Real-time operation optimization method and system for multi-energy complementary electrolytic hydrogen production system

Abstract

The invention discloses a real-time operation optimization method and system for a multi-energy complementary electrolytic hydrogen production system, and the method comprises the steps: collecting real-time parameters of each module of the system, and inputting structured process knowledge to guide a neural network to predict a parameter change trend; based on a prediction result, constructing an optimization main model by using a distributed robust opportunity constraint programming algorithm, decomposing sub-models through a multi-main-body collaborative optimization algorithm based on ADMM, and solving the sub-models to obtain preliminary optimization parameters; and the photothermal effect is combined to induce the water phase regulation and control platform to regulate the medium state and the system operation parameters, closed-loop regulation is formed, and the final optimization parameters are output after the parameter deviation reaches the standard through iteration. The system comprises a multi-parameter collaborative acquisition unit, a structured knowledge driven prediction unit, a distributed robust optimization modeling unit, a multi-subject collaborative solving unit, a closed-loop parameter adjustment unit and an iterative optimization output unit, and all the units collaboratively realize real-time optimization of the system, so that the operation stability and economy are improved.

Description

Technical Field

[0001] This invention relates to the field of hydrogen production system operation optimization technology, and in particular to a real-time operation optimization method and system for multi-energy complementary electrolysis hydrogen production systems. Background Technology

[0002] In the current process of transitioning the energy structure towards clean and low-carbon energy, multi-energy complementary electrolysis hydrogen production systems, with their ability to integrate and utilize renewable energy sources such as solar and wind power, have become a key technological direction for achieving large-scale green hydrogen production. Such systems need to simultaneously coordinate supply fluctuations from multiple energy input modules, operating condition changes in the electrolysis hydrogen production module, and the state control of the aqueous medium to ensure operational stability and economic efficiency. However, during system operation, parameters such as energy supply and hydrogen production load demand undergo real-time dynamic changes, and the photothermal effect-induced state of the aqueous medium in the control platform further affects electrolysis efficiency. Therefore, achieving real-time collaborative optimization of multiple parameters and modules has become a core requirement for improving the overall system performance, and has also driven the research and development of related optimization methods and systems.

[0003] Existing technologies for real-time optimization of multi-energy complementary electrolysis hydrogen production systems suffer from two significant drawbacks. Firstly, most optimization methods fail to effectively integrate structured process knowledge and data-driven models, relying solely on single algorithms for parameter prediction or optimization. This makes it difficult to accurately capture the correlation between multi-energy conversion rules and the kinetic characteristics of the electrolysis hydrogen production reaction, leading to discrepancies between predicted results and actual system operational requirements, and insufficient adaptability of optimized parameters. Secondly, existing systems lack efficient multi-entity collaborative control mechanisms. When dealing with constrained coupling problems between multiple subsystems, inconsistencies often arise between subsystem optimization objectives and overall system objectives. Furthermore, they fail to establish a complete optimization process encompassing parameter prediction, model building, collaborative solving, and closed-loop regulation, hindering dynamic real-time adjustment of operating parameters for each module and impacting system stability and economic efficiency. Summary of the Invention

[0004] In order to overcome the shortcomings and deficiencies of existing technologies, this invention provides a real-time operation optimization method and system for a multi-energy complementary electrolysis hydrogen production system.

[0005] The technical solution adopted in this invention is a real-time operation optimization method for a multi-energy complementary electrolysis hydrogen production system, comprising: Step S1: collecting real-time operating parameters of each energy input module, operating condition parameters of the electrolysis hydrogen production module, and medium state parameters of the photothermal effect-induced aqueous phase control platform in the multi-energy complementary electrolysis hydrogen production system to construct a real-time operating parameter set for the system; Step S2: inputting the real-time operating parameter set of the system into a structured process knowledge-guided neural network, and through the multi-energy complementary energy conversion rules and electrolysis hydrogen production reaction kinetics knowledge embedded in the network, predicting the energy supply, electrolysis hydrogen production load demand, and aqueous phase medium characteristic change trends of the system within a future preset time period, and outputting a predicted parameter sequence; Step S3: based on the predicted parameter sequence, using a distributed bar chance constraint programming algorithm, with the goal of minimizing system operating costs, and combining the multi-energy complementary energy supply and demand fluctuation range and electrolysis hydrogen production process constraints, establishing a main model for system operation optimization, and determining the fluctuation of uncertainty parameters in the model. The scope and constraints satisfy the probability threshold; Step S4: Decompose the main model of system operation optimization into multiple sub-models. Each sub-model corresponds to a subsystem in the multi-energy complementary electrolysis hydrogen production system. Adopt the multi-agent collaborative optimization algorithm based on ADMM, set the collaborative optimization parameters and iterative convergence criteria between each sub-model, and solve the preliminary operation optimization parameters of each subsystem through alternating iterative calculation and information interaction of each sub-model; Step S5: Input the preliminary operation optimization parameters into the control module of the photothermal effect-induced aqueous phase control platform, adjust the flow rate, temperature and concentration of the aqueous phase medium in the platform according to the parameters, and simultaneously feed back the control of the electrode voltage, current density and output power of each energy input module of the electrolysis hydrogen production module to form a closed-loop parameter regulation; Step S6: Repeat steps S2 to S5, perform deviation analysis on the system operation parameters obtained in each iteration, and when the parameter deviation of multiple consecutive iterations is within the preset range, output the final real-time operation optimization parameters of the multi-energy complementary electrolysis hydrogen production system;

[0006] Furthermore, the prediction output expression of the structured process knowledge-guided neural network is: y pred =σ(W3·tanh(W2·(W1x+b1)+K·f proc (x)+b2)+b3), where y pred The system is defined as a sequence of predicted parameters for a pre-defined time period, including predicted energy supply, predicted demand for hydrogen production via electrolysis, and predicted properties of the aqueous medium; x is the set of real-time operating parameters input from the system; W1, W2, and W3 are the weight matrices of each hidden layer and output layer of the neural network, respectively; b1, b2, and b3 are the bias vectors of each hidden layer and output layer of the neural network, respectively; σ(·) is the activation function of the output layer; tanh(·) is the activation function of the hidden layer; K is the weight coefficient of the structured process knowledge; f proc(x) is the process function corresponding to the embedded multi-energy complementary energy conversion rule and the knowledge of the electrolytic hydrogen production reaction kinetics.

[0007] Furthermore, the main model expression for system operation optimization constructed by the bibliometric chance-constrained programming algorithm is as follows:

[0008]

[0009] Where u represents the system operation optimization decision variable, including the output power of each energy input module, the operating parameters of the electrolysis hydrogen production module, and the control parameters of the photothermal effect-induced aqueous phase regulation platform; U represents the feasible region of the decision variable; c represents the cost coefficient vector of the decision variable; c T u is the objective function for system operating cost; This represents a family of distributions for the uncertainty parameter ξ, which includes fluctuations in energy supply, fluctuations in hydrogen production load due to electrolysis, and fluctuations in the properties of the aqueous medium; g i (u, ξ) are the system operation constraint functions for the i-th system; ∈ i Let be the threshold for the allowed violation probability of the i-th constraint; n is the total number of constraints; It is a probability measure.

[0010] Furthermore, the iterative update expression of the ADMM-based multi-agent collaborative optimization algorithm is:

[0011]

[0012] Among them, u k U is the decision variable for the k-th subsystem; k Let be the feasible region of the decision variable for the k-th subsystem; m be the number of subsystems; t be the number of iterations; λ be the Lagrange multiplier vector; ρ be the penalty parameter; L ρ (·) represents the augmented Lagrangian function; A k Let be the correlation matrix of the decision variables of the k-th subsystem; b is the right-hand vector of the inter-subsystem collaborative constraint equation. Let λ be the value of the decision variable in the (t+1)th iteration of the k-th subsystem; t+1 Let be the Lagrange multiplier value for the (t+1)th iteration.

[0013] Furthermore, the medium state regulation model expression for the photothermal effect-induced aqueous phase control platform is as follows:

[0014]

[0015] Among them, T t+1 T t These represent the temperatures of the aqueous medium within the platform at time t+1 and time t, respectively; α is the photothermal conversion efficiency coefficient; Q opt The light energy input to the photothermal module; β is the heat dissipation coefficient of the medium; Tenv γ represents ambient temperature; γ represents the heat transfer coefficient of the medium flow rate; q flow T is the medium circulation flow rate; in To compensate for the temperature of the medium; C t+1 C t δ represents the concentration of the aqueous medium within the platform at time t+1 and time t, respectively; C represents the mixing coefficient of the supplementary medium; add To replenish the medium concentration; q add To supplement the medium flow rate; q total ∈ represents the total flow rate of the medium within the platform; ∈ represents the reaction consumption coefficient; r react Δt represents the rate at which the medium participates in the electrolysis-assisted reaction; Δt represents the time step.

[0016] Furthermore, the correlation model expression between the total energy consumption and hydrogen production capacity of the multi-energy complementary electrolysis hydrogen production system is as follows:

[0017]

[0018] Among them, H prod η represents the total hydrogen production within the preset time period [t0, t1]. total The overall hydrogen production efficiency of the system is represented by p; the number of energy input modules is represented by P. j (t) represents the output power of the j-th energy input module at time t; η j P represents the energy transfer efficiency from the j-th energy input module to the electrolysis hydrogen production module. loss (t) represents the total energy loss of the system at time t, including line loss, equipment loss, and energy consumption of the water phase control platform induced by photothermal effect.

[0019] Further, step S3 includes the following sub-steps: S31: Extract the predicted values ​​of multi-energy complementary energy supply, electrolysis hydrogen production load demand, and aqueous medium characteristics from the predicted parameter sequence output in step S2. Use these predicted values ​​as basic data to divide the fluctuation range of each parameter within a preset time period in the future, and determine the probability distribution type of each fluctuation range; S32: Based on the operating requirements of the multi-energy complementary electrolysis hydrogen production system, clarify the hard constraints and soft constraints in the system operation process. Hard constraints include the maximum current density limit of the electrolysis hydrogen production module and the upper limit of the output power of each energy input module. Soft constraints include the fluctuation range of system operating costs. S33: Determine the probability threshold for satisfying each constraint under the sub-Bruker framework; S34: With minimizing the system operating cost as the objective function, integrate the parameter fluctuation range and probability distribution type determined in step S31, the constraint conditions and probability thresholds set in step S32 into the sub-Bruker chance constraint programming model framework, construct the main system operation optimization model including uncertainty parameters, and define the value range and correlation of each variable in the model; S35: Perform a feasibility analysis on the constructed main system operation optimization model, and ensure that the model has a feasible solution under the preset system operation scenario by adjusting the fluctuation range of uncertainty parameters and the constraint satisfaction probability thresholds, thus forming the final main system operation optimization model.

