A SOEC-methanol system collaborative control method for a renewable energy scene

Abstract

The embodiment of the specification discloses a SOEC-methanol system collaborative control method for a renewable energy scene, the technical scheme of the application generates synthetic irradiance through dynamic sky classification, K-means clustering and the like, combines multiple data to construct an input matrix, and predicts photovoltaic power generation power through a double-layer LSTM neural network; then, the prediction is used as a disturbance, a model predictive control (MPC) algorithm is used, a thermal-electric coupling state space model is used as a basis, multiple operation constraints are considered, and control instructions are rollingly optimized; finally, the instructions are executed by a PID controller, SOEC-related equipment is adjusted, and the system state is fed back, so that closed-loop roll optimization is realized.

Description

Technical Field

[0001] This application relates to the field of new energy and chemical system integration technology, and in particular to a collaborative control method for SOEC-methanol systems in renewable energy scenarios. Background Technology

[0002] As the global energy transition progresses, the proportion of renewable energy sources such as solar power in the energy structure continues to increase. However, their significant volatility, intermittency, and unschedulable nature pose challenges to the stable operation of energy systems. Especially in scenarios with a high proportion of grid connection, the mismatch between power supply and load timing leads to a decline in energy utilization efficiency. To achieve grid connection, additional investment in energy storage, peak shaving, and other facilities is required, increasing the overall system cost. Against this backdrop, promoting the coupling of renewable energy with energy-intensive chemical processes, and transforming fluctuating green electricity into high-value-added products that are easy to store and transport, has become an important direction for the integrated development of energy and chemical industries. As clean fuels and chemical feedstocks, green hydrogen and green methanol have attracted much attention for their system integration and synergistic control technologies. Solid oxide electrolysis cells (SOECs) have become an important development direction for green hydrogen and syngas preparation due to their high electro-thermal synergistic efficiency and direct co-electrolysis to syngas production capability. They can be combined with downstream methanol synthesis reaction systems to construct an integrated "electrolysis-synthesis" green methanol production pathway.

[0003] However, in systems dominated by renewable energy sources, most existing technologies treat the SOEC unit and the methanol reactor as independent systems when enabling the SOEC co-electrolysis system and the methanol synthesis system to operate in synergy, lacking a cross-system synergistic control mechanism, resulting in low overall system operating efficiency. Summary of the Invention

[0004] This specification provides an embodiment of a collaborative control method for SOEC-methanol systems in renewable energy scenarios to solve at least one of the technical problems mentioned above.

[0005] To solve the above-mentioned technical problems, the embodiments in this specification are implemented as follows:

[0006] This invention provides a collaborative control method for SOEC-methanol systems in renewable energy scenarios, comprising:

[0007] S1. Using the trained Long Short-Term Memory (LSTM) network, predict the sequence of photovoltaic power generation values ​​for the next H steps after the current time t, denoted as d. t , Among them, symbols This represents the predicted photovoltaic power generation at the i-th time step after the current time t, and satisfies 1≤i≤H, where H represents the future time step of the photovoltaic power generation predicted by the LSTM model starting from the current time t; wherein, the trained Long Short-Term Memory (LSTM) network is used to predict the power generation of photovoltaic power plants based on the input synthetic meteorological data vector, historical weather forecast data vector, historical photovoltaic power generation data sequence, and time-encoding features;

[0008] S2, using the photovoltaic power generation prediction value sequence d t As a disturbance input, the Model Predictive Control (MPC) algorithm is used to construct a predictive model based on the thermo-electrical coupling state-space model of the SOEC co-electrolysis system and the methanol synthesis unit, generating the optimal control input sequence for the next N+1 steps. Specifically, only the control command of the current step in the control command sequence is selected. Output to the underlying actuator control system;

[0009] The discrete-time expression of the thermo-electric coupling state-space model of the SOEC co-electrolysis system and the methanol synthesis unit is as follows:

[0010]

[0011] Where, symbol x t This represents the state variable vector of the SOEC co-electrolysis system at time t. symbol x t+1 This represents the state variable vector of the SOEC co-electrolysis system at time t+1; symbol u t This represents the control input vector at time t. symbol ω t The modeling error is represented by the symbols A, B, and E, which represent the coefficient matrix; the state variable vector x at time t is... t Chinese symbol T t V represents the stack temperature of an SOEC system. t Indicates the voltage of a single fuel cell stack, symbol Represents the mole fraction of hydrogen, symbol The mole fraction of carbon dioxide is represented by the symbol P. an,t Indicates anode pressure, symbol P ca,t The cathode pressure is represented; the control input vector u at time t is... t Chinese symbol I t Represents electric current, symbol Q heater,t Indicates heating power, symbol Indicates hydrogen mass flow rate, symbol AFR represents the mass flow rate of carbon dioxide.t Represents the air-fuel ratio; the symbol C represents the output matrix, used to convert the state variable vector x of the SOEC system. t Mapped to the observable output vector y t ;

[0012] S3, The underlying actuator control system executes the control command for the current step. The system collects and feeds back real-time operating status data of the SOEC co-electrolysis system and methanol synthesis unit, including stack operating temperature, gas feed flow rate, reactor pressure, and syngas composition. This provides feedback information for the rolling optimization of model predictive control in the next control cycle, thereby performing rolling optimization of the control sequence for several future steps within a preset control cycle.

