) i7-13700H 2.40 GHz processor and 32GB RAM. The computer program in the memory calls each computing module to complete the following optimization process of pressure swing adsorption coupled hydrogen production.

[0025] This embodiment focuses on the deep separation and purification of syngas from the gasification of specific low-grade waste gas. The goal is to obtain fuel cell-grade hydrogen (H2 ≥ 99.97%, CO ≤ 0.2 ppm) that meets the standards for proton exchange membrane fuel cells, while maximizing the system's hydrogen recovery rate and minimizing hydrogen separation cost (SC). The basic feed boundary conditions for the feed gas are set as follows: feed temperature 298.15 K, pressure 101.325 kPa, and total flow rate 20000 Nm³. 3 / h; initial component mole fractions: CO 28.53%, CO2 10.36%, H2 31.07%, N2 26.93%, CH4 3.11%.

[0026] The specific implementation steps of this embodiment strictly follow the method described in the claims of this invention, and are detailed below: S1: This embodiment uses MATLAB as the platform to build a main control unit, a pressure swing adsorption dynamic model solver, and a response surface optimization module. It establishes high-frequency bidirectional data transmission with the steady-state process solver (UniSimDesign in this embodiment) and the economic accounting module (based on Excel) through the Component Object Model (COM) protocol. Its data interaction architecture is as follows: Figure 1 As shown.

[0027] The stream parameters mentioned in this method include the temperature, pressure, flow rate, and content of each gas phase component of the corresponding stream. A coupled hydrogen production process physical topology is constructed in the steady-state process solver (as shown in Figure 2): the feed gas is pressurized by a compressor and enters the first-stage membrane separator (HM-1). The permeate is discharged as a fuel gas byproduct, and the permeate enters the second-stage membrane separator (HM-2). The permeate from the second-stage membrane is refluxed to the front end of the first-stage membrane, and its permeate is sent to the downstream pressure swing adsorption (PSA) unit. The desorbed gas generated by the PSA unit is fully refluxed to the mixer at the front end of the second-stage membrane, forming a closed-loop material reflux.

[0028] To enable mathematical closed-loop solution of the system, the main control unit sets a disconnection point on the reflux line as the tear stream and initializes the predicted values ​​of the tear stream parameters. Subsequently, the feed stream parameters, equipment parameters of each stage of membrane separation unit, and the predicted values ​​of the tear stream parameters are input into the main control unit and transmitted to the steady-state process solver for steady-state process calculation. The output parameters of the pressure swing adsorption (PSA) unit inlet stream are obtained and sent to the PSA dynamic model solver.

[0029] In this embodiment, the first and second stage membrane separators use commercially available polyimide (PI) polymer gas separation membranes. The gas permeability of the PI membrane to each component in this embodiment is shown in Table 1.

[0030] Table 1: Gas permeability of PI membrane

[0031] In the pressure swing adsorption (PSA) dynamic model solver, the PSA unit in this embodiment is configured as a "6 towers, 12 steps" cyclic timing sequence. A complete cycle (480 s) for a single tower is divided into 12 independent steps. The specific cyclic timing table is shown in Table 2. The boundary interaction calculation logic for each step is as follows: Adsorption (AD): The gas stream from the inlet of the pressure swing adsorption unit is introduced into the bottom of the column, and the pressure is maintained at the adsorption pressure. The gas flowing out from the top of the column is collected as product gas before deducting internal consumption. Pressure reduction (ED1 / ED2 / ED3): The bottom valve is closed, and the top valve is opened sequentially, reducing the pressure in the bed layer in sequence. The transient multi-component molar flux flowing out of the top of the column is integrated over time step by step and accumulated and stored in the corresponding "virtual buffer tank models" ED1, ED2, and ED3 respectively; Forward venting (PP): The top of the column continues to release pressure in the forward direction. The discharged high-purity gas is integrated and recorded to provide a flushing source for other columns. Reverse depressurization (BD): The top of the column is closed, and the bottom of the column is depressurized in reverse to near atmospheric pressure. The desorbed gas is collected by integration. Purging (PR): The top of the column receives purging gas generated in the PP step from other columns, while the bottom of the column continues to discharge desorbed gas; Equal pressure boosting (ER3 / ER2 / ER1): The normalized composition and temperature of the mixed gas stored in buffer tanks ED3, ED2 and ED1 are extracted sequentially and used as the feed boundary conditions for reverse entry into the top of this column to achieve step-by-step pressure boosting of the bed. Final pressurization (FR): Pure hydrogen gas is introduced from the product gas header at the top of the tower to restore the bed pressure to the adsorption pressure; Standby (ID): Inlet and outlet valves are closed, and transient balance calculations of the closed system are performed.

