A method for photoelectric hydrogen production and methanol production

Abstract

This invention relates to a method for photoelectric hydrogen production and methanol production. The method collects operational data from a photoelectric hydrogen production unit, a hydrogen buffer unit, a carbon dioxide supply unit, a methanol synthesis unit, and a circulating separation unit to construct a system state vector and a disturbance vector. Based on material conservation, energy conservation, and the hydrogen-carbon ratio, a cross-unit causal graph is constructed to extract the causal effects of disturbances on hydrogen production flow rate, hydrogen supply capacity, and methanol yield. A mechanism-constrained time-series prediction network is used to obtain hydrogen production flow rate, hydrogen purity, methanol yield, overall energy consumption, carbon dioxide utilization rate, and prediction uncertainty. Based on the prediction results, a multi-objective optimization model with opportunity constraints is constructed, and a two-layer rolling solution is used to generate control actions. Under event-triggered conditions, control reconstruction, safety projection, and counterfactual online correction are performed to achieve integrated adaptive and coordinated control of the photoelectric hydrogen production and methanol production processes. This method can improve methanol yield and carbon dioxide utilization rate, and reduce overall energy consumption and operational risks.

Description

Technical Field

[0001] This invention relates to the fields of green chemical engineering and intelligent control technology, and in particular to a photoelectric hydrogen-to-methanol production method for operating conditions with fluctuating light energy. Specifically, it relates to an integrated collaborative control method for photoelectric hydrogen-to-methanol production based on cross-unit causal modeling, mechanism-constrained time-series prediction, sub-Bruker optimization, event-triggered reconstruction, and counterfactual online correction. Background Technology

[0002] Against the backdrop of the "dual carbon" goals and the high proportion of renewable energy integration, using solar energy to drive hydrogen production and further combining it with the resource utilization of carbon dioxide to produce methanol has become an important technological route for green fuels and low-carbon chemicals. Methanol has advantages such as convenient liquid storage and transportation, high energy density, and strong adaptability to chemical feedstocks. Coupled photovoltaic hydrogen production with methanol synthesis, it can transform fluctuating renewable energy into chemicals that are easy to store and transport, which is of great significance for improving the capacity for new energy absorption and realizing the high-value utilization of carbon dioxide.

[0003] In existing technologies, photoelectric hydrogen production and methanol synthesis are typically designed and controlled as relatively independent process units. Photoelectric hydrogen production often employs fixed operating point control, empirical rule control, or unit-level adjustments targeting local voltage, current, and temperature parameters. Methanol synthesis, on the other hand, focuses on steady-state optimization or conventional feedback control around variables such as reactor temperature, pressure, feed ratio, and recycle ratio. While these approaches can maintain the operation of each subsystem to some extent, the lack of unified modeling of the dynamic coupling relationships between hydrogen production, hydrogen storage, carbon dioxide supply, methanol synthesis, and recycle separation makes it difficult to achieve coordinated matching between hydrogen production, buffering, synthesis, and reflux at the system level.

[0004] Furthermore, the photovoltaic hydrogen-to-methanol system exhibits significant characteristics of multi-timescale, strong coupling, strong nonlinearity, and disturbance uncertainty. On the one hand, external conditions such as solar irradiance, spectral composition, and ambient temperature change rapidly over time, leading to significant fluctuations in hydrogen production capacity. On the other hand, the methanol synthesis unit is affected by factors such as temperature, pressure, hydrogen-to-carbon ratio, catalyst activity, and tail gas composition, resulting in problems such as high thermal inertia, slow state changes, and complex constraint boundaries. If only conventional PID control, empirical threshold control, or predictive control that does not consider uncertainties is used, it is often difficult to simultaneously achieve methanol yield, overall energy consumption, carbon dioxide utilization rate, operational stability, and equipment safety boundaries.

[0005] Some existing improvement schemes attempt to introduce data-driven models, model predictive control, or digital twin technology, but most only optimize a single device or a single objective, and generally still have the following shortcomings: First, there is a lack of structured modeling methods that can characterize the causal transmission paths and coupling directions between units, making it difficult to distinguish between the direct and indirect effects of disturbances on system performance; Second, there is a lack of bibliometric optimization and probabilistic safety constraint mechanisms for strong fluctuation conditions and model mismatch conditions, resulting in insufficient adaptability of control strategies to extreme conditions; Third, there is a lack of ability to synchronously reconstruct the model, constraints, and control sequences when events such as sudden changes in illumination, out-of-bounds raw material composition, accelerated catalyst deactivation, and reactor hotspot formation occur; Fourth, there is a lack of mechanisms to use counterfactual samples to correct control decisions online, making it difficult for models and cost parameters to adapt in a timely manner to equipment aging and long-term operating environment changes.