[0020] Further, step S4 includes the following sub-steps: S41: Based on the structural composition of the multi-energy complementary electrolysis hydrogen production system, the main model for system operation optimization is decomposed into corresponding sub-models according to the energy input subsystem, the electrolysis hydrogen production subsystem, and the photothermal effect-induced aqueous phase control subsystem, and the decision variables, objective function, and local constraints of each sub-model are defined; S42: The initial parameters of the multi-agent collaborative optimization algorithm based on ADMM are set, including the initial values ​​of the decision variables of each sub-model, the initial values ​​of the Lagrange multipliers, and the values ​​of the penalty parameters. At the same time, the iterative convergence criteria of the algorithm are determined, which include the iterative deviation threshold of the decision variables and the constraint satisfaction deviation threshold. S43: Each sub-model solves its own objective function based on the initial parameters and the boundary parameters passed from other sub-models, obtaining its own preliminary decision variable values. These preliminary decision variable values ​​are then passed to the collaborative optimization center, which calculates the constraint deviations between the sub-models. S44: Based on the constraint deviations calculated by the collaborative optimization center, the Lagrange multipliers are updated. Each sub-model then resolves its objective function based on the updated Lagrange multipliers and penalty parameters, obtaining new decision variable values. The parameter passing, constraint deviation calculation, and variable update process is repeated until the iterative convergence criterion is met, and the preliminary operating optimization parameters of each subsystem are output.

[0021] Further, step S5 includes the following sub-steps: S51: Classify and organize the preliminary operation optimization parameters of each subsystem output in step S4, and screen out the control parameters related to the photothermal effect-induced aqueous phase control platform, including medium flow control parameters, temperature adjustment parameters, and concentration adjustment parameters. Simultaneously, extract the electrode voltage and current density parameters of the electrolysis hydrogen production module and the output power parameters of each energy input module; S52: Input the screened photothermal effect-induced aqueous phase control platform control parameters to the platform's actuator. The actuator adjusts the speed of the medium circulation pump to change the medium flow rate, adjusts the heating power of the photothermal module to change the medium temperature, and adjusts the opening of the replenishing medium valve according to the parameter instructions. S53: Change the medium concentration; Send the extracted electrode voltage and current density parameters of the electrolysis hydrogen production module to the controller of the electrolysis hydrogen production module. The controller adjusts the voltage and current output of the electrode power supply circuit to make the operating parameters of the electrolysis hydrogen production module reach the preliminary optimized value. At the same time, the output power parameters of each energy input module are transmitted to the corresponding energy control module to adjust the energy output; S54: Collect the medium state parameters after adjustment by the photothermal effect induced aqueous phase control platform, the actual operating parameters of the electrolysis hydrogen production module, and the actual output parameters of each energy input module. Compare these parameters with the preliminary operating optimization parameters, calculate the parameter deviation, and feed the deviation information back to step S4 for subsequent parameter iterative adjustment.

[0022] The real-time operation optimization system for a multi-energy complementary electrolysis hydrogen production system includes: a multi-parameter collaborative acquisition unit, which is connected to the energy input modules, electrolysis hydrogen production modules, and photothermal effect-induced aqueous phase control platform of the multi-energy complementary electrolysis hydrogen production system, and is used to acquire the real-time operating parameters of each module and construct a real-time operating parameter set for the system; a structured knowledge-driven prediction unit, which is connected to the multi-parameter collaborative acquisition unit, has a built-in structured process knowledge-guided neural network, receives the real-time operating parameter set output by the multi-parameter collaborative acquisition unit, predicts the parameter change trend of the system within a preset time period, and outputs a predicted parameter sequence; and a sub-Bruker bar optimization modeling unit, which is connected to the structured knowledge-driven prediction unit, uses the sub-Bruker bar chance-constrained programming algorithm to construct a main model for system operation optimization based on the predicted parameter sequence, and determines the fluctuation range of uncertain parameters in the model. The system comprises: a probability threshold for constraint satisfaction; a multi-agent collaborative solution unit connected to the sub-Bruker bar optimization modeling unit, which decomposes the main system operation optimization model into sub-models, solves the sub-models using an ADMM-based multi-agent collaborative optimization algorithm, and outputs the preliminary operation optimization parameters for each subsystem; a closed-loop parameter adjustment unit connected to the multi-agent collaborative solution unit, the electrolysis hydrogen production module, and the photothermal effect-induced aqueous phase control platform, which adjusts the operation parameters of the electrolysis hydrogen production module and the photothermal effect-induced aqueous phase control platform according to the preliminary operation optimization parameters and feeds back the parameter deviations; and an iterative optimization output unit connected to the closed-loop parameter adjustment unit, which receives parameter deviation information, controls the repeated operation of the structured knowledge-driven prediction unit, the sub-Bruker bar optimization modeling unit, the multi-agent collaborative solution unit, and the closed-loop parameter adjustment unit, and outputs the final optimization parameters when the parameter deviations meet the requirements.

[0023] Beneficial Effects: This invention proposes a real-time operation optimization method and system for a multi-energy complementary electrolysis hydrogen production system. It employs a structured process knowledge-guided neural network, embedding multi-energy complementary energy conversion rules and electrolysis hydrogen production reaction kinetics into the model. This effectively integrates structured process knowledge with a data-driven model, avoiding the limitations of single algorithms, accurately capturing the correlation between parameters, improving the matching degree between prediction results and actual operational needs, and solving the problem of insufficient adaptability of optimization parameters in existing technologies. The system operation optimization master model is constructed using a split-Browser chance-constrained programming algorithm, and then combined with an ADMM-based multi-agent collaborative optimization algorithm to decompose and collaboratively solve the master model into sub-models. This, combined with the multi-agent collaborative solution unit, establishes a high-performance... A multi-entity collaborative control mechanism is established to coordinate the optimization objectives of each subsystem with the overall system, avoiding the problem of incoordination between subsystems and overall objectives in constrained coupling scenarios. Through the photothermal effect-induced closed-loop parameter adjustment of the aqueous phase control platform, the electrolysis hydrogen production module, and the energy input module, combined with the iterative optimization output unit, a complete optimization process of "parameter acquisition-prediction-modeling-solving-adjustment-iteration" is formed, realizing the dynamic real-time adjustment of the operating parameters of each module. At the same time, the photothermal effect-induced precise adjustment of the aqueous phase medium state by the aqueous phase control platform further ensures electrolysis efficiency, improves the overall system operation stability and economy, and comprehensively makes up for the deficiencies of existing technologies in optimization accuracy, coordination, and process integrity. Attached Figure Description

[0024] Figure 1 This is a diagram illustrating the method steps of the present invention;

[0025] Figure 2 This is a diagram showing the system unit composition of the present invention. Detailed Implementation

[0026] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] like Figure 1 As shown, the real-time operation optimization method for a multi-energy complementary electrolysis hydrogen production system includes:

[0028] Step S1: Collect the real-time operating parameters of each energy input module in the multi-energy complementary electrolysis hydrogen production system, the operating condition parameters of the electrolysis hydrogen production module, and the medium state parameters of the photothermal effect-induced aqueous phase control platform to construct a set of real-time operating parameters for the system.

[0029] Specifically, step S1 first determines the core modules of the multi-energy complementary electrolysis hydrogen production system, including energy input modules such as solar photovoltaic modules, wind power generation modules, and grid-supplemented energy modules; the electrolysis hydrogen production module composed of electrolyzers; and the photothermal effect-induced aqueous phase control platform composed of a water storage tank, heating device, circulating pump, and concentration sensor. During the data acquisition process, for the energy input modules, the real-time irradiance intensity of the solar photovoltaic modules needs to be obtained (typically ranging from 200-1000 W / m²). 2 The system needs to collect the following data: output power (range 0-500kW), real-time wind speed (range 3-25m / s) and output power (range 0-800kW) of the wind power generation module; real-time supply voltage (range 380-400V) and current (range 0-1200A) of the grid-supplied power module; and for the electrolysis hydrogen production module, real-time electrode temperature (range 60-85℃), electrode voltage (range 1.8-2.2V), and current density (range 1000-3000A / m²) of the electrolyzer. 2 Hydrogen production (range 0-50 Nm³) 3 / h); For the photothermal effect-induced aqueous phase control platform, it is necessary to collect real-time temperature (range 50-75℃), concentration (range 20%-30%), and circulation flow rate (range 5-15m³) of the aqueous medium. 3 The data acquisition frequency is set to once per minute. Data is acquired through sensors deployed in various modules (such as radiation sensors, wind speed sensors, voltage sensors, temperature sensors, concentration sensors, and flow sensors), and transmitted to the central control system via a data acquisition card. The system integrates and deduplicates the data, ultimately constructing a real-time system operating parameter set including more than 30 parameters. This provides complete and accurate basic data support for subsequent prediction and optimization. This step must ensure that the sensor sampling error does not exceed ±2% and the data transmission delay does not exceed 100ms to guarantee the real-time performance and reliability of the parameter set.

[0030] Step S2: Input the real-time operating parameter set of the system into the structured process knowledge-guided neural network. Through the multi-energy complementary energy conversion rules and electrolysis hydrogen production reaction kinetics knowledge embedded in the network, predict the energy supply, electrolysis hydrogen production load demand and water phase medium characteristic change trend of the system in the future preset time period, and output the prediction parameter sequence.