[0032] Table 2: Adsorption Cycle Time Table

[0033] Table 2: Continued

[0034] S2: In the pressure swing adsorption dynamic model solver, the adsorption bed in this embodiment is filled axially with a double layer of adsorbent consisting of ordinary activated carbon (AC) and cuprous ion-modified activated carbon (CuCl@AC). Discrete spatial mesh (number of nodes) is then generated. m = 40), to address the issue of a sudden increase in local stiffness in the partial differential equation caused by directly and rigidly switching the isotherms of the two adsorbent layers, a Sigmoid-type logic smoothing transition function is adopted: in, To control the constant of steepness (preferred in this embodiment) k =50), This represents the interface location. Using this function, the physical property matrices (bed porosity, skeleton density, specific heat capacity, etc.) and adsorption isotherm parameters of the bottom AC and top CuCl@AC layers are nonlinearly interpolated and mapped onto the mesh along the axial direction. In the formula, This represents the calculation result of any of the above physical quantities or isotherms. This mechanism eliminates abrupt parameter changes at the interfaces between different adsorbents, obtaining a smoothly transitioned physical property matrix and the spatial distribution of adsorption isotherm parameters.

[0035] S3: In the pressure swing adsorption dynamic model solver, to avoid the pseudo-steady state caused by the false accumulation of materials in the simulation of recirculation containing tail gas, at the initial moment of the cycle ( t=0) The initial gas phase composition determined based on the inlet flow parameters of the pressure swing adsorption unit is used to initialize the state variable matrix in the adsorption bed, so that the initial adsorption amount of the solid phase reaches a local true thermodynamic equilibrium, and the initial state matrix of the pressure swing adsorption dynamic model solver is obtained.

[0036] S4: After initialization, the system enters a bidirectional nested loop between the pressure swing adsorption dynamic model solver and the steady-state process solver. The main control unit performs the following scheduling and decision-making: (a) Integral loop of inner layer operation steps: The pressure swing adsorption dynamic model solver receives the current inner layer tolerance from the master control terminal (the first loop uses the preset first inner layer tolerance) and reads the loop timing table (Table 2). Based on the geometric parameters of step S2, the physical property matrix after smooth transition, the spatial distribution of adsorption isotherm parameters, the initial state matrix of step S3, and the current timing steps, the real-time boundary conditions of the partial differential equation system are constructed. For the extreme pressure step caused by instantaneous valve switching, the pressure boundary is smoothed using continuous harmonic functions: The equations are discretized onto a spatial grid using a first-order upwind difference scheme, and the direction of convection is determined. Adaptive integration is performed using the ode15s variable-step numerical integration algorithm. After integration, the spatial physics state at the end of the current time step is recorded as the initial value for the next time step. The instantaneous molar flow rate integral in the end spatial physics state is accumulated into the virtual buffer tank model for use by other steps, and the total product gas and desorbed gas (before deducting internal consumption) are calculated simultaneously. The predicted values ​​of the tear stream parameters for a single complete cycle are output. This process is repeated until the relative residuals of the stream parameters for two consecutive complete cycles satisfy the current inner-layer tolerance. The calculated values ​​of the tear stream parameters for the last single complete cycle are output, and the process proceeds to step (b). The pressure evolution curve of the pressure swing adsorption bed is shown in [reference needed]. Figure 3 .

[0037] (b) Outer loop steady-state determination and dynamic tolerance allocation: The master control terminal calculates the relative error between the tearing flow parameters. The first relative error calculation is the relative error between the initial tearing flow parameter guess value in step S1 and the tearing flow parameter guess value in the last single complete cycle of the pressure swing adsorption dynamic model solver. The subsequent relative error calculation is the relative error between the parameters in the tearing flow parameter guess values ​​output by the pressure swing adsorption dynamic model solver in the two outer loops before and after.

[0038] When the maximum value of the relative error is greater than or equal to the set outer layer error threshold (set to 10 in this embodiment), -4 When the outer loop enters its quasi-steady-state iteration period, the master control unit sends a relatively loose first inner-loop tolerance (10) to the inner loop. -3To avoid ineffective long-time integration under unsteady conditions; the relaxation factor allocation threshold is set to 5%, and a first relaxation factor is assigned to parameters whose relative error exceeds the relaxation factor allocation threshold (first relaxation factor). For parameters whose relative error is not greater than the relaxation factor allocation threshold, a second relaxation factor is assigned (the second relaxation factor). ).implement The weighted updated predicted values ​​of the tear stream parameters, after including forced normalization of the mole fraction of gas phase components, are fed back to the steady-state process solver by the master control unit for recalculation. After obtaining the new inlet stream parameters of the pressure swing adsorption unit, the process returns to step (a) for inner-layer circulation. Figure 4 shows the convergence process of the relative error of the tear stream parameters during the quasi-steady-state iteration period.