[0013] In some optional implementations, the synthetic meteorological data vector includes synthetic solar irradiance, and the method for generating the synthetic meteorological data vector includes the following steps:

[0014] Historical solar irradiance data for the area where the photovoltaic power station is located were collected and divided into data segments by season and hour. The K-means clustering algorithm was used to cluster the hourly irradiance data into 5 categories, corresponding to the NOAA sky type standards of sunny, partly cloudy, cloudy, partly cloudy, and overcast. The irradiance center point C of each category was calculated. k k = 1, 2, 3, 4, 5, which are the typical irradiance values ​​for this category;

[0015] Obtain the sky type of the city-level weather forecast at the current time t, match it to the clustering category in step 1, and extract the irradiance center point C of the corresponding category. k As a composite solar irradiance

[0016] Synthetic solar irradiance The data are spliced ​​together with temperature, humidity, and wind speed data collected at the same time to form a synthetic meteorological data vector containing quantitative meteorological parameters.

[0017] In some optional implementations, the training method of the Long Short-Term Memory (LSTM) network includes the following steps:

[0018] S11. Prepare training data, including:

[0019] The training dataset is obtained from a historical database. Each sample contains input features and output labels. The input features include synthetic meteorological data vectors, historical weather forecast data vectors, historical photovoltaic power generation data sequences, and time-coded features. The synthetic meteorological data vectors include numerical irradiance, temperature, wind speed, and relative humidity. The historical weather forecast data vectors include simulated forecast values ​​of irradiance, temperature, wind speed, and relative humidity at historical times. The time-coded features include hourly one-hot encoded vectors and seasonal encoded vectors.

[0020] The output label includes the actual sequence of future H-step photovoltaic power generation;

[0021] S12. Data preprocessing, specifically including:

[0022] S121. Numerical feature normalization, including:

[0023] Min-Max scaling is applied to continuous features in synthetic meteorological data vectors, historical weather forecast data vectors, and historical photovoltaic power generation data sequences, using the following formula: Map to the interval [0, 1];

[0024] S122. Category feature coding, including:

[0025] Convert the five sky types obtained by K-means clustering into one-hot vectors;

[0026] S13. Model training, including:

[0027] A two-layer cascaded LSTM network is constructed, with the first layer configured with 75 memory units and the second layer configured with 70 memory units. The network weights are optimized using the time backpropagation algorithm, and the loss function is the multi-step prediction mean squared error.

[0028]

[0029] Among them, symbols Represents the predicted value of photovoltaic power generation, symbol The actual value of photovoltaic power generation is represented by the symbol H, which represents the prediction step size. The Adam optimizer is used, and L2 regularization is introduced to suppress overfitting. The training cycle is no less than the preset number of times, and the batch size is set to the preset value.

[0030] S14. Rolling update mechanism, including:

[0031] At fixed intervals, the latest collected synthetic meteorological data, historical weather forecast data, photovoltaic power generation data, and time-coded features are integrated into the training set, and the model parameters are updated through incremental learning.

[0032] In some optional implementations, the optimization objective of the MPC algorithm is defined by the following objective function:

[0033]

[0034] Wherein, the symbol u t:t+N The symbol x represents the control input sequence for the next N control cycles. t+k The system state vector at time k in the future is represented by the symbol. Represents the desired system state trajectory, symbol The weighted norm represents the state deviation, and the symbol Q denotes the first weighting matrix used to weight the state vector x. t+k With the expected system state trajectory The deviations are weighted, and the signs are... The weighted norm of the control input is represented by the symbol R, which denotes the second weighting matrix used to weight the control input u. t+k Weighted sum of squares;

[0035] The constraints include: the stack temperature range is T. min ≤T t ≤T max The range of gas components is: The safe range for current and voltage is I. min ≤I t ≤I max V t ≤V safe Photovoltaic power generation power limit constraint is Air-fuel ratio constraint is AFR min ≤AFR t+k ≤AFR max ; where the symbol T t The symbol T represents the real-time operating condition of the SOEC co-electrolysis system's fuel cell stack at time t. min The symbol T represents the lowest temperature threshold allowed for stable operation of the SOEC co-electrolysis system. max This represents the highest temperature threshold allowed for stable operation of the SOEC co-electrolysis system; symbol The symbol represents the mole fraction of hydrogen in the feed of the SOEC co-electrolysis system at time t. Indicates the mole fraction of carbon dioxide in the feed of the SOEC co-electrolysis system at time t; symbol I t The symbol I represents the electrolytic current of the SOEC co-electrolysis system at time t. min The symbol I represents the safe lower limit of the electrolytic current of the SOEC co-electrolysis system at time t. max V represents the safe upper limit of the electrolytic current of the SOEC co-electrolysis system at time t.t V represents the total voltage of the SOEC co-electrolysis system at time t; symbol V safe The symbol Q represents the safe voltage threshold of the SOEC co-electrolysis system. heater,t The symbol represents the heater power of the stack in the SOEC co-electrolysis system at time t. AFR represents the predicted photovoltaic power generation at time t. min The minimum permissible value for air-fuel ratio, symbol AFR max This indicates the maximum permissible value for the air-fuel ratio.