[0031] Specifically, in step S2, the system's real-time operating parameter set constructed in S1 is first standardized to ensure that all parameters are input into the structured process knowledge-guided neural network in numerical form. This network adopts a three-layer hidden layer structure. The number of input layer nodes is consistent with the number of parameter items in the system's real-time operating parameter set (e.g., 30 parameters correspond to 30 input nodes). The number of output layer nodes is 6, corresponding to 6 predicted parameters for the next hour (the preset time period is set to 1 hour, and the time interval is divided into 12 5-minute intervals): solar photovoltaic module supply, wind power generation module supply, grid supplementary energy module supply, electrolysis hydrogen production load demand, aqueous medium temperature change, and aqueous medium concentration change. The multi-energy complementary energy conversion rules embedded in the network include the linear correlation rule between solar irradiance and photovoltaic output power, the cubic correlation rule between wind speed and wind power output power, and the correlation rule between grid power supply efficiency and voltage stability. The electrolysis hydrogen production reaction kinetics knowledge includes the positive correlation between electrode temperature and reaction rate, the linear relationship between current density and hydrogen production, and the correlation between aqueous medium concentration and ionic conductivity. The network training process employs the gradient descent algorithm, using historical system data from the past three months (with a sample size of no less than 100,000 sets). The training objective is to ensure that the root mean square error of the prediction error does not exceed 5%. During implementation, the standardized real-time parameter set is input into the trained network. The network performs forward propagation calculations, first passing parameters from the input layer to the first hidden layer, then processing them through an activation function before passing them to the second hidden layer. Simultaneously, structured process knowledge is incorporated to correct intermediate parameters, which are then passed to the third hidden layer for further processing. Finally, the output layer outputs a sequence of predicted parameters, including six parameters across 12 time periods. This step requires periodic (every 7 days) retraining of the network to update knowledge rules and weights, ensuring prediction accuracy. When the predicted energy supply value for a given time period fluctuates by more than ±10%, it is marked as a high-risk fluctuation period, providing a key focus for subsequent optimization modeling. The entire prediction process must be completed within 10 seconds to meet real-time optimization requirements.

[0032] Step S3: Based on the predicted parameter sequence, the split-bar chance-constrained programming algorithm is adopted. With the goal of minimizing the system operating cost, the main model for system operation optimization is established by combining the fluctuation range of multi-energy complementary energy supply and demand and the constraints of the electrolysis hydrogen production process. The fluctuation range of the uncertainty parameters in the model and the probability threshold of constraint satisfaction are determined.

[0033] Specifically, step S3 requires the predicted parameter sequence output from S2 as its core basis. First, the fluctuation ranges of various energy supplies (e.g., solar photovoltaic supply fluctuation range of 50-150kW, wind power supply fluctuation range of 100-300kW), the fluctuation range of hydrogen electrolysis load demand (e.g., 200-400kW), and the fluctuation range of aqueous medium characteristics (e.g., temperature fluctuation range ±3℃, concentration fluctuation range ±2%) are extracted from the predicted parameter sequence. These ranges are then determined as the fluctuation ranges of uncertainty parameters in the main model for system operation optimization. Next, the objective function for system operation cost is set. The cost components include grid power purchase cost (unit price set at 0.5 yuan / kWh), equipment operation and maintenance cost (photovoltaic module operation and maintenance cost 0.02 yuan / kWh, wind power module 0.03 yuan / kWh, electrolyzer 0.05 yuan / kWh), and solar thermal regulation cost (0.04 yuan / kWh). The objective function is to minimize the total operating cost. Then, the constraints of the electrolytic hydrogen production process are defined, with hard constraints including a maximum current density in the electrolyzer not exceeding 3500 A / m. 2 Minimum current density not less than 800 A / m 2 The maximum output power of each energy input module shall not exceed its rated power (500kW for photovoltaic, 800kW for wind power, and 600kW for grid power). The temperature of the aqueous medium shall not exceed 80℃ and not be lower than 45℃, and the concentration shall not exceed 35% and not be lower than 15%. Soft constraints include daily fluctuations in system operating costs not exceeding ±8%. When constructing the model using the distributed bar chance-constrained programming algorithm, the distribution family of uncertainty parameters is set as an interval distribution, and the constraint satisfaction probability threshold is set to 95% (i.e., ensuring that the constraint conditions are met in more than 95% of operating scenarios). During implementation, the objective function and constraint conditions are first transformed into a mathematical model framework through the algorithm, and then the fluctuation range of uncertainty parameters and the probability threshold are substituted to determine the coefficients and boundary values ​​of each variable in the model. For example, the grid power purchase cost coefficient is set to 0.5, and the photovoltaic operation and maintenance cost coefficient is set to 0.02. The model is then initially constructed, and its feasibility is verified by Monte Carlo simulation (at least 1000 simulations). If the probability of constraint satisfaction is less than 95% in the simulation results, the range of uncertainty parameter fluctuations (e.g., reducing the range of wind power supply fluctuations to 80-280kW) or the constraint satisfaction probability threshold (which can be adjusted up to 92%) is adjusted until the model meets the feasibility requirements in the simulation scenario. Finally, a main system operation optimization model is formed, including an objective function, more than 20 constraints, and more than 15 decision variables. The implementation time of this step must be controlled within 30 seconds to ensure the progress of the real-time optimization process.

[0034] Step S4: Decompose the main model of system operation optimization into multiple sub-models. Each sub-model corresponds to a subsystem in the multi-energy complementary electrolysis hydrogen production system. Adopt the multi-agent collaborative optimization algorithm based on ADMM, set the collaborative optimization parameters and iterative convergence criteria among the sub-models, and obtain the preliminary operation optimization parameters of each subsystem through alternating iterative calculation and information interaction of each sub-model.

[0035] Specifically, in step S4, based on the physical structure and functional division of the multi-energy complementary electrolysis hydrogen production system, the main system operation optimization model constructed in S3 is decomposed into three sub-models, corresponding to the energy input subsystem (including photovoltaic, wind power, and grid modules), the electrolysis hydrogen production subsystem (including electrolyzer groups), and the photothermal effect-induced aqueous phase control subsystem (including water storage tanks, heating devices, etc.). The decision variables for each sub-model are: photovoltaic output power allocation, wind power output power allocation, and grid input power allocation for the energy input sub-model; electrolyzer current density setting and electrode temperature setting for the electrolysis hydrogen production sub-model; and medium circulation flow rate setting, heating power setting, and supplementary medium flow rate setting for the photothermal control sub-model. The objective function for each sub-model is to minimize the operating cost of the corresponding subsystem, and the local constraints are the hardware limitations of each subsystem (such as the power upper limit constraint for the energy input sub-model and the current density constraint for the electrolysis hydrogen production sub-model). When using the ADMM-based multi-agent collaborative optimization algorithm, the initial parameters are set as follows: the initial values ​​of the decision variables of each sub-model are the current actual operating parameter values, the initial value of the Lagrange multiplier is 0.1, the penalty parameter is 10, and the iterative convergence criterion is set as follows: the absolute value of the deviation of all decision variables between two iterations is less than 0.01 (i.e. 1%) and the constraint deviation between sub-models is less than 0.05. During implementation, the collaborative optimization center first distributes initial parameters to the three sub-models. Each sub-model solves the objective function based on the initial parameters and its own local constraints, obtaining the first round of preliminary decision variable values ​​(e.g., the energy input sub-model outputs 100kW photovoltaic, 200kW wind power, and 150kW grid power), which are then uploaded to the collaborative optimization center. The center calculates the constraint deviations between the sub-models (e.g., the deviation between the total energy input power and the load demand of electrolysis hydrogen production), and updates the Lagrange multipliers based on the deviations (the update formula is the current multiplier plus the product of the penalty parameter and the deviation). The updated multipliers are then distributed to each sub-model, which re-solves the objective function to obtain the second round of decision variable values. This process is repeated (the number of iterations is usually 5-10 times) until the convergence criterion is met, and the preliminary operating optimization parameters of each subsystem are output (e.g., the energy input subsystem outputs 120kW photovoltaic, 220kW wind power, and 130kW grid power, and the electrolysis hydrogen production subsystem outputs a current density of 2500A / m). 2 The flow rate of the photothermal control subsystem is 10m³. 3This step requires ensuring that the solution time for each sub-model does not exceed 5 seconds and the overall iteration process does not exceed 60 seconds, in order to meet the requirements of real-time optimization.

[0036] Step S5: Input the preliminary operation optimization parameters into the control module of the photothermal effect induced aqueous phase regulation platform, adjust the flow rate, temperature and concentration of the aqueous medium in the platform according to the parameters, and at the same time, feed back and regulate the electrode voltage, current density and output power of each energy input module of the electrolysis hydrogen production module to form a closed-loop parameter regulation;

[0037] Specifically, step S5 is implemented based on the preliminary operation optimization parameters output by S4. First, the preliminary parameters are classified and the control parameters of the photothermal effect-induced aqueous phase control platform (such as the medium circulation flow rate of 10m³) are selected. 3 / h, heating power 80kW, replenishing medium flow rate 1m 3 / h, replenishing medium concentration 25%), and control parameters of the electrolysis hydrogen production module (e.g., electrolyzer current density 2500A / m³). 2 The control parameters for the energy input module (e.g., photovoltaic output power 120kW, wind power output power 220kW, grid input power 130kW) are set as follows: For the solar thermal control platform, these control parameters are sent to the platform's PLC controller. The controller then drives the circulating pump to adjust its speed (range 1000-3000rpm, 10m). 3 / h flow rate corresponds to 2000rpm) to change the medium circulation flow rate, drive the heating device to adjust the heating tube power (80kW corresponds to 4 heating tubes being turned on, with a single tube power of 20kW) to control the medium temperature, and drive the replenishment medium valve to adjust the opening degree (1m 3 The flow rate (corresponding to a valve opening of 30%) is adjusted using a concentration sensor to maintain the replenishing medium concentration at 25%. During adjustment, the medium temperature and concentration are monitored in real-time (sampling frequency 10 seconds / time) to ensure the temperature remains stable at 75℃±1℃ and the concentration at 25%±0.5%. For the electrolysis hydrogen production module, current density and temperature parameters are sent to the electrolyzer controller, which adjusts the rectifier output current (2500A / m³). 2 The corresponding total current is 5000A, and the electrolytic cell area is 2m². 2To control the current density, the electrode heating element power (set to 5kW) is adjusted to maintain the electrode temperature at 75℃. During adjustment, hydrogen production is monitored in real-time (data collected every minute) to ensure it meets optimized expectations. For the energy input module, power parameters are sent to the photovoltaic inverter, wind power converter, and grid connection controller. The inverter adjusts the photovoltaic output power to 120kW (achieved by adjusting MPPT tracking accuracy), the converter adjusts the wind power output power to 220kW (achieved by adjusting the pitch angle), and the grid controller adjusts the input power to 130kW (achieved by adjusting the circuit breaker opening degree). Meanwhile, sensors deployed in each module collect the actual operating parameters after adjustment in real time (such as the actual temperature of the photothermal platform medium, the actual current density of the electrolytic cell, and the actual output power of each energy module). The actual parameters are compared with the preliminary operation optimization parameters, and the deviation value is calculated (such as temperature deviation = actual temperature - 75℃). The deviation data is fed back to the collaborative optimization center of S4 through the data transmission module for parameter correction in the next round of iterative optimization. The adjustment response time of this step must be less than 2 seconds, and the parameter stabilization time must be less than 30 seconds to ensure the timeliness and effectiveness of closed-loop adjustment.