[0039] When the relative errors of all parameters in the predicted tear flow parameters are less than the outer layer error threshold (less than 10), -4 When the outer loop reaches a certain threshold, the system determines that it has entered the steady-state iteration period. At this time, the master control unit sends the second inner-loop tolerance (10) to the inner loop. -5 The pressure swing adsorption unit inlet stream parameters output by the current steady-state process solver are locked as constant feed boundary conditions. Step (a) is continued until the inner layer cycle meets the second inner layer tolerance, achieving a strict cyclic steady state, and the inner and outer layer cycles are terminated. Figures (a) to (h) in Figure 5 show in detail the evolution trajectory of the stream parameters of the exhaust gas at the top of the tower, which monotonically and asymptotically converge to the cyclic steady state within the final 30 cycles during the steady-state iteration period.

[0040] S5: Once the system reaches a strictly stable cyclic state, the main control unit extracts the state variable matrix and parameters of each stream. When calculating the product gas flow rate, the amount of gas recirculated from the product gas main to the adsorption tower for process consumption within the pressure swing adsorption cycle is forcibly deducted. This obtains the objective and true net product gas yield and purity. Subsequently, the stream parameters of the deducted net product gas, desorbed gas, and by-products are compiled into the total stream parameters.

[0041] S6: The main control unit transmits the packetized total flow parameters, equipment parameters, and utility consumption parameters extracted from the steady-state process solver to the economic accounting module to calculate the total annual cost. TAC Given that the desorbed gas is rich in a large number of inert, non-flammable components, this embodiment introduces a discount factor to characterize the degradation of the calorific value of the by-products. Using the formula: The cost of hydrogen separation with objective physical significance was calculated.

[0042] Finally, the response surface optimization module is invoked to build a proxy model to maximize the recovery rate and minimize the impact of the response surface optimization. SCTo achieve the goal of collaborative optimization, the optimal operating parameter configuration is output. This embodiment focuses on the equipment parameters of the membrane separation unit (primary membrane area, secondary membrane area, inlet pressure, and permeate-side pressure) and the equipment parameters of the pressure swing adsorption unit (adsorption pressure, flushing ratio). P / F BBD experiments and response surface methodology were conducted using the tower diameter and AC / CuCl@AC adsorbent packing ratio. Through multi-objective collaborative optimization using a response surface surrogate model, the globally optimal process configuration was finally determined: a first-stage membrane area of ​​13433.9 m². 2 Secondary membrane area: 1084.3 m² 2 Membrane inlet pressure 2260.0 kPa, permeate side pressure 214.2 kPa, adsorption pressure 26 bar, flushing ratio P / F With an adsorbent loading ratio of 0.146, a column diameter of 2.355 m, and an AC / CuCl@AC ratio of 0.50, the system achieved a total plant recovery rate of 90.340% and a stable output of 5614.04 Nm³ under optimal parameters for membrane separation and pressure swing adsorption units. 3 / h of fuel cell-grade hydrogen with a purity of 99.9993% (CO concentration stably controlled at 0.197 ppm), and SC As low as 3.116 / kg H2.

[0043] (The above embodiments are intended to illustrate the technical implementation path of the present invention in detail. Conventional equivalent substitutions made by those skilled in the art for model parameters, software platform selection, loop timing and target setting without departing from the core idea, algorithm framework and physicochemical principles of the present invention should all fall within the protection scope of the claims of the present invention.)