[0006] Therefore, there is an urgent need for a photoelectric hydrogen-to-methanol method and system that can uniformly model, predict, optimize, and adaptively update the photoelectric hydrogen production end, hydrogen buffer end, and methanol synthesis end under conditions of light disturbance, raw material fluctuation, and equipment time-varying degradation, so as to improve methanol yield, carbon dioxide utilization rate, and system operation stability, and reduce overall energy consumption and safety risks. Summary of the Invention

[0007] The purpose of this invention is to provide a photoelectric hydrogen production and methanol production method to solve the problems in the prior art, such as the disconnect between the photoelectric hydrogen production end and the methanol synthesis end, insufficient characterization of the dynamic coupling relationship, poor adaptability to fluctuating operating conditions and equipment degradation, and difficulty in balancing yield, energy consumption, carbon utilization and operational safety. This invention achieves integrated, adaptive and robust collaborative control of the entire process of photoelectric hydrogen production, hydrogen buffering and methanol synthesis.

[0008] To achieve the above-mentioned objectives, this invention provides a method for photoelectric hydrogen production and methanol production, characterized by being executed by a processor and including: constructing a system state vector based on the operating data of the photoelectric hydrogen production unit, hydrogen buffer unit, carbon dioxide supply unit, methanol synthesis unit, and circulating separation unit. With disturbance vector A cross-unit causal graph was constructed based on the laws of material conservation, energy conservation, and the hydrogen-carbon ratio. Extract the causal effect characteristics of the perturbation variables on hydrogen production flow rate, hydrogen supply capacity, and methanol yield; and convert the system state vector... Perturbation vector The causal effect features are input into the mechanism-constrained time-series prediction network to obtain future... Within a given step, the predicted values ​​for hydrogen production flow rate, hydrogen purity, methanol yield, unit methanol comprehensive energy consumption, carbon dioxide utilization rate, and prediction uncertainty are calculated. Based on these prediction results, a multi-objective optimization model with opportunity constraints is constructed, and a two-layer rolling solution is used to generate control actions. When the comprehensive event trigger index exceeds a threshold, the causal graph parameters, prediction model parameters, and optimization cost parameters are updated synchronously, and candidate control actions are projected to the safe and feasible region before execution. Based on the actual output after execution and the counterfactual output generated by the candidate actions under the same state and disturbance conditions, the prediction model and optimization cost parameters are corrected online. This process is repeated cyclically to achieve integrated adaptive and coordinated control of the photoelectric hydrogen production and methanol production processes.

[0009] Compared with the prior art, the present invention has at least the following beneficial effects:

[0010] 1. This invention constructs a cross-unit causal graph model, unifying the photoelectric hydrogen production end, the hydrogen buffer end, and the methanol synthesis end into the same collaborative control framework. This enables a more accurate characterization of the dynamic coupling relationship and disturbance transmission path between each process unit, thereby improving the accuracy and consistency of system-level control decisions.

[0011] 2. By introducing causal effect information into the mechanism-constrained time-series prediction network, this invention can not only achieve multi-step prediction of hydrogen production, methanol yield, comprehensive energy consumption and carbon utilization, but also take into account the physical interpretability of the prediction results and the consistency of process constraints, thereby improving the reliability of prediction under complex operating conditions.

[0012] 3. This invention constructs a partial Bruker optimization problem with opportunity constraints, which can still take into account methanol yield, comprehensive energy consumption, carbon dioxide utilization rate and operational risks even under conditions of uncertain perturbation distribution, model bias and rapid changes in operating conditions, thereby improving the robustness of the control strategy to extreme and uncertain operating conditions.

[0013] 4. This invention employs a two-layer rolling solution mechanism to achieve collaborative planning for slow variables and real-time closed-loop adjustment for fast variables, thereby simultaneously considering both the overall economic efficiency and local real-time performance of the system, and improving the matching efficiency between photoelectric hydrogen production and methanol synthesis.

[0014] 5. By setting up an event-driven control reconfiguration mechanism, this invention can synchronously update the graphical model, prediction model, and constraint boundaries when there are sudden changes in illumination, abnormal raw materials, accelerated equipment degradation, or hotspot trends, and perform safe projection on control actions, thereby improving the safety and emergency adaptability of system operation.

[0015] 6. By introducing counterfactual sample construction and online correction mechanisms, this invention can adaptively correct the prediction model and optimization cost without increasing the cost of real trial and error, thereby improving the adaptability to equipment aging, environmental changes and operating condition migration under long-term operating conditions.

[0016] 7. This invention enables integrated adaptive and coordinated control of the entire process of photoelectric hydrogen production to methanol production under conditions of light fluctuations, raw material fluctuations, and equipment time-varying degradation, thereby improving methanol yield and carbon dioxide utilization, and reducing the overall energy consumption per unit of methanol and the risk of operational fluctuations. Attached Figure Description

[0017] Figure 1 This is a block diagram of the photoelectric hydrogen and methanol production apparatus provided by the present invention.

[0018] Figure 2 This is a flowchart of the photoelectric hydrogen production and methanol production method provided by the present invention. Detailed Implementation

[0019] The present invention will now be described in detail with reference to the accompanying drawings.