[0038] Step S6: Repeat steps S2 to S5, perform deviation analysis on the system operating parameters obtained in each iteration, and when the parameter deviation of multiple consecutive iterations is within the preset range, output the final real-time operating optimization parameters of the multi-energy complementary electrolysis hydrogen production system.

[0039] Specifically, when implementing step S6, the control rules for iterative operation are first set. The iteration trigger condition is that the absolute value of any parameter deviation in the parameter deviations fed back in S5 exceeds a preset threshold (such as temperature deviation threshold ±2℃, current density deviation threshold ±50A / m). 2The power deviation threshold is ±10kW. The iteration termination condition is that the absolute value of all parameter deviations in three consecutive iterations is less than the preset threshold, and the interval between each iteration is set to 5 minutes (consistent with the time interval predicted in S2). During implementation, the first iteration is executed again from S2, that is, a new real-time operating parameter set of the system is constructed based on the actual operating parameters fed back from S5 (replacing the parameter set of the initial S1). The structured process knowledge is input to guide the neural network to make a new round of predictions, and the updated prediction parameter sequence is obtained. Then, S3 is executed, and the fluctuation range and constraints of the uncertainty parameters are adjusted based on the new prediction parameter sequence to update the main model of system operation optimization. Then, S4 is executed, and the sub-models are decomposed using the updated main model and solved again to obtain the corrected preliminary operation optimization parameters. Then, S5 is executed, and closed-loop adjustment is performed based on the corrected parameters and new parameter deviations are fed back. Repeat steps S2 to S5 above, recording all parameter deviations after each iteration. When the deviation data for three consecutive iterations meet the threshold requirements, the system operating parameters are considered to have reached a stable optimization state. The iteration is then stopped, and the final real-time optimized operating parameters are output (including the final output power of each energy input module, the final current density and temperature of the electrolysis hydrogen production module, and the final medium flow rate and concentration of the photothermal control platform). These final parameters must be stored in the system database (with a storage period of one year) for subsequent operation and maintenance analysis and model optimization. This step requires setting a maximum number of iterations (usually 10). If the maximum number of iterations is reached but the termination condition is not met, an alarm mechanism is triggered, indicating an anomaly in the system (such as sensor failure or equipment failure), requiring manual intervention for troubleshooting. Simultaneously, the current optimal parameters are output to maintain system operation. The total iteration time must be controlled within 30 minutes (corresponding to a maximum of 10 iterations, with each iteration lasting 5 minutes) to ensure the continuity and stability of system operation and avoid affecting hydrogen production efficiency due to excessive iteration time.

[0040] Preferably, the prediction output expression of the structured process knowledge-guided neural network is: y pred =σ(W3·tanh(W2·(W1x+b1)+K·f proc (x)+b2)+b3), where y pred The system is defined as a sequence of predicted parameters for a pre-defined future time period, including predicted energy supply, predicted demand for hydrogen production via electrolysis, and predicted properties of the aqueous medium; x is the set of real-time operating parameters input from the system; E1, W2, and W3 are the weight matrices of each hidden layer and output layer of the neural network, respectively; b1, b2, and b3 are the bias vectors of each hidden layer and output layer of the neural network, respectively; σ(·) is the activation function of the output layer; tanh(·) is the activation function of the hidden layer; K is the weight coefficient of the structured process knowledge; f proc (x) is the process function corresponding to the embedded multi-energy complementary energy conversion rule and the knowledge of the electrolytic hydrogen production reaction kinetics.

[0041] Specifically, structured process knowledge guides neural network prediction outputs. During implementation, the specific parameters and functions of each network component are clearly defined: the predicted output includes the energy supply for 12 five-minute intervals within the next hour, the electrolysis hydrogen production load demand, and the predicted characteristics of the aqueous medium. The input set of real-time system operating parameters must include more than 30 core parameters (such as solar irradiance, wind power output, and electrolyzer current density). The weight matrices for each layer of the network need to be determined through training with historical data. The dimensions of the weight matrices from the input layer to the first hidden layer match the number of input parameter terms (e.g., 30 parameters correspond to a 30-row, 50-column matrix), the first to second hidden layer is a 50-row, 40-column matrix, the second to third layer is a 40-row, 30-column matrix, and the third to output layer is a 30-row, 6-column matrix. The dimension of the bias vector is consistent with the number of output nodes in the corresponding layer (e.g., the first layer bias vector is 50-dimensional, and the output layer is 6-dimensional). For activation function selection, the output layer uses the Sigmoid function (suitable for parameter range constraints), and the hidden layer uses the hyperbolic tangent function (to enhance nonlinear fitting ability). The weight coefficients of structured process knowledge need to be set according to the importance of the knowledge. For example, the coefficient corresponding to the energy conversion rule is set to 0.6, and the coefficient corresponding to the electrolytic hydrogen production reaction kinetics knowledge is set to 0.4. The process function needs to quantify the knowledge association, such as converting the cubic relationship between wind speed and wind power output into a specific calculation formula. In implementation, the input parameters are first substituted into the network for forward propagation. Each layer of calculation incorporates the corresponding weights and biases. At the same time, structured knowledge is introduced through the process function to correct intermediate results. Finally, the predicted sequence is output. This process must ensure that the root mean square error of the prediction does not exceed 5%, and the system is retrained every 7 days to update the weights and knowledge coefficients to ensure that the prediction accuracy dynamically adapts to the system's operating status.

[0042] Preferably, the main model expression for system operation optimization constructed by the split-bar chance-constrained programming algorithm is:

[0043]

[0044] Where u represents the system operation optimization decision variable, including the output power of each energy input module, the operating parameters of the electrolysis hydrogen production module, and the control parameters of the photothermal effect-induced aqueous phase regulation platform; U represents the feasible region of the decision variable; c represents the cost coefficient vector of the decision variable; c T u is the objective function for system operating cost; This represents a family of distributions for the uncertainty parameter ξ, which includes fluctuations in energy supply, fluctuations in hydrogen production load due to electrolysis, and fluctuations in the properties of the aqueous medium; g i (u, ξ) are the system operation constraint functions for the i-th system; ∈ i Let be the threshold for the allowed violation probability of the i-th constraint; n is the total number of constraints; It is a probability measure.

[0045] Specifically, the main model for system operation optimization is constructed using the Bruker chance-constrained programming algorithm. During implementation, the specific composition of the decision variables and objective function is clearly defined: the decision variables include more than 15 parameters, including the output power of each energy input module (photovoltaic 0-500kW, wind power 0-800kW, grid 0-600kW) and the operating parameters of the electrolysis hydrogen production module (current density 800-3500A / m³). 2 Electrode temperature 60-85℃) and control parameters of the photothermal regulation platform (medium flow rate 5-15m³ / h) 3 / h, temperature 50-75℃, concentration 20%-30%); the feasible region of decision variables must strictly match the equipment hardware limitations, such as the photovoltaic output power not exceeding its rated value of 500kW. The cost coefficient vector needs to be set in combination with actual cost data, with the grid purchase cost coefficient at 0.5 yuan / kWh, photovoltaic operation and maintenance cost coefficient at 0.02 yuan / kWh, wind power at 0.03 yuan / kWh, electrolyzer at 0.05 yuan / kWh, and solar thermal regulation at 0.04 yuan / kWh. The objective function calculates the total operating cost by weighted summation and minimizes it. The uncertainty parameter distribution family is set as an interval distribution, including energy supply fluctuations (photovoltaic ±20kW, wind power ±50kW), hydrogen production load fluctuations (±30kW), and water phase medium characteristic fluctuations (temperature ±3℃, concentration ±2%). The constraint function needs to quantify the system operation limitations, such as the electrolyzer current density constraint function being set as the current density value and 3500A / m 2 The difference is less than 0, and is related to 800A / m 2 The difference is greater than 0; the allowed violation probability threshold is distinguished according to the importance of the constraint: hard constraints (such as current density, power limit) are set at 95%, and soft constraints (such as cost fluctuation) are set at 90%. In implementation, the objective function and constraint function are first transformed into mathematical expressions, and the uncertainty parameters fluctuation range and probability threshold are substituted. The main model is constructed through the algorithm, and then Monte Carlo simulation (more than 1000 simulations) is used to verify the feasibility. If the constraint satisfaction probability does not meet the standard, the fluctuation range (such as reducing wind power fluctuation to ±40kW) or the threshold (not less than 85%) is adjusted. Finally, a solvable optimization main model is formed to ensure that the model is feasible in more than 90% of operating scenarios.

[0046] Preferably, the iterative update expression of the ADMM-based multi-agent collaborative optimization algorithm is:

[0047]

[0048] Among them, u k U is the decision variable for the k-th subsystem; k Let be the feasible region of the decision variable for the k-th subsystem; m be the number of subsystems; t be the number of iterations; λ be the Lagrange multiplier vector; ρ be the penalty parameter; L ρ (·) represents the augmented Lagrangian function; Ak Let be the correlation matrix of the decision variables of the k-th subsystem; b is the right-hand vector of the inter-subsystem collaborative constraint equation. Let λ be the value of the decision variable in the (t+1)th iteration of the k-th subsystem; t+1 Let be the Lagrange multiplier value for the (t+1)th iteration.