Claims

1. An optimization method for pressure swing adsorption coupled hydrogen production process based on a cross-platform dynamic decoupling algorithm, characterized in that, Includes the following steps: S1: Construct a cross-platform joint simulation architecture, which integrates a master control terminal, a steady-state process solver, a pressure swing adsorption dynamic model solver, an economic accounting module, and a response surface optimization module. For a pressure swing adsorption coupling process that includes a circulating reflux stream, a break point on the reflux closed loop is set as the tear stream at the main control terminal, and the estimated values ​​of the tear stream parameters are initialized. The feed stream and equipment parameters are input into the main control terminal and transmitted to the steady-state process solver for steady-state process calculation. The pressure swing adsorption unit inlet stream parameters output by the steady-state process solver are obtained and transmitted to the pressure swing adsorption dynamic model solver. The stream parameters include the temperature, pressure, flow rate and content of each gas phase component of the corresponding stream. The guessed value of the tear stream parameter in step S1 is the initial stream parameter guessed value set for the tear stream. S2: In the pressure swing adsorption dynamic model solver, based on the inlet flow parameters of the pressure swing adsorption unit, the physical property matrix, geometric parameters, and adsorption isotherm parameters are initialized; the adsorption bed is divided into discrete spatial grids, and a smooth transition function is used to perform nonlinear interpolation mapping of the physical property matrix and adsorption isotherm parameters of different adsorbent layers along the axial direction on the discrete spatial grid to obtain the spatial distribution of the physical property matrix and adsorption isotherm parameters after smooth transition; S3: In the pressure swing adsorption dynamic model solver, initialize the state variable matrix covering the system state: At the initial moment of the cycle, use the initial gas phase composition determined according to the start-up purging condition or the inlet flow parameters of the pressure swing adsorption unit to initialize the state variable matrix and obtain the initial state matrix of the pressure swing adsorption dynamic model solver. S4: Enter the nested loop solution process between the pressure swing adsorption dynamic model solver and the steady-state process solver, and the main control terminal performs variable transfer and judgment between the above modules; S5: Once a strict cyclic steady state is reached, the main control unit extracts the state variable matrix, product gas and desorbed gas stream parameters from the pressure swing adsorption dynamic model solver, and by-product stream parameters from the steady-state process solver for processing and calculation. When calculating the product gas flow rate, the amount of gas returning from the product gas main pipe to the adsorption tower for process consumption within the pressure swing adsorption cycle is deducted to obtain the net product gas yield and purity. The stream parameters of the net product gas, desorbed gas, and by-products after deduction are then processed into total stream parameters. S6: The main control unit transmits the total flow parameters, equipment parameters, and utility consumption parameters extracted from the steady-state process solver to the economic accounting module; finally, it calls the response surface optimization module to construct a response surface proxy model, uses preset indicators as the collaborative optimization target, and outputs the optimal operation parameter configuration.

2. The method for optimizing a pressure swing adsorption coupled hydrogen production process based on a cross-platform dynamic decoupling algorithm according to claim 1, characterized in that, The smooth transition function mentioned in step S2 is a Sigmoid type logic function, and its formula is: ,in This is the transition steepness constant. It is the axial distance. This refers to the location of the interface.

3. The method for optimizing a pressure swing adsorption coupled hydrogen production process based on a cross-platform dynamic decoupling algorithm according to claim 1, characterized in that, Step S4 specifically includes: (a) Inner layer operation step integral loop: The pressure swing adsorption dynamic model solver receives the inner layer tolerance sent by the main control terminal, wherein the first calculation is preset as the first inner layer tolerance. It reads the loop timing table and constructs the real-time boundary conditions of the partial differential equation system based on the geometric parameters of step S2, the physical property matrix after smooth transition, the spatial distribution of adsorption isotherm parameters, the initial state matrix of step S3, and the current timing step. The real-time pressure boundary is smoothed by continuous harmonic function constraint. Based on the set real-time boundary conditions and the physical property matrix, geometric parameters, and adsorption isotherm parameters, the partial differential equation system and its real-time boundary conditions are discretized onto the spatial grid using a first-order upwind difference scheme and the convection term is determined. The algorithm proceeds in the forward direction, calling the variable step size numerical integration algorithm for rigid ordinary differential equations for adaptive numerical integration; after integration, the spatial physical field state at the end of the current time step is recorded as the initial value for the next time step; the instantaneous molar flow rate integral in the end spatial physical field state is accumulated into the virtual buffer tank model for use by other steps, and the total amount of product gas and desorbed gas before deducting internal consumption is calculated simultaneously; the predicted value of the tear stream parameters for a single complete cycle is output; this process is repeated until the relative residuals of all tear stream parameters in the two complete cycles before and after the pressure swing adsorption dynamic model solver meet the current inner layer tolerance, the predicted value of the tear stream parameters for the last single complete cycle is output, and the process proceeds to step (b). (b) Outer loop steady state determination and dynamic tolerance allocation: The first relative error calculation is to calculate the relative error of the parameters in the initial guess value of the tearing flow parameters in step S1 and the guess value of the tearing flow parameters in the last single complete cycle of the pressure swing adsorption dynamic model solver. The subsequent relative error calculation is to calculate the relative error of the parameters in the guess value of the tearing flow parameters output by the pressure swing adsorption dynamic model solver in the two outer loops before and after. When the relative error of any parameter in the predicted parameters of the tearing flow stream is greater than or equal to the outer layer error threshold, the quasi-steady-state iteration period of the outer layer loop is entered, and the first inner layer tolerance is issued to the inner layer loop; a first relaxation factor is assigned to the parameters in the predicted parameters of the tearing flow stream that have a relative error greater than the relaxation factor allocation threshold, and a second relaxation factor is assigned to the parameters whose relative error is not greater than the relaxation factor allocation threshold, and the first relaxation factor is greater than the second relaxation factor; the predicted parameters of the tearing flow stream are updated by weighting using the allocated relaxation factors, and fed back to the steady-state process solver by the main control end for calculation, and after re-acquiring the pressure swing adsorption unit inlet flow stream parameters output by the steady-state process solver, the process returns to step (a) to perform the inner layer loop; When the relative errors of all parameters in the predicted tear flow parameters are less than the outer layer error threshold, the outer layer loop enters the steady-state iteration period, and a second inner layer tolerance less than the first inner layer tolerance is issued to the inner layer loop. The pressure swing adsorption unit inlet flow parameters output by the current steady-state process solver are locked as the constant feed flow parameters of the pressure swing adsorption dynamic model solver. Step (a) continues to run until the inner layer loop meets the second inner layer tolerance, reaches a strict cyclic steady state, and ends the inner and outer layer loops.