[0036] In some optional implementations, the step of performing rolling optimization of the control sequence for several future steps with a preset control period specifically includes:

[0037] The underlying actuator control system executes the control command for the current step. Then, the operating status data is fed back to the MPC controller to achieve closed-loop rolling optimization for the next control cycle.

[0038] One embodiment of this specification can achieve at least the following beneficial effects: The technical solution of this application constructs a multi-layer control architecture of "prediction-optimization-execution", uses LSTM neural network to fuse multimodal inputs to realize forward prediction of the power generation of renewable energy such as photovoltaic power generation, and perceives the trend of energy fluctuations in advance; then, based on the SOEC thermo-electric coupling state space model, combined with the objective function including state deviation and control quantity changes, it coordinates the operation of SOEC co-electrolysis system and methanol synthesis unit, and coordinates the stack operation constraints (temperature, gas composition, current and voltage, etc.), thereby realizing dynamic matching of cross-system parameters and improving energy conversion efficiency; the bottom actuator accurately executes control commands through the controller and feeds back the system status in real time to form a closed loop, suppressing operation fluctuations, thereby improving the overall operating efficiency. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a schematic diagram illustrating the application scenario of the SOEC-methanol system collaborative control method for renewable energy scenarios provided by this invention;

[0041] Figure 2This is a flowchart of a collaborative control method for SOEC-methanol systems in renewable energy scenarios provided by the present invention. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of one or more embodiments of this specification clearer, the technical solutions of one or more embodiments of this specification will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of them. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of one or more embodiments of this specification.

[0043] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another.

[0044] The background technology section mentioned that "promoting the coupling of renewable energy with energy-intensive chemical processes, and transforming fluctuating green electricity into high-value-added products that are easy to store and transport, has become an important direction for the integrated development of energy and chemical industries." The output power of photovoltaic power generation is significantly affected by meteorological conditions (such as cloud cover and wind speed changes), exhibiting short-term drastic fluctuations and long-term unpredictability (such as sudden weather changes). If the system relies solely on real-time power signal regulation, the control lag will lead to the following adverse consequences: the SOEC stack current may rise and fall sharply, causing drastic fluctuations in the temperature field, accelerating stack aging or even damage; or it may cause fluctuations in synthesizer output and composition (H2 / CO), resulting in decreased efficiency of the methanol synthesis reactor and substandard product purity.

[0045] Existing technologies often treat SOEC and methanol synthesis as independent systems, using simple controllers such as PID controllers to adjust single variables (e.g., current, temperature) in real time. This approach suffers from drawbacks such as response lag, cross-system mismatch, and energy efficiency losses. Specifically, regarding response lag, the system adjusts parameters only after power fluctuations, leading to large deviations in stack temperature and significant fluctuations in syngas composition. Regarding cross-system mismatch, SOEC gas production changes become disconnected from methanol synthesis requirements; for example, a sudden increase in stack gas production can cause a decrease in synthesis efficiency in the methanol reactor due to a lack of prior temperature / pressure adjustments. Finally, regarding energy efficiency losses, frequent start-ups and shutdowns or significant parameter adjustments can cause the system to deviate from its optimal operating point over a long period, resulting in overall energy efficiency reduction.

[0046] This invention provides a coordinated control method for an SOEC-methanol system in renewable energy scenarios. It can dynamically adjust the coordinated operation between the SOEC electrolytic reactor, syngas production unit, and methanol synthesis reactor under conditions of high volatility and unstable supply and demand in renewable energy, thereby achieving efficient and stable operation of the entire system. The technical solution of this application is described below with reference to the accompanying drawings. Figure 1 This is a schematic diagram illustrating the application scenario of the SOEC-methanol system collaborative control method for renewable energy scenarios provided by this invention. Figure 2 This is a flowchart of the SOEC-methanol system coordinated control method for renewable energy scenarios provided by the present invention. (Refer to...) Figure 1 and Figure 2 The method may include the following steps:

[0047] Step S1: Use the trained Long Short-Term Memory (LSTM) network to predict the sequence of photovoltaic power generation values ​​H steps after the current time t, denoted as d. t , Among them, symbols This represents the predicted photovoltaic power generation at the i-th time step after the current time t, and satisfies 1≤i≤H, where H represents the future time step of the photovoltaic power generation predicted by the LSTM model starting from the current time t; wherein, the trained Long Short-Term Memory (LSTM) network is used to predict the power generation of photovoltaic power plants based on the input synthetic meteorological data vector, historical weather forecast data vector, historical photovoltaic power generation data sequence, and time-encoding features.

[0048] The main purpose of this step is to use a pre-trained LSTM network to output a sequence of predicted photovoltaic power generation values ​​for the next H steps after the current time t. This provides perturbation input for the subsequent "rolling optimization" of MPC. The meaning of each data input to the LSTM network is explained below.