[0049] Specifically, based on the ADMM-based multi-agent collaborative optimization algorithm, the implementation involves dividing the system into subsystems and setting initial parameters: the subsystems are divided into three parts: energy input, electrolysis hydrogen production, and solar thermal regulation. The decision variables of each subsystem are consistent with the corresponding part of the main model (e.g., the energy input subsystem includes three variables: photovoltaic, wind power, and grid power). The feasible region must match the hardware limitations of the subsystem (e.g., the medium flow rate of the solar thermal regulation subsystem is 5-15 m³ / s). 3 / h). In the initial parameters of the iteration, the initial values ​​of the decision variables are set to the current actual operating parameters (e.g., if the photovoltaic output is currently 100kW, then the initial value is set to 100kW). The initial value of the Lagrange multiplier is uniformly set to 0.1 (to ensure that the initial iteration step size is appropriate). The penalty parameter is set to 10 according to the requirements of collaborative accuracy (the larger the value, the stricter the collaborative constraint). The augmented Lagrange function needs to integrate the subsystem objective function and collaborative constraints. For example, the energy input subsystem function includes its operating cost term and the term multiplied by the Lagrange multiplier by the difference between the total energy power and the hydrogen production load. The subsystem correlation matrix needs to quantify the parameter correlation between subsystems. For example, the correlation matrix between the energy input subsystem and the electrolysis hydrogen production subsystem is set to 1 (indicating that the energy power needs to match the hydrogen production load). The vector on the right side of the collaborative constraint equation is set to the predicted value of the hydrogen production load (e.g., 250kW). The iterative convergence criterion is set as follows: the absolute value of the iteration deviation of the decision variable is less than 0.01 (i.e., 1%, such as the difference between two iterations of photovoltaic power being less than 1kW), and the constraint deviation is less than 0.05 (such as the difference between total energy power and hydrogen production load being less than 12.5kW). During implementation, the collaborative center first issues initial parameters. Each subsystem solves its own function to obtain the first round of decision variables. After uploading, the center calculates the constraint deviation and updates the Lagrange multipliers (the update amount is the penalty parameter multiplied by the deviation value). The updated multipliers are then issued, and the subsystems repeat the solution and upload process. Typically, after 5-10 iterations, the convergence criterion is met, and preliminary optimization parameters are output. This process must ensure that the solution time for each subsystem does not exceed 5 seconds, and the overall iteration time does not exceed 60 seconds, guaranteeing real-time optimization efficiency.

[0050] Preferably, the medium state regulation model expression of the photothermal effect-induced aqueous phase control platform is:

[0051]

[0052] Among them, T t+1 T t These represent the temperatures of the aqueous medium within the platform at time t+1 and time t, respectively; α is the photothermal conversion efficiency coefficient; Qopt The light energy input to the photothermal module; β is the heat dissipation coefficient of the medium; T env γ represents ambient temperature; γ represents the heat transfer coefficient of the medium flow rate; q flow T is the medium circulation flow rate; in To compensate for the temperature of the medium; C t+1 C t δ represents the concentration of the aqueous medium within the platform at time t+1 and time t, respectively; C represents the mixing coefficient of the supplementary medium; add To replenish the medium concentration; q add To supplement the medium flow rate; q total ∈ represents the total flow rate of the medium within the platform; ∈ represents the reaction consumption coefficient; r react Δt represents the rate at which the medium participates in the electrolysis-assisted reaction; Δt represents the time step.

[0053] Specifically, in the implementation of the medium state regulation model for the photothermal effect-induced aqueous phase control platform, the specific values ​​and physical meanings of the model parameters are clearly defined: In the temperature regulation model, the photothermal conversion efficiency coefficient is set to 0.8 based on the performance of the photothermal module (representing that 80% of the light energy is converted into heat energy), and the input light energy of the photothermal module is calculated based on the light intensity (e.g., irradiance of 800 W / m²). 2 The input energy is 400kW; the heat dissipation coefficient needs to be set to 0.05 (unit: kW / ℃) based on the platform's insulation performance, and the ambient temperature should be taken as the real-time monitoring value (e.g., 25℃); the flow rate heat transfer coefficient is set to 0.03 (unit: kW / (m³)) based on the heat exchanger efficiency. 3 / h·℃), the medium circulation flow rate is the current control value (e.g., 10m³). 3 / h), with the replenishing medium temperature at ambient temperature (25℃). In the concentration adjustment model, the replenishing medium mixing coefficient is set to 0.7 based on the mixing device efficiency (indicating that 70% of the replenishing medium is uniformly mixed with the original medium), the replenishing medium concentration is set to 35% (higher than the original concentration to achieve concentration increase), and the replenishing medium flow rate is adjusted according to the concentration deviation (e.g., set to 1m when the concentration is below 25%). 3 / h), the total medium flow rate is the sum of the circulating flow rate and the replenishment flow rate (e.g., 11m³ / h). 3 / h); the reaction consumption coefficient is set to 0.02 (in % / h) based on the reaction rate, and the rate at which the medium participates in the electrolytic auxiliary reaction is calculated based on the electrolytic current density (e.g., 2500 A / m). 2The flow rate is 0.5% / h, and the time step is set to 5 minutes (consistent with the system optimization interval). During implementation, the current medium temperature and concentration are first collected and substituted into the model to calculate the parameters for the next moment: the temperature calculation needs to add the photothermal heating amount, heat loss and flow heat exchange, and the concentration calculation needs to consider the mixing of the supplementary medium and the consumption of the reaction. The calculation results are used as the target control values ​​for the next moment, and the PLC controller drives the actuators (circulating pump, heating pipe, valve) to adjust. During the adjustment process, the actual parameters are collected every 10 seconds and compared with the model calculation values. If the deviation exceeds ±1℃ (temperature) or ±0.5% (concentration), the parameters are resubmitted into the model for correction to ensure that the medium state is stable within the optimization range and to provide a suitable aqueous phase environment for electrolytic hydrogen production.

[0054] Preferably, the correlation model expression between the total energy consumption and hydrogen production capacity of the multi-energy complementary electrolysis hydrogen production system is as follows:

[0055]

[0056] Among them, H prod η represents the total hydrogen production within the preset time period [t0, t1]. total The overall hydrogen production efficiency of the system is represented by p; the number of energy input modules is represented by P. j (t) represents the output power of the j-th energy input module at time t; η j P represents the energy transfer efficiency from the j-th energy input module to the electrolysis hydrogen production module. loss (t) represents the total energy loss of the system at time t, including line loss, equipment loss, and energy consumption of the water phase control platform induced by photothermal effect.

[0057] Specifically, the correlation model between total energy consumption and hydrogen production in a multi-energy complementary electrolysis hydrogen production system requires determining the specific values ​​and calculation methods for each parameter during implementation. The overall hydrogen production efficiency needs to be calculated comprehensively based on the efficiency of each component of the system, including energy transmission efficiency (photovoltaic 0.9, wind power 0.92, grid 0.98) and electrolyzer efficiency (0.75-0.85, varying with current density, e.g., 2500 A / m²). 2The overall efficiency is obtained by weighted averaging the energy input modules (0.8 for solar power, 0.8 for solar thermal regulation, and 0.78 for solar thermal regulation), typically set to 0.78. The number of energy input modules is determined by the system configuration and is 3 (photovoltaic, wind power, and grid). The output power of each module needs to be collected in real time (sampling frequency 1 minute / time) and multiplied by the corresponding transmission efficiency to obtain the actual energy input to the electrolysis hydrogen production module; for example, if the photovoltaic output is 120kW and the efficiency is 0.9, the actual input energy is 108kW. The total system energy loss needs to be calculated separately: line loss is set according to the transmission distance (e.g., 0.02 for a 100-meter line, 2.4kW for a 120kW transmission loss), equipment loss (electrolyzer 0.03kW, circulating pump 0.01kW), and solar thermal regulation platform energy consumption (heating device 80kW, pump 0.5kW). The total energy loss is obtained by summing these losses. The time interval is set to 1 hour (consistent with the predicted time period). The integral calculation is achieved by accumulating the (total input energy - total energy loss) value for each minute within this interval, and then multiplying it by the overall hydrogen production efficiency to obtain the total hydrogen production. During implementation, power, efficiency, and loss data of each module are collected every hour, substituted into the model to calculate the hydrogen production for that period, and compared with the actual metering value (such as hydrogen flow meter data). If the deviation exceeds ±3%, the overall hydrogen production efficiency is adjusted (e.g., corrected to 0.77) to ensure that the model can accurately reflect the correlation between energy consumption and hydrogen production, providing data support for system operating cost accounting and optimization target adjustment.

[0058] Preferably, step S3 includes the following sub-steps: S31: Extract the predicted values ​​of multi-energy complementary energy supply, electrolysis hydrogen production load demand, and aqueous medium characteristics from the predicted parameter sequence output in step S2. Use these predicted values ​​as basic data to divide the fluctuation range of each parameter within a preset time period in the future, and determine the probability distribution type of each fluctuation range; S32: Based on the operating requirements of the multi-energy complementary electrolysis hydrogen production system, clarify the hard constraints and soft constraints in the system operation process. Hard constraints include the maximum current density limit of the electrolysis hydrogen production module and the upper limit of the output power of each energy input module. Soft constraints include the fluctuation range of system operating costs. Simultaneously, set... S33: Using the minimization of system operating cost as the objective function, the parameter fluctuation range and probability distribution type determined in step S31, the constraint conditions and probability thresholds set in step S32 are integrated into the split-bar chance-constrained programming model framework to construct a main system operation optimization model including uncertain parameters, and the value range and correlation of each variable in the model are defined; S34: Feasibility analysis is performed on the constructed main system operation optimization model. By adjusting the fluctuation range of uncertain parameters and the constraint satisfaction probability thresholds, it is ensured that the model has a feasible solution under the preset system operation scenario, thus forming the final main system operation optimization model.

[0059] Specifically, in step S3, S31 is executed first, extracting predicted values ​​for energy supply (photovoltaic 50-150kW, wind power 100-300kW), electrolysis hydrogen production load demand (200-400kW), and aqueous medium characteristics (temperature 50-75℃, concentration 20%-30%) from the predicted parameter sequence output in step S2. The fluctuation range of each parameter is then divided into 12 five-minute intervals within the next hour, and the distribution type (photovoltaic normal distribution, wind power Weibull distribution) is determined based on historical data from the past three months. Next, in S32, hard constraints are defined (electrolyzer current density 800-3500A / m³). 2 The power limits for each energy module are set as follows: photovoltaic 500kW, wind power 800kW, and grid power 600kW. Soft constraints (daily cost fluctuation ±8%) are also defined. Hard constraints are satisfied with a probability threshold of 95%, and soft constraints with a probability threshold of 90%, based on importance. Then, in S33, with the goal of minimizing total operating costs (grid power purchase 0.5 yuan / kWh, photovoltaic operation and maintenance 0.02 yuan / kWh, etc.), the intervals and distributions of S31, and the constraints and thresholds of S32 are integrated into a distributed bar framework to construct an optimization master model containing more than 15 decision variables (such as solar thermal medium flow rate) and more than 20 constraints. The variable range is defined as (flow rate 5-15m³ / h). 3 / h) and correlation (current density is positively correlated with hydrogen production); finally, S34, the feasibility is verified by more than 1000 Monte Carlo simulations. If the constraint satisfaction probability does not meet the standard, the fluctuation range is reduced (wind power is adjusted to 80-280kW) or the threshold is fine-tuned (minimum 85%) to ensure that the model has a feasible solution and lay the foundation for subsequent solutions.