4. The method for optimizing a pressure swing adsorption coupled hydrogen production process based on a cross-platform dynamic decoupling algorithm according to claim 3, characterized in that, The partial differential equations set mentioned in step S4(a) include the overall mass continuity equation, the overall energy balance equation of the mixed bed, the transient heat conduction equation of the tower wall, and the mass transfer kinetic equation; and in numerical calculation, they are solved together with the local ideal gas equation of state as algebraic constraints.

5. The method for optimizing a pressure swing adsorption coupled hydrogen production process based on a cross-platform dynamic decoupling algorithm according to claim 3, characterized in that, The continuous harmonic function of the real-time pressure boundary mentioned in step S4(a) is: ,in and These are the initial and target pressures, respectively. For normalized dimensionless time.

6. The method for optimizing a pressure swing adsorption coupled hydrogen production process based on a cross-platform dynamic decoupling algorithm according to claim 3, characterized in that, In step S4(b), the order of magnitude of the outer layer error threshold is set to 10 -4 up to 10 -3 , the first inner layer tolerance is greater than the second inner layer tolerance; the relaxation factor assignment threshold is less than 15%, the first relaxation factor is set to 0.60 to 0.99, and the second relaxation factor is set to 0.10 to 0.60; The estimated values ​​of the tear flow parameters, which are then updated using a relaxation factor weighting method, are calculated using the following formula: ,in, These are the guessed values ​​for the updated tearing flow parameters. The first relaxation factor or the second relaxation factor is allocated based on the relaxation factor allocation threshold. These are the estimated values ​​of the tearing flow parameters output from a single complete cycle of the pressure swing adsorption dynamic model solver. The parameters of the tear stream in the previous outer cycle are estimated; after updating the estimated parameters of the tear stream, a forced normalization step for the mole fraction of the gas phase components is also included.

7. The method for optimizing a pressure swing adsorption coupled hydrogen production process based on a cross-platform dynamic decoupling algorithm according to claim 1, characterized in that, In step S6, the formula is used. Calculate the cost of hydrogen separation SC ;in, The total annual cost is the sum of the annualized depreciation of the equipment, the annual cost of raw gas, and the annual operating expenses. For net product hydrogen production, The by-product calorific value revenue is calculated based on the benchmark gas price. The discount factor is used to characterize the deterioration of the calorific value of by-products.

8. An optimization system for pressure swing adsorption coupled hydrogen production process based on a cross-platform dynamic decoupling algorithm, characterized in that, The system, applicable to the method of any one of claims 1 to 7, comprises: The main control unit is configured to perform cross-platform data integration, parameter distribution, scheduling system timing calculations, and multi-objective collaborative optimization. The steady-state process solver is configured to receive the predicted values ​​of the tear stream, perform macroscopic steady-state process simulation calculations, and output the inlet stream parameters, by-product stream parameters, and utility consumption of the pressure swing adsorption unit. The dynamic model solver for pressure swing adsorption is embedded with a linear method spatial discrete grid and a rigid ordinary differential equation variable step size numerical integrator, configured to perform inner and outer double-layer nested loop dynamic solution to extract spatiotemporal physical field and flow characteristic information. The economic accounting module is configured to receive steady-state flow parameters and utility consumption, and calculate and output technical and economic evaluation indicators. The response surface optimization module is configured to build a proxy model based on multi-condition data and perform multi-objective collaborative optimization.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 7.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method according to any one of claims 1 to 7.

Images