[0049] The synthetic meteorological data vector refers to a sequence of quantitative meteorological parameters for the future time interval t→t+H. Core parameters may include temperature, humidity, wind speed, and synthetic quantitative solar irradiance for the future time interval t→t+H. Common weather terms such as temperature, humidity, and wind speed will not be explained here; the meaning of the synthetic quantitative solar irradiance data will be explained in detail below. Since some data in public weather forecasts are qualitative descriptions (such as "sunny to cloudy"), they cannot be directly input into the LSTM model. To resolve the contradiction between "uncertainty of future weather" and "the need for quantitative data in model input," the synthetic solar irradiance is data that maps qualitative sky type data such as "sunny to cloudy" from "semantic" to "numerical" based on statistical laws. For example, mapping "sunny to cloudy" to 600W / m².2 Or map "sunny" to 800W / m 2 Historical weather forecast data vectors can refer to meteorological data for specific time periods over a number of past periods, including data such as irradiance, temperature, wind speed, and humidity. Historical photovoltaic power generation data sequences can refer to the output power data of photovoltaic power plants over consecutive past periods, arranged chronologically as a vector (e.g., P∈R). T′1 This is used to enable the LSTM model to capture the temporal dependencies of power (such as ramp-up trends and fluctuation cycles). Temporal encoded features can refer to "hourly + seasonal" features represented by one-hot encoding. This involves transforming "time attributes" into numerical vectors through structured encoding, serving as one of the input features of the LSTM model. This explicitly injects "time-periodic prior knowledge," addressing the periodic fluctuations in photovoltaic power caused by diurnal and seasonal variations. It transforms "temporal semantics" into "computable numerical features," assisting the LSTM model in learning the power variation patterns over time.

[0050] Step S2: Using the photovoltaic power generation prediction value sequence d t As a disturbance input, the Model Predictive Control (MPC) algorithm is used to construct a predictive model based on the thermo-electrical coupling state-space model of the SOEC co-electrolysis system and the methanol synthesis unit, generating the optimal control input sequence for the next N+1 steps. Specifically, only the control command of the current step in the control command sequence is selected. Output to the underlying actuator control system;

[0051] The discrete-time expression of the thermo-electric coupling state-space model of the SOEC co-electrolysis system and the methanol synthesis unit is as follows:

[0052]

[0053] Where, symbol x t Let x represent the state variable vector of the SOEC co-electrolysis system at time t. t = symbol x t+1 This represents the state variable vector of the SOEC co-electrolysis system at time t+1; symbol u t This represents the control input vector at time t. symbol ω t The modeling error is represented by the symbols A, B, and E, which represent the coefficient matrix; the state variable vector x at time t is... t Chinese symbol T t V represents the stack temperature of an SOEC system. t Indicates the voltage of a single fuel cell stack, symbol Represents the mole fraction of hydrogen, symbol The mole fraction of carbon dioxide is represented by the symbol P. an,t Indicates anode pressure, symbol P ca,t The cathode pressure is represented; the control input vector u at time t is... t Chinese symbol I t Represents electric current, symbol Q heater,t Indicates heating power, symbol Indicates hydrogen mass flow rate, symbol AFR represents the mass flow rate of carbon dioxide. t Represents the air-fuel ratio; the symbol C represents the output matrix, used to convert the state variable vector x of the SOEC system. t Mapped to the observable output vector y t .

[0054] The purpose of this step is to achieve coordinated optimization of the SOEC co-electrolysis system and the methanol synthesis system based on the model predictive control algorithm (MPC) and the photovoltaic power generation prediction sequence. Through "future photovoltaic power prediction → multi-module dynamic modeling → rolling optimization decision", the global balance between energy supply and demand and equipment constraints is achieved.

[0055] As previously mentioned, in systems primarily powered by renewable energy, most existing technologies treat the SOEC unit and methanol reactor as independent systems when implementing coordinated operation of SOEC co-electrolysis and methanol synthesis systems. This lack of cross-system coordinated control mechanisms leads to low overall system efficiency. To address this issue, this application's technical solution constructs a predictive model based on a thermo-electrical coupling state-space model of the SOEC co-electrolysis system and methanol synthesis unit. This state-space model describes the thermo-electrical and mass flow coupling relationships between the SOEC co-electrolysis and methanol synthesis units using discrete-time equations. Its discrete-time expression is as follows:

[0056]

[0057] The meanings of the symbols in this expression have been explained in detail above. The characteristics of each variable will now be introduced. Among them, the symbol x... t This represents the state variable vector of the SOEC co-electrolysis system at time t. This vector describes the dynamic state inside the system (which cannot be directly controlled and requires monitoring). The control input vector is u. t As an actively adjustable operating variable, from the perspective of SOEC co-electrolysis system and methanol synthesis system, the disturbance input d t As an uncontrollable disturbance, it drives the dynamic changes of the system. Output vector y t It is the measurable system response.

[0058] In this step, Model Predictive Control (MPC) is a model-based rolling time-domain optimization control strategy. Its core idea is to establish a dynamic model of the system, optimize the control sequence for a future period within each control cycle, and execute only the control command for the current moment. In the technical solution of this application, MPC serves as the mid-level control core, connecting the upper-level photovoltaic prediction and the lower-level execution equipment to achieve cross-scale collaborative optimization of the SOEC co-electrolysis system and the methanol synthesis system.

[0059] Symbol A represents the state transition matrix, describing state x. t The natural evolution of the SOEC stack temperature (e.g., the decay of SOEC stack temperature over time due to heat dissipation, and the natural fluctuations in methanol synthesis tower pressure). Control matrix B is used to quantify u. t For x t The effects (e.g., a 10A increase in current, a 5°C / minute rise in stack temperature) are mapped by the output map C to represent the internal state x. t Transform into observable y t (e.g., stack temperature → temperature sensor reading).