[0060] Preferably, step S4 includes the following sub-steps: S41: Based on the structural composition of the multi-energy complementary electrolysis hydrogen production system, the main model for system operation optimization is decomposed into corresponding sub-models according to the energy input subsystem, the electrolysis hydrogen production subsystem, and the photothermal effect-induced aqueous phase control subsystem, and the decision variables, objective function, and local constraints of each sub-model are defined; S42: The initial parameters of the multi-agent collaborative optimization algorithm based on ADMM are set, including the initial values ​​of the decision variables of each sub-model, the initial values ​​of the Lagrange multipliers, and the values ​​of the penalty parameters. At the same time, the iterative convergence criteria of the algorithm are determined, which include the iterative deviation threshold of the decision variables and the constraint satisfaction deviation threshold. S43: Each sub-model solves its own objective function based on the initial parameters and the boundary parameters passed from other sub-models, obtaining its own preliminary decision variable values. These preliminary decision variable values ​​are then passed to the collaborative optimization center, which calculates the constraint deviations between the sub-models. S44: Based on the constraint deviations calculated by the collaborative optimization center, the Lagrange multipliers are updated. Each sub-model then resolves its objective function based on the updated Lagrange multipliers and penalty parameters, obtaining new decision variable values. The parameter passing, constraint deviation calculation, and variable update process is repeated until the iterative convergence criterion is met, and the preliminary operating optimization parameters of each subsystem are output.

[0061] Specifically, in step S4, S41 is executed first, decomposing the main optimization model into sub-models for energy input, electrolysis hydrogen production, and solar thermal regulation according to the system structure. The decision variables (energy sub-model includes photovoltaic, wind power, and grid power; electrolysis sub-model includes current density and electrode temperature; solar thermal sub-model includes medium flow rate and temperature), objective function (minimizing subsystem cost), and local constraints (solar thermal medium temperature 45-80℃) for each sub-model are defined. Next, in S42, the initial parameters of ADMM are set. The initial values ​​of the decision variables are taken as the current actual values ​​(e.g., if the photovoltaic current is 100kW, then the initial value is 100kW), the Lagrange multiplier is 0.1, the penalty parameter is 10, and the convergence criterion is the absolute value of the iteration deviation of the decision variables. <0.01 (e.g., photovoltaic power difference <1kW), constraint deviation <0.05 (total energy power and hydrogen production load difference <12.5kW); then in S43, each sub-model solves the objective function based on the initial parameters and the boundary parameters of other sub-models (e.g., the energy sub-model receives the load demand of the electrolysis sub-model), obtains the preliminary decision variables, and uploads them to the collaboration center, which calculates the constraint deviation; finally, in S44, the Lagrange multipliers are updated according to the deviation (update amount = penalty parameter × deviation), and the sub-models are re-solved based on the updated parameters. The process of transmission, calculation, and updating is repeated, usually 5-10 iterations to meet the criteria, and the preliminary optimized parameters are output (e.g., photovoltaic 120kW, electrolysis current density 2500A / m). 2 This ensures that the solution time for the sub-model is ≤5 seconds and the overall iteration time is ≤60 seconds, guaranteeing real-time performance.

[0062] Preferably, step S5 includes the following sub-steps: S51: Classify and organize the preliminary operation optimization parameters of each subsystem output in step S4, and screen out the control parameters related to the photothermal effect-induced aqueous phase control platform, including medium flow control parameters, temperature adjustment parameters, and concentration adjustment parameters. At the same time, extract the electrode voltage and current density parameters of the electrolysis hydrogen production module and the output power parameters of each energy input module; S52: Input the screened control parameters of the photothermal effect-induced aqueous phase control platform to the actuator of the platform. The actuator adjusts the speed of the medium circulation pump to change the medium flow, adjusts the heating power of the photothermal module to change the medium temperature, and adjusts the opening of the replenishing medium valve to change the medium flow according to the parameter instructions. S53: The extracted electrode voltage and current density parameters of the electrolysis hydrogen production module are sent to the controller of the electrolysis hydrogen production module. The controller adjusts the voltage and current output of the electrode power supply circuit to make the operating parameters of the electrolysis hydrogen production module reach the preliminary optimized value. At the same time, the output power parameters of each energy input module are transmitted to the corresponding energy control module to adjust the energy output. S54: The medium state parameters after adjustment by the photothermal effect-induced aqueous phase control platform, the actual operating parameters of the electrolysis hydrogen production module, and the actual output parameters of each energy input module are collected. These parameters are compared with the preliminary operating optimization parameters, the parameter deviation is calculated, and the deviation information is fed back to step S4 for subsequent parameter iterative adjustment.

[0063] Specifically, in step S5, S51 is executed first to classify and organize preliminary optimization parameters and screen photothermal control parameters (medium flow rate 10m). 3 / h, temperature 75℃, concentration 25%, makeup flow rate 1m 3 / h), electrolysis parameters (current density 2500A / m 2 Electrode temperature 75℃), energy parameters (photovoltaic 120kW, wind power 220kW, grid 130kW); then S52, the photothermal parameters are input into the PLC controller, driving the circulation pump speed to 2000rpm (corresponding to 10m). 3 / h), turn on 4 20kW heating elements (total 80kW) to control the temperature, and adjust the opening of the supplementary valve to 30% (corresponding to 1m 3 / h), combined with the concentration sensor feedback to maintain a concentration of 25%, and collect the medium temperature and concentration every 10 seconds to ensure stability at 75℃±1℃ and 25%±0.5%; then S53, the electrolysis parameters are sent to the electrolytic cell controller, and the rectifier output is adjusted to 5000A (corresponding to 2500A / m). 2 The area of ​​the tank is 2m² 2The system controls the heating element power to 5kW to maintain 75℃, monitors hydrogen production every minute, and sends energy parameters to the inverter (MPPT adjusted to 120kW), converter (pitch angle adjusted to 220kW), and grid controller (circuit breaker adjusted to 130kW). Finally, in S54, the actual parameters of each module are collected and compared with the preliminary optimized parameters to calculate the deviation (e.g., temperature deviation = actual - 75℃). The deviation data is fed back to the coordination center for the next round of iteration correction, ensuring that the adjustment response is <2 seconds, the parameter stability is <30 seconds, and the closed-loop effectiveness is guaranteed.

[0064] The structured process knowledge-guided neural network in this invention integrates professional knowledge of multi-energy complementary electrolytic hydrogen production systems with data-driven predictive models. It embeds structured process knowledge of system operation into the traditional neural network architecture, rather than simply relying on data training. In implementation, the three-layer hidden layer structure of the network is first determined. The number of input layer nodes matches the number of real-time system operating parameters (e.g., 30 items), and the output layer nodes correspond to six key prediction parameters (energy supply, hydrogen production load demand, changes in aqueous medium characteristics, etc.). Then, the multi-energy complementary energy conversion rules (e.g., the correlation between irradiance intensity and photovoltaic power, wind speed and wind power) and electrolytic hydrogen production reaction kinetics knowledge (e.g., the relationship between electrode temperature and reaction rate, current density and hydrogen production) are transformed into computable process functions, and knowledge weight coefficients (e.g., 0.6 for energy rules, 0.4 for kinetics knowledge) are incorporated into the calculation of the intermediate layers of the network. Finally, the network is trained using historical data from the past three months (no less than 100,000 sets), controlling the root mean square error of prediction to be no more than 5%, and the parameters are retrained and updated every seven days. The network's function is to accurately predict the changing trends of key parameters over 12 time periods within the next hour, based on the system's real-time operating parameters, providing reliable data support for subsequent optimization modeling. It breaks away from the limitations of traditional single-data-driven models, improving prediction accuracy and parameter adaptability through knowledge embedding. This avoids prediction biases caused by data noise or sudden changes in operating conditions, ensuring that the optimization direction highly aligns with the system's actual needs, and laying the foundation for the overall real-time optimization process.

[0065] The Weibull chance-constrained programming algorithm in this invention is a modeling method for addressing system parameter uncertainties and ensuring the feasibility of the optimization model. It considers parameter fluctuations while ensuring that constraints are met in most operating scenarios, rather than simply pursuing the optimal solution under ideal conditions. In implementation, it first extracts the fluctuation ranges of energy supply, hydrogen production load, and aqueous medium characteristics (e.g., 50-150kW for photovoltaic, 100-300kW for wind power) from the predicted parameter sequence to determine the parameter distribution type (e.g., normal distribution for photovoltaic, Weibull distribution for wind power); then it distinguishes the system's hard constraints (e.g., electrolyzer current density 800-3500A / m³). 2The algorithm employs several methods to optimize system operation. It addresses constraints such as the upper limit of energy module power and soft constraints (daily cost fluctuation ±8%), setting constraint satisfaction probability thresholds based on importance (95% for hard constraints, 90% for soft constraints). Then, with the goal of minimizing total operating costs (including grid power purchase, equipment maintenance, and solar thermal control costs), it integrates parameter fluctuation ranges, distribution types, constraints, and thresholds into the algorithm framework, constructing an optimization master model with more than 15 decision variables and more than 20 constraints. Finally, it verifies the model's feasibility through more than 1000 Monte Carlo simulations. If the constraint satisfaction probability is insufficient, the parameter fluctuation range is reduced (e.g., wind power is adjusted to 80-280kW) or the threshold is fine-tuned (minimum 85%) to ensure model feasibility. The algorithm's purpose is to construct a system operation optimization master model that balances uncertainty and constraint satisfaction, providing a scientific optimization framework for subsequent solutions. It solves the problem that traditional deterministic optimization models struggle to handle parameter fluctuations, improves the model's applicability under complex real-world conditions, and ensures that the optimization results reduce operating costs while meeting system hardware limitations and operational requirements, avoiding system failures or efficiency losses due to parameter fluctuations.