[0060] In the state transition model, the output matrix C is used to represent the system's state variables x. t Mapped to the observable output variable y t , that is, y t =Cx t Output variable y t It is the state variable x t A subset of, i.e. Select key states in the system that can be directly measured or observed (such as stack temperature, methanol production, gas composition, etc.). This is achieved by controlling the output variable y. t The accuracy of the state transition model is verified by comparing it with actual measured values. For example, the actual stack temperature. Compared with model predictions (symbol C) T Compare the row vectors corresponding to the temperatures in C, and calculate the modeling error ω. t Update the model parameters (such as A, B, E) to improve prediction accuracy. (The last part, "y", appears to be a typo and can be omitted.) t As a feedback signal, the control input u is adjusted in real time. t . Assuming we are interested in the stack temperature T t and hydrogen components The output matrix C is then... The first row of this matrix is ​​used to extract temperature T. t The second row of this matrix is ​​used to extract x. t The third element, namely hydrogen.

[0061] Step S3: The underlying actuator control system executes the control command for the current step. The system collects and feeds back real-time operating status data of the SOEC co-electrolysis system and methanol synthesis unit, including stack operating temperature, gas feed flow rate, reactor pressure, and syngas composition. This provides feedback information for the rolling optimization of model predictive control in the next control cycle, thereby performing rolling optimization of the control sequence for several future steps within a preset control cycle.

[0062] In this step, the rolling optimization of the control sequence for several future steps with a preset control period may specifically include:

[0063] The underlying actuator control system executes the control command for the current step. Then, the operating status data is fed back to the MPC controller to achieve closed-loop rolling optimization for the next control cycle.

[0064] In this embodiment, the underlying actuator control system first executes the control command for the current step. Subsequently, high-frequency real-time acquisition of operational status data from the SOEC co-electrolysis system and methanol synthesis unit is used. Specifically, the stack operating temperature can be monitored using high-precision thermocouples, reflecting electrolysis efficiency and equipment lifespan; gas feed flow rate can be acquired using a mass flow meter, a parameter crucial for the stability of the H2 / CO ratio in the syngas; reactor pressure can be obtained using a pressure transmitter; and syngas components can be analyzed using an online gas chromatograph. The acquired operational status data is fed back to the MPC controller. Based on a state-space model, the MPC corrects the deviation between the "model-predicted state" and the "actual acquired state" (e.g., if the actual stack temperature is higher than the predicted value, the auxiliary heating power or electrolysis current in subsequent control cycles is reduced). It then recalculates the optimal control sequence for the next N+1 steps, achieving a closed-loop rolling iteration of "execution → feedback → correction → re-optimization." This continuously offsets interference from photovoltaic power fluctuations, equipment parameter drift, and model mismatch, ensuring optimal system operation under multiple constraints.

[0065] The technical solution of this application constructs a multi-layered control architecture of "prediction-optimization-execution". It uses an LSTM neural network to fuse multimodal inputs to achieve forward prediction of the power generation of renewable energy such as photovoltaic power generation, and perceives the trend of energy fluctuations in advance. Then, based on the SOEC thermo-electric coupling state-space model, combined with an objective function that includes state deviations and changes in control quantities, it coordinates the operation of SOEC co-electrolysis system and methanol synthesis unit, and takes into account the stack operation constraints (temperature, gas composition, current, voltage, etc.), thereby achieving dynamic matching of cross-system parameters and improving energy conversion efficiency. The bottom actuator accurately executes control commands through the controller and feeds back the system status in real time to form a closed loop, suppressing operational fluctuations and thus improving overall operating efficiency.

[0066] Based on the foregoing technical solutions, this invention also provides some more specific technical solutions, which are described below.

[0067] In an optional embodiment, the synthetic meteorological data vector includes synthetic solar irradiance, and the method for generating the synthetic meteorological data vector may include the following steps:

[0068] Historical solar irradiance data for the area where the photovoltaic power station is located were collected and divided into data segments by season and hour. The K-means clustering algorithm was used to cluster the hourly irradiance data into 5 categories, corresponding to the NOAA sky type standards of sunny, partly cloudy, cloudy, partly cloudy, and overcast. The irradiance center point C of each category was calculated. k k = 1, 2, 3, 4, 5, which are the typical irradiance values ​​for this category;

[0069] Obtain the sky type of the city-level weather forecast at the current time t, match it to the corresponding cluster category, and extract the irradiance center point C of the corresponding category. k As a composite solar irradiance

[0070] Synthetic solar irradiance The data are spliced ​​together with temperature, humidity, and wind speed data collected at the same time to form a synthetic meteorological data vector containing quantitative meteorological parameters.

[0071] In an optional embodiment, the training method for the Long Short-Term Memory (LSTM) network may include the following steps:

[0072] Step S11: Prepare training data, which may include:

[0073] The training dataset is obtained from a historical database. Each sample contains input features and output labels. The input features include synthetic meteorological data vectors, historical weather forecast data vectors, historical photovoltaic power generation data sequences, and time-coded features. The synthetic meteorological data vectors include numerical irradiance, temperature, wind speed, and relative humidity. The historical weather forecast data vectors include simulated forecast values ​​of irradiance, temperature, wind speed, and relative humidity at historical times. The time-coded features include hourly one-hot encoded vectors and seasonal encoded vectors.

[0074] The output label includes the actual sequence of future H-step photovoltaic power generation;

[0075] Step S12: Data preprocessing, which may include the following sub-steps:

[0076] Step S121: Numerical feature normalization, which may include:

[0077] Min-Max scaling is applied to continuous features in synthetic meteorological data vectors, historical weather forecast data vectors, and historical photovoltaic power generation data sequences, using the following formula: Map to the interval [0, 1];

[0078] Step S122: Category feature encoding, which may include:

[0079] Convert the five sky types obtained by K-means clustering into one-hot vectors;

[0080] Step S13, model training, may include:

[0081] A two-layer cascaded LSTM network is constructed, with the first layer configured with 75 memory units and the second layer configured with 70 memory units. The network weights are optimized using the time backpropagation algorithm, and the loss function is the multi-step prediction mean squared error.

[0082]

[0083] Among them, symbols Represents the predicted value of photovoltaic power generation, symbol The actual value of photovoltaic power generation is represented by the symbol H, which represents the prediction step size. The Adam optimizer is used, and L2 regularization is introduced to suppress overfitting. The training cycle is no less than the preset number of times, and the batch size is set to the preset value.

[0084] S14. Rolling update mechanism, including:

[0085] At fixed intervals, the latest collected synthetic meteorological data, historical weather forecast data, photovoltaic power generation data, and time-coded features are integrated into the training set, and the model parameters are updated through incremental learning.

[0086] In an optional embodiment, the optimization objective of the MPC algorithm is defined by the following objective function:

[0087]

[0088] Wherein, the symbol u t:t+N The symbol x represents the control input sequence for the next N control cycles. t+k The system state vector at time k in the future is represented by the symbol. Represents the desired system state trajectory, symbol The weighted norm represents the state deviation, and the symbol Q denotes the first weighting matrix used to weight the state vector x. t+k With the expected system state trajectory The deviations are weighted, and the signs are... The weighted norm of the control input is represented by the symbol R, which denotes the second weighting matrix used to weight the control input u.t+k Weighted sum of squares;

[0089] The constraints include: the stack temperature range is T. min ≤T t ≤T max The range of gas components is: The safe range for current and voltage is I. min ≤I t ≤I max V t ≤V safe symbol V safe This represents the voltage safety threshold; the photovoltaic power generation limit constraint is... Air-fuel ratio constraint is AFR min ≤AFR t+k ≤AFR max ; where the symbol T t The symbol T represents the real-time operating condition of the SOEC co-electrolysis system's fuel cell stack at time t. min The symbol T represents the lowest temperature threshold allowed for stable operation of the SOEC co-electrolysis system. max This represents the highest temperature threshold allowed for stable operation of the SOEC co-electrolysis system; symbol The symbol represents the mole fraction of hydrogen in the feed of the SOEC co-electrolysis system at time t. Indicates the mole fraction of carbon dioxide in the feed of the SOEC co-electrolysis system at time t; symbol I t The symbol I represents the electrolytic current of the SOEC co-electrolysis system at time t. min The symbol I represents the safe lower limit of the electrolytic current of the SOEC co-electrolysis system at time t. max V represents the safe upper limit of the electrolytic current of the SOEC co-electrolysis system at time t. t V represents the total voltage of the SOEC co-electrolysis system at time t; symbol V safe The symbol Q represents the safe voltage threshold of the SOEC co-electrolysis system. heater,t The symbol represents the heater power of the stack in the SOEC co-electrolysis system at time t. AFR represents the predicted photovoltaic power generation at time t. min The minimum permissible value for air-fuel ratio, symbol AFR max This indicates the maximum permissible value for the air-fuel ratio.

[0090] In this embodiment, the optimization objective of the MPC algorithm is defined by the following objective function.

[0091]

[0092] The design objective of this objective function is explained below. This objective function can be broken down into: and Two items, of which, The design goal of this section is to make the system state approach the reference value. By penalizing the "deviation between the actual state and the reference state", the system state (such as the stack temperature and the syngas ratio) is forced to be as close as possible to the optimal operating point. For example, if the stack temperature deviates from 800°C (reference value), the weight of the temperature dimension in the Q matrix will amplify the deviation penalty, drive the control input adjustment (such as adjusting the heating power), and restore the temperature to the high-efficiency range. If the H2 / CO ratio deviates from 2.0 (reference value), the Q matrix will penalize the deviation, forcing the SOEC current or feed valve to be adjusted to maintain the syngas quality. The design goal of this section is to control input smoothness by penalizing the "amplitude / rate of change of the control input" to prevent drastic changes in the device's operating conditions. For example, if the SOEC current frequently changes... When the current fluctuates, the R matrix will amplify the control cost, forcing the MPC to choose a smoother current adjustment strategy (such as gradually increasing from 60A to 75A) to reduce electrode life loss; if the methanol feed valve opening changes abruptly, the R matrix penalty will drive the MPC to adopt a "small step, multi-cycle" adjustment method to avoid catalyst deactivation caused by syngas flow shock.