[0066] The ADMM-based multi-agent collaborative optimization algorithm in this invention is a solution method that decomposes and solves complex system optimization problems and achieves collaboration between subsystems and the overall goal. It uses alternating iteration and information interaction to allow each subsystem to solve independently while taking into account the overall constraints, rather than directly solving the complex overall model. In implementation, the main optimization model is first decomposed into three sub-models based on the system structure: energy input, electrolytic hydrogen production, and solar thermal regulation. The decision variables (e.g., the energy sub-model includes photovoltaic, wind power, and grid power), objective function (minimizing subsystem cost), and local constraints (e.g., solar thermal medium temperature 45-80℃) for each sub-model are defined. Then, initial parameters are set (initial values ​​of decision variables are the current actual values, Lagrange multipliers are 0.1, and penalty parameter is 10) and convergence criteria (iteration deviation of decision variables < 0.01, constraint deviation < 0.05). Next, each sub-model solves the objective function based on the initial parameters and the boundary parameters of other sub-models. The preliminary decision variables are uploaded to the collaborative center, which calculates the constraint deviation and updates the Lagrange multipliers. Finally, the sub-models are re-solved based on the updated parameters, iterating 5-10 times until the criteria are met, and the preliminary optimization parameters are output. The algorithm's function is to efficiently solve the main optimization model, obtain the preliminary operational optimization parameters for each subsystem, and ensure both solution efficiency and global optimality. To address the issues of traditional overall solution methods being too time-consuming and unable to adapt to real-time optimization requirements, this method balances solution efficiency and optimization effectiveness through decomposition and collaborative optimization. This ensures that the optimization objectives of each subsystem are consistent with the overall system objective, avoids overall efficiency loss caused by independent optimization of subsystems, and meets the timeliness requirements of real-time system optimization.

[0067] The photothermal effect-induced aqueous phase control platform of this invention is an auxiliary control device that adjusts the state of the aqueous medium to ensure the efficiency of hydrogen production through electrolysis. It utilizes photothermal conversion and medium parameter adjustment to provide a suitable aqueous environment for the hydrogen production process, rather than relying solely on the electrolysis module's own adjustment. In implementation, the platform consists of a water storage tank, a heating device (including multiple 20kW heating tubes), a circulating pump (speed 1000-3000rpm), a concentration sensor, and a replenishing medium valve (opening degree 0-100%). It first receives the medium control parameters (e.g., flow rate 10m³) from the preliminary optimization parameters. 3 / h, temperature 75℃, concentration 25%, makeup flow rate 1m 3 The system, driven by a PLC controller, operates as follows: the circulating pump adjusts its speed to 2000 rpm to control the flow rate; the heating device activates four heating tubes (total power 80kW) to control the temperature; and the supplementary valve adjusts its opening to 30%, maintaining the concentration based on feedback from the concentration sensor. During adjustment, medium temperature and concentration data are collected every 10 seconds to ensure stability at 75℃±1℃ and 25%±0.5%. Simultaneously, the actual parameters are compared with optimized parameters to calculate deviations, which are then fed back to the collaborative center for iterative correction. The platform's function is to precisely regulate the aqueous medium state, providing a suitable ion conduction environment for the electrolytic hydrogen production module. It also coordinates with the energy input and electrolysis module to achieve closed-loop parameter regulation. This overcomes the shortcomings of traditional systems that neglect the influence of the aqueous medium. By optimizing the medium state, it improves electrolysis efficiency (e.g., suitable concentration increases ion conductivity, stable temperature ensures reaction rate), and by enhancing system parameter synergy through closed-loop regulation, it avoids decreased electrolysis efficiency or equipment damage due to fluctuations in the medium state, further improving the overall system's operational stability and economy.

[0068] like Figure 2 As shown, the real-time operation optimization system for the multi-energy complementary electrolysis hydrogen production system includes:

[0069] A multi-parameter collaborative acquisition unit, connected to the energy input modules, electrolysis hydrogen production module, and photothermal effect-induced aqueous phase control platform of the multi-energy complementary electrolysis hydrogen production system, is used to acquire real-time operating parameters of each module and construct a real-time operating parameter set for the system. A structured knowledge-driven prediction unit, connected to the multi-parameter collaborative acquisition unit, incorporates a structured process knowledge-guided neural network. It receives the real-time operating parameter set output by the multi-parameter collaborative acquisition unit, predicts the parameter change trends within a preset future time period, and outputs a predicted parameter sequence. A sub-Bruker bar optimization modeling unit, connected to the structured knowledge-driven prediction unit, uses a sub-Bruker bar chance-constrained programming algorithm to construct a master model for system operation optimization based on the predicted parameter sequence, determining the fluctuation range of uncertain parameters and the probability threshold for constraint satisfaction in the model. (Multi-master...) The system comprises four subsystems: a multi-agent collaborative solution unit (MAS unit), connected to the multi-agent bar optimization modeling unit, and a closed-loop parameter adjustment unit. The MAS unit decomposes the main system operation optimization model into sub-models, solves the sub-models using an ADMM-based multi-agent collaborative optimization algorithm, and outputs the preliminary operation optimization parameters for each subsystem. A closed-loop parameter adjustment unit, connected to the multi-agent collaborative solution unit, the electrolysis hydrogen production module, and the photothermal effect-induced aqueous phase control platform, adjusts the operation parameters of these modules based on the preliminary operation optimization parameters and provides feedback on parameter deviations. An iterative optimization output unit, connected to the closed-loop parameter adjustment unit, receives parameter deviation information and controls the repeated operation of the structured knowledge-driven prediction unit, the multi-agent bar optimization modeling unit, the multi-agent collaborative solution unit, and the closed-loop parameter adjustment unit. When the parameter deviation meets the requirements, the final optimization parameters are output.

[0070] This invention discloses a real-time operation optimization method and system for a multi-energy complementary electrolytic hydrogen production system. By guiding a neural network with structured process knowledge, it deeply embeds the rules of multi-energy complementary energy conversion and the kinetics of electrolytic hydrogen production into the model, breaking the limitations of traditional single algorithms and achieving an effective fusion of structured process knowledge and data-driven models. This fusion accurately captures the correlations between various parameters within the system, making the prediction results more closely match actual operational needs. It fundamentally solves the problem of insufficient adaptability of optimization parameters in the background technology, providing a precise parameter basis for subsequent optimization modeling. Simultaneously, it enhances the system's predictive ability for energy supply fluctuations and hydrogen production load changes, ensuring a high degree of matching between the optimization direction and the actual operating conditions of the system.

[0071] In the optimization modeling and solution stages, a master model for system operation optimization is constructed using the Bruker chance-constrained programming algorithm. Simultaneously, an ADMM-based multi-agent collaborative optimization algorithm is employed to decompose and collaboratively solve the master model into sub-models. This, combined with a multi-agent collaborative solution unit, forms an efficient multi-agent collaborative control mechanism. This mechanism effectively coordinates the optimization objectives of each subsystem with the overall system, avoiding the constraint coupling problem caused by independent optimization of factor systems in traditional techniques. It effectively solves the deficiency of incoordination between subsystem objectives and overall objectives in the background technology, allowing each subsystem to work collaboratively to reduce overall system operating costs and improve efficiency, ensuring the global optimality of the optimization results.

[0072] This invention also incorporates a photothermal effect-induced closed-loop parameter adjustment of the aqueous phase control platform, the electrolysis hydrogen production module, and the energy input module, combined with an iterative optimization output unit to form a complete optimization process of "parameter acquisition-prediction-modeling-solving-adjustment-iteration." This process enables dynamic real-time adjustment of the operating parameters of each module. Simultaneously, the photothermal effect-induced precise adjustment of the aqueous phase medium state by the aqueous phase control platform further ensures electrolysis efficiency, comprehensively overcoming the shortcomings of incomplete optimization processes and the inability to dynamically adjust parameters in background technologies. The overall process allows the system to continuously optimize operating parameters based on real-time conditions, ultimately significantly improving the system's stability and economy, ensuring high efficiency during long-term operation.

[0073] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0074] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for real-time operation optimization of a multi-capacity complementary electrolytic hydrogen production system, characterized in that, The method comprises the following steps: Step S1: Collecting real-time operation parameters of each energy input module in the multi-energy complementary electrolytic hydrogen production system, working condition parameters of the electrolytic hydrogen production module, and medium state parameters of the photothermal effect induced water phase regulation platform, and constructing a system real-time operation parameter set; Step S2: Inputting the system real-time operation parameter set into a structured process knowledge guided neural network, and predicting the energy supply amount, electrolytic hydrogen production load demand and water phase medium characteristic change trend in a future preset time period through the multi-energy complementary energy conversion rule and electrolytic hydrogen production reaction kinetics knowledge embedded in the network, and outputting a prediction parameter sequence; Step S3: Based on the prediction parameter sequence, using a distribution robust chance constraint programming algorithm, taking the minimization of system operation cost as the target, combining the multi-energy complementary energy supply and demand fluctuation interval and the electrolytic hydrogen production process constraint condition, establishing a system operation optimization main model, and determining the fluctuation range of the uncertainty parameters in the model and the constraint satisfaction probability threshold; Step S4: Decomposing the system operation optimization main model into a plurality of sub-models, each sub-model corresponding to a subsystem in the multi-energy complementary electrolytic hydrogen production system, using a multi-agent collaborative optimization algorithm based on ADMM, setting the collaborative optimization parameters and iteration convergence criteria between the sub-models, and solving the preliminary operation optimization parameters of each subsystem through alternating iteration calculation and information interaction of each sub-model.

2. The method for real-time operation optimization of a multiple function complementary electrolytic hydrogen generation system according to claim 1, characterized in that, The method further comprises the following steps: Step S5: Inputting the preliminary operation optimization parameters into the photothermal effect induced water phase regulation platform control module, adjusting the flow, temperature and concentration of the water phase medium in the platform according to the parameters, and simultaneously feeding back and regulating the electrode voltage, current density of the electrolytic hydrogen production module and the output power of each energy input module, forming a parameter closed-loop regulation; Step S6: Repeating steps S2 to S5, and performing deviation analysis on the system operation parameters obtained in each iteration, and outputting the final multi-energy complementary electrolytic hydrogen production system real-time operation optimization parameters when the parameter deviations of continuous multiple iterations are within a preset range; The prediction output expression of the structured process knowledge guided neural network is: y pred = σ(W3·tanh(W2·(W1x+b1)+K·f proc (x)+b2)+b3), wherein y pred is a predicted parameter sequence of the system in a future preset time period, including a predicted value of energy supply, a predicted value of electrolytic hydrogen load demand, and a predicted value of water phase medium characteristics; x is an input set of real-time operation parameters of the system; W1, W2, and W3 are weight matrices of each hidden layer and an output layer of the neural network respectively; b1, b2, and b3 are bias vectors of each hidden layer and the output layer of the neural network respectively; σ(·) is an output layer activation function; tanh(·) is a hidden layer activation function; K is a structured process knowledge weight coefficient; and f proc (x) is a process function corresponding to embedded multi-energy complementary energy conversion rules and electrolytic hydrogen reaction kinetics knowledge.