[0093] By solving this quadratic programming problem, the optimal control input sequence for the next N+1 steps can be obtained: Each For a control command at a certain moment, although the technical solution of this application can solve for the control sequence of multiple future steps, the actual control quantity executed at the current step is different. Subsequent steps (such as) As a pre-aiming command, in the next control cycle, the system will adjust the target based on the new state x. t+1 Using the updated prediction data, the optimization problem is resolved to generate a new control sequence. This "rolling optimization + first-step execution only" mechanism enables dynamic closed-loop updates, avoiding error accumulation. Thus, "rolling optimization" ensures that the system is based on the latest state x in each control cycle. t+1 (such as actual stack temperature, real-time syngas composition) and updated prediction data (the photovoltaic power sequence d corrected in step S1). t (Based on synthetic meteorological data), the control sequence for the next N+1 steps is recalculated, rather than relying on the static prediction at the initial moment. If the system state changes due to equipment aging (such as increased internal resistance of SOEC stacks) or disturbances (such as photovoltaic power prediction errors), the new state feedback will correct the model prediction bias, making the control sequence adapt to the current system characteristics. On the other hand, only the first step of control is executed. Subsequent aiming commands (such as) Instead of executing the commands directly, the system re-optimizes them in the next cycle. Even if the initial predicted future operating conditions (such as photovoltaic power and weather conditions) deviate from the actual conditions (such as a sudden drop in irradiance due to premature cloud cover), the rolling optimization in the next cycle will adjust subsequent control decisions based on the new information, avoiding cascading errors caused by "executing multiple steps according to incorrect pre-targeting instructions" (such as an incorrect increase in SOEC current leading to reactor temperature runaway). By "periodically updating the decision basis + executing only the immediate optimal instructions," this mechanism continuously offsets the effects of time-varying system states (equipment aging, parameter drift) and prediction errors (photovoltaic power, weather uncertainty), ensuring that the control strategy always fits the real-time operating conditions, thereby improving the system's robustness against dynamic disturbances.

[0094] In this embodiment, the control quantity values ​​obtained through solving can directly guide the adjustment of operating parameters of the SOEC and methanol synthesis system, for example:

[0095]

[0096] Solving this formula yields an "optimal control command sequence for future prediction," transforming the "uncertainty" of photovoltaic fluctuations into the "schedulability" of control quantities. This coordinates the cross-stage operation of SOEC and methanol synthesis systems (such as the synchronous adjustment of stack current and syngas flow rate), adapting to the coupling characteristics of the system.

[0097] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of one embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing the present invention.

[0098] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A collaborative control method for SOEC-methanol systems in renewable energy scenarios, characterized in that, The method includes: S1. Predict the current time using the trained Long Short-Term Memory (LSTM) network. After the future The sequence of predicted photovoltaic power generation values ​​for step is denoted as Among them, symbols Relative to the current time After that, the The predicted photovoltaic power generation at each time step, and satisfying ,symbol Indicates from the current time The initial step is the number of future time steps of the photovoltaic power generation predicted by the LSTM model; wherein, the trained Long Short-Term Memory (LSTM) network is used to predict the power generation of the photovoltaic power station based on the input synthetic meteorological data vector, historical weather forecast data vector, historical photovoltaic power generation data sequence and time coding features. S2, using the photovoltaic power generation prediction value sequence As a disturbance input, the Model Predictive Control (MPC) algorithm is used to construct a predictive model based on the thermo-electrical coupling state-space model of the SOEC co-electrolysis system and the methanol synthesis unit, generating the future... Optimal control input sequence of steps ; wherein, only the control command of the current step in the optimal control input sequence is used. Output to the underlying actuator control system; The discrete-time expression of the thermo-electric coupling state-space model of the SOEC co-electrolysis system and the methanol synthesis unit is as follows: ； Among them, symbols express The state variable vector of the SOEC co-electrolysis system at the given time, ;symbol This indicates that the SOEC co-electrolysis system is in The state variable vector at each moment; symbol express Control the input vector at all times. ;symbol Indicates modeling error; symbol , , Represents the coefficient matrix; the Time-state variable vector Chinese symbol Indicates the stack temperature of the SOEC system, symbol Indicates the voltage of a single fuel cell stack, symbol Represents the mole fraction of hydrogen, symbol Represents the mole fraction of carbon dioxide, symbol Indicates anode pressure, symbol Indicates cathode pressure; the Control input vector at any time Chinese symbol Represents electric current, symbol Indicates heating power, symbol Indicates hydrogen mass flow rate, symbol Represents carbon dioxide mass flow rate, symbol Indicates air-fuel ratio; symbol The output matrix represents the vector of state variables of the SOEC system. Mapped to the observable output vector ; S3, The underlying actuator control system executes the control command for the current step. The system collects and feeds back real-time operating status data of the SOEC co-electrolysis system and methanol synthesis unit, including stack operating temperature, gas feed flow rate, reactor pressure, and syngas composition, to provide feedback information for the model predictive control rolling optimization of the next control cycle, thereby performing rolling optimization of the control sequence for several future steps with a preset control cycle. The optimization objective of the MPC algorithm in step S2 is defined by the following objective function: ； Among them, symbols Indicates the future The control input sequence for each control cycle, symbol Indicates the future The system state vector at time t, symbol Represents the desired system state trajectory, symbol The weighted norm representing the state deviation, with the sign... This represents the first weighting matrix, used to weight the state vector. With the expected system state trajectory The deviations are weighted, and the signs are... The weighted norm of the control input is represented by the symbol. This represents the second weighting matrix, used for the control input. The sum of squares is weighted.

2. The SOEC-methanol system collaborative control method for renewable energy scenarios according to claim 1, characterized in that, The synthetic meteorological data vector includes synthetic solar irradiance, and the method for generating the synthetic meteorological data vector includes the following steps: Historical solar irradiance data for the area where the photovoltaic power station is located were collected and divided into data segments by season and hour. The K-means clustering algorithm was used to cluster the hourly irradiance data into five categories, corresponding to the NOAA sky type standards: sunny, partly cloudy, cloudy, partly cloudy, and overcast. The irradiance center point of each category was calculated. , This serves as a typical irradiance value for this category. Get the current time The city-level weather forecast sky type is matched to the corresponding cluster category, and the irradiance center point of the corresponding category is extracted. As a composite solar irradiance ; Synthetic solar irradiance The data are spliced ​​together with temperature, humidity, and wind speed data collected at the same time to form a synthetic meteorological data vector containing quantitative meteorological parameters.

3. The SOEC-methanol system collaborative control method for renewable energy scenarios according to claim 2, characterized in that, The training method for the Long Short-Term Memory (LSTM) network includes the following steps: S11. Prepare training data, including: The training dataset is obtained from a historical database. Each sample contains input features and output labels. The input features include synthetic meteorological data vectors, historical weather forecast data vectors, historical photovoltaic power generation data sequences, and time-coded features. The synthetic meteorological data vectors include numerical irradiance, temperature, wind speed, and relative humidity. The historical weather forecast data vectors include simulated forecast values ​​of irradiance, temperature, wind speed, and relative humidity at historical times. The time-coded features include hourly one-hot encoded vectors and seasonal encoded vectors. The output label includes the actual sequence of future H-step photovoltaic power generation; S12. Data preprocessing, specifically including: S121. Numerical feature normalization, including: Min-Max scaling is applied to continuous features in synthetic meteorological data vectors, historical weather forecast data vectors, and historical photovoltaic power generation data sequences, using the following formula: Mapped to interval; S122. Category feature coding, including: Convert the five sky types obtained by K-means clustering into one-hot vectors; S13. Model training, including: A two-layer cascaded LSTM network is constructed, with the first layer configured with 75 memory units and the second layer configured with 70 memory units. The network weights are optimized using the time backpropagation algorithm, and the loss function is the multi-step prediction mean squared error. ； ； Among them, symbols Represents the predicted value of photovoltaic power generation, symbol Represents the actual value of photovoltaic power generation, symbol This indicates the prediction step size; the Adam optimizer is used, L2 regularization is introduced to suppress overfitting, the training cycle is no less than the preset number of times, and the batch size is set to the preset value; S14. Rolling update mechanism, including: At fixed intervals, the latest collected synthetic meteorological data, historical weather forecast data, photovoltaic power generation data, and time-coded features are integrated into the training set, and the model parameters are updated through incremental learning.

4. The SOEC-methanol system collaborative control method for renewable energy scenarios according to claim 1, characterized in that, The constraints of the MPC algorithm in step S2 include: the stack temperature range is... The range of gas components is: The safe range for current and voltage is: , Photovoltaic power generation power limit constraint is Air-fuel ratio constraint is Among them, symbols express At any given time, the real-time operation of the stack in the SOEC co-electrolysis system is discussed, with the symbol... This represents the lowest permissible temperature threshold for stable operation of the SOEC co-electrolysis system, denoted by [symbol]. This represents the highest temperature threshold allowed for stable operation of the SOEC co-electrolysis system; symbol Indicates time The mole fraction of hydrogen in the feed of the SOEC co-electrolysis system, denoted by [symbol]. Indicates time The mole fraction of carbon dioxide in the feed of the SOEC co-electrolysis system; symbol Indicates time The electrolytic current of the SOEC co-electrolysis system is denoted by [symbol]. Indicates time The safe lower limit of the electrolytic current of the SOEC co-electrolysis system, denoted by [symbol missing]. Indicates time The safe upper limit of the electrolytic current of the SOEC co-electrolysis system is given by the symbol [symbol missing]. Indicates time The total voltage of the SOEC co-electrolysis system at that time; symbol The symbol represents the safe threshold voltage of the SOEC co-electrolysis system. Indicates time The heater power of the stack in the SOEC co-electrolysis system; symbol Indicates time Predicted photovoltaic power generation, symbol The minimum permissible value for air-fuel ratio, symbol This indicates the maximum permissible value for the air-fuel ratio.

5. The SOEC-methanol system collaborative control method for renewable energy scenarios according to claim 1, characterized in that, The rolling optimization of the control sequence for several future steps with a preset control period specifically includes: The underlying actuator control system executes the control command for the current step. Then, the operating status data is fed back to the MPC controller to achieve closed-loop rolling optimization for the next control cycle.

Images