3. The method for real-time operation optimization of a multiple function complementary electrolytic hydrogen generation system according to claim 1, characterized in that, The system operation optimization main model expression constructed by the distribution robust chance constraint programming algorithm is: wherein u is a system operation optimization decision variable, including the output power of each energy input module, the operation parameters of the electrolytic hydrogen production module, and the control parameters of the water phase regulation platform induced by the photo-thermal effect; U is the feasible region of the decision variable; c is a decision variable cost coefficient vector; c T u is a system operation cost objective function; is a distribution family of uncertain parameters ξ, including energy supply fluctuation, electrolytic hydrogen production load fluctuation, and water phase medium characteristic fluctuation; g i (u, ξ) is the i-th system operation constraint function; ∈ i is the allowable violation probability threshold of the i-th constraint; n is the total number of constraints; is a probability measure.

4. The method for real-time operation optimization of a multiple function complementary electrolytic hydrogen generation system according to claim 1, characterized in that, The iteration update expression of the multi-agent collaborative optimization algorithm based on ADMM is: where u k is the decision variable of the kth subsystem; U k is the feasible region of the kth subsystem decision variable; m is the number of subsystems; t is the iteration number; λ is the Lagrange multiplier vector; ρ is the penalty parameter; L ρ (·) is the augmented Lagrangian function; A k is the associated matrix of the kth subsystem decision variable; b is the right-hand side vector of the inter-subsystem coordination constraint equation; is the value of the decision variable of the kth subsystem at the t+1th iteration; λ t+1 is the value of the Lagrange multiplier at the t+1th iteration.

5. The method for real-time operation optimization of a multiple functionally complementary electrolytic hydrogen generation system according to claim 1, characterized in that, The medium state adjustment model expression of the photothermal effect induced water phase regulation platform is: T t+1 = T t + α · Q opt - β · (T t - T env ) - γ · q flow · (T t - T in ) Wherein, T t+1 , T t are the temperatures of the aqueous medium in the platform at the t+1th moment and the tth moment respectively; α is the light-heat conversion efficiency coefficient; Q opt is the light energy input of the light-heat module; β is the medium heat dissipation coefficient; T env is the ambient temperature; γ is the medium flow heat exchange coefficient; q flow is the medium circulation flow; T in is the supplementary medium temperature; C t+1 , C t are the concentrations of the aqueous medium in the platform at the t+1th moment and the tth moment respectively; δ is the supplementary medium mixing coefficient; C add is the supplementary medium concentration; q add is the supplementary medium flow; q total is the total flow of the medium in the platform; ∈ is the reaction consumption coefficient; r react is the rate of the medium participating in the electrolysis auxiliary reaction; Δt is the time step.

6. The method for real-time operation optimization of a multiple functionally complementary electrolytic hydrogen generation system according to claim 1, characterized in that, The total energy consumption and hydrogen production amount correlation model expression of the multi-energy complementary electrolytic hydrogen production system is: wherein H prod is the total hydrogen production in the preset time period [t0, t1]; η total is the overall hydrogen production efficiency of the system; p is the number of energy input modules; P j (t) is the output power of the jth energy input module at time t; η j is the energy transmission efficiency from the jth energy input module to the electrolytic hydrogen production module; P loss (t) is the total energy consumption loss of the system at time t, including line loss, equipment loss, and photothermal effect-induced water phase regulation platform energy consumption.

7. The method for real-time operation optimization of a multiple functionally complementary electrolytic hydrogen generation system according to claim 1, characterized in that, Step S3 comprises the following steps: S31: Extracting the multi-energy complementary energy supply amount prediction value, electrolytic hydrogen production load demand prediction value and water phase medium characteristic prediction value from the prediction parameter sequence output in step S2, taking these prediction values as basic data, dividing the fluctuation interval of each parameter in the future preset time period, and determining the probability distribution type of each fluctuation interval; S32: According to the operation requirements of the multi-energy complementary electrolytic hydrogen production system, the hard constraints and soft constraints in the system operation process are determined, the hard constraints include the maximum current density limit of the electrolytic hydrogen production module, the output power upper limit of each energy input module, the soft constraints include the fluctuation range of the system operation cost, and the probability threshold of meeting each constraint under the distribution robust framework is set; S33: Taking the minimization of the system operation cost as the objective function, the parameter fluctuation interval and probability distribution type determined in step S31, the constraint conditions and probability threshold set in step S32 are integrated into the distribution robust chance constraint programming model framework, a system operation optimization main model including uncertain parameters is constructed, and the value range and correlation of each variable in the model are defined; S34: The feasibility of the constructed system operation optimization main model is analyzed, the fluctuation range of the uncertain parameters and the constraint satisfaction probability threshold are adjusted to ensure that the model has a feasible solution under the preset system operation scenario, and the final system operation optimization main model is formed.

8. The method for real-time operation optimization of a multiple functionally complementary electrolytic hydrogen generation system according to claim 1, characterized in that, Step S4 includes the following sub-steps: S41: According to the structural composition of the multi-energy complementary electrolytic hydrogen production system, the system operation optimization main model is decomposed into corresponding sub-models according to the energy input subsystem, the electrolytic hydrogen production subsystem and the light-heat effect induced water phase regulation subsystem, and the decision variables, objective function and local constraint conditions of each sub-model are determined; S42: Set the initial parameters of the multi-agent collaborative optimization algorithm based on ADMM, including the initial values of the decision variables of each sub-model, the initial values of the Lagrange multipliers and the values of the penalty parameters, and determine the iteration convergence criteria of the algorithm, including the iteration deviation threshold of the decision variables and the constraint satisfaction deviation threshold; S43: Based on the initial parameters and the boundary parameters transmitted by other sub-models, each sub-model solves its own objective function to obtain the preliminary decision variable values, and transmits these preliminary decision variable values to the collaborative optimization center, and the collaborative optimization center calculates the constraint deviation between sub-models; S44: According to the constraint deviation calculated by the collaborative optimization center, update the Lagrange multipliers, and based on the updated Lagrange multipliers and penalty parameters, each sub-model re-solves the objective function to obtain new decision variable values, and the parameter transmission, constraint deviation calculation and variable updating process are repeated until the iteration convergence criteria are met, and the preliminary operation optimization parameters of each subsystem are output.

9. The method for real-time operation optimization of a multiple functionally complementary electrolytic hydrogen generation system according to claim 2, characterized in that, Step S5 Comprise the following steps: S51: the preliminary operation optimization parameters of each subsystem output by step S4 are classified and arranged, the control parameters related to the photothermal effect induced water phase regulation platform are screened out, including medium flow control parameters, temperature regulation parameters and concentration adjustment parameters, at the same time, the electrode voltage, current density parameters of electrolytic hydrogen production module and the output power parameters of each energy input module are extracted; S52: the screened photothermal effect induced water phase regulation platform control parameters are input to the executing mechanism of the platform, the executing mechanism adjusts the rotating speed of the medium circulating pump to change the medium flow according to the parameter instruction, adjusts the heating power of the photothermal module to change the medium temperature, adjusts the valve opening of the supplementary medium to change the medium concentration; S53: the extracted electrode voltage, current density parameters of electrolytic hydrogen production module are sent to the controller of electrolytic hydrogen production module, the controller adjusts the voltage and current output of electrode power supply circuit to make the operation parameters of electrolytic hydrogen production module reach the preliminary optimization value, at the same time, the output power parameters of each energy input module are transmitted to the corresponding energy control module to adjust the energy output; S54: the medium state parameters after the adjustment of photothermal effect induced water phase regulation platform, the actual operation parameters of electrolytic hydrogen production module and the actual output parameters of each energy input module are collected, these parameters are compared with the preliminary operation optimization parameters, the parameter deviation is calculated, and the deviation information is fed back to step S4 for subsequent parameter iteration adjustment.

10. A real-time operation optimization system for a multi-energy complementary electrolytic hydrogen production system, characterized in that, Comprise: A multi-parameter collaborative acquisition unit is connected with each energy input module, electrolytic hydrogen production module and light-heat effect induced water phase regulation platform of the multi-energy complementary electrolytic hydrogen production system, used for acquiring real-time operation parameters of each module and constructing a system real-time operation parameter set; a structured knowledge driven prediction unit is connected with the multi-parameter collaborative acquisition unit, has a structured process knowledge guided neural network built-in, receives the system real-time operation parameter set output by the multi-parameter collaborative acquisition unit, predicts parameter variation trends in a future preset time period of the system and outputs a predicted parameter sequence; a distributed robust optimization modeling unit is connected with the structured knowledge driven prediction unit, uses a distributed robust chance-constrained programming algorithm, constructs a system operation optimization main model based on the predicted parameter sequence, and determines an uncertainty parameter fluctuation range and a constraint satisfaction probability threshold in the model; a multi-agent collaborative solving unit is connected with the distributed robust optimization modeling unit, decomposes the system operation optimization main model into sub-models, uses a multi-agent collaborative optimization algorithm based on ADMM to solve the sub-models, and outputs preliminary operation optimization parameters of each subsystem; a closed-loop parameter adjustment unit is connected with the multi-agent collaborative solving unit, electrolytic hydrogen production module and light-heat effect induced water phase regulation platform respectively, adjusts operation parameters of the electrolytic hydrogen production module and light-heat effect induced water phase regulation platform according to the preliminary operation optimization parameters, and feeds back parameter deviations; and an iterative optimization output unit is connected with the closed-loop parameter adjustment unit, receives parameter deviation information, controls the structured knowledge driven prediction unit, distributed robust optimization modeling unit, multi-agent collaborative solving unit and closed-loop parameter adjustment unit to repeatedly run, and outputs final optimization parameters when the parameter deviations meet the requirements.

Images