[0020] Figure 1 This is a block diagram of the photoelectric hydrogen-to-methanol production device provided by the present invention, as shown below. Figure 1 As shown, the photoelectric hydrogen-to-methanol production apparatus provided by the present invention includes at least:

[0021] The photoelectric hydrogen production unit is used to carry out photoelectric decomposition reaction under sunlight and / or external bias to generate hydrogen, and outputs operating parameters related to the hydrogen production process, such as output voltage, current density, hydrogen production flow rate, hydrogen purity and photoelectrode temperature.

[0022] A hydrogen buffer unit, connected to the photoelectric hydrogen production unit, is used to buffer, stabilize, peak-shaving, and release the hydrogen produced by the photoelectric hydrogen production unit as needed, so as to balance the hydrogen supply fluctuations between the photoelectric hydrogen production end and the methanol synthesis end on the time scale.

[0023] The carbon dioxide supply unit is used to provide carbon dioxide feedstock to the methanol synthesis process and output parameters such as carbon dioxide feed flow rate and carbon dioxide purity.

[0024] The methanol synthesis unit is connected to the hydrogen buffer unit and the carbon dioxide supply unit respectively, and is used to carry out the methanol synthesis reaction using hydrogen and carbon dioxide, and output operating parameters such as methanol yield, reactor temperature, and reactor pressure.

[0025] The circulating separation unit is connected to the methanol synthesis unit and is used to separate the product stream after methanol synthesis to obtain methanol product. It also returns unreacted gas and / or circulating tail gas to the methanol synthesis process to form a recycling loop, while outputting parameters such as tail gas composition and recycling ratio.

[0026] The sensing and detection unit is installed in the photoelectric hydrogen production unit, hydrogen buffer unit, carbon dioxide supply unit, methanol synthesis unit and circulation separation unit, and is used to collect solar irradiance, spectral characteristics, ambient temperature, photoelectrode temperature, output voltage, current density, hydrogen production flow rate, hydrogen purity, buffer tank pressure, carbon dioxide feed flow rate, carbon dioxide purity, reactor temperature, reactor pressure, circulation ratio, methanol yield, tail gas composition and catalyst activity index.

[0027] The control and processing unit is connected to the photoelectric hydrogen production unit, hydrogen buffer unit, carbon dioxide supply unit, methanol synthesis unit, circulation separation unit, and sensing and detection unit, respectively. It is used to construct system state vectors and disturbance vectors based on the collected operating data, construct cross-unit causal graphs, perform mechanism-constrained timing prediction, sub-Bruker optimization, double-layer rolling solution, event-triggered reconstruction, safety projection, and counterfactual online correction, and generate external bias voltage setpoints, current density setpoints, electrolyte flow rate setpoints, buffer hydrogen release flow rate setpoints, carbon dioxide feed setpoints, circulation ratio setpoints, reactor temperature setpoints, and reactor pressure setpoints to achieve integrated adaptive and coordinated control of the photoelectric hydrogen production and methanol synthesis processes.

[0028] Furthermore, the device may also include:

[0029] The data storage unit is used to store historical running data, model parameters, optimization cost parameters, counterfactual samples, and control logs;

[0030] The digital twin simulation unit, connected to the control and processing unit, is used to generate simulation samples under scenarios of extreme light disturbance, raw material fluctuation, equipment degradation, and changes in exhaust gas composition, and to generate counterfactual outputs to support the control and processing unit in offline training and online correction.

[0031] Figure 2 This is a flowchart of the photoelectric hydrogen-to-methanol method provided by the present invention, as follows: Figure 2 As shown, the photoelectric hydrogen-to-methanol method provided by the present invention includes steps S1 to S8.

[0032] S1. Multi-source time-series data acquisition and unified state representation

[0033] Specifically, it includes:

[0034] Multi-source operational data for the photoelectric hydrogen production unit, hydrogen buffer unit, carbon dioxide supply unit, methanol synthesis unit, and circulating separation unit are acquired at the current time and within historical time periods. This multi-source operational data includes at least solar irradiance, spectral characteristics, ambient temperature, photoelectrode temperature, output voltage, current density, electrolyte flow rate, hydrogen production flow rate, hydrogen purity, buffer tank pressure, carbon dioxide feed flow rate, carbon dioxide purity, reactor temperature, reactor pressure, recycle ratio, methanol yield, tail gas composition, and catalyst activity indicators. The multi-source operational data undergoes unified timestamp alignment, outlier removal, missing value completion, dimensional normalization, and noise filtering to form a time-series sample sequence with a unified sampling period.

[0035] Based on the preprocessed time series samples, construct the system state vector. With disturbance vector ,in:

[0036]

[0037] In the formula, Indicates time Solar irradiance, Indicates spectral characteristics, Indicates ambient temperature. Indicates the temperature of the photoelectrode. This indicates the output voltage of the photoelectric hydrogen production unit. Indicates current density, Indicates hydrogen production flow rate, Indicates the purity of hydrogen. Indicates the pressure of the buffer tank. This indicates the carbon dioxide feed flow rate. Indicates the purity of carbon dioxide. This indicates the temperature of the methanol synthesis reactor. This indicates the pressure in the methanol synthesis reactor. Indicates the cycle ratio. Indicates methanol yield. Indicates the characteristics of exhaust gas composition. This indicates the catalyst activity index.

[0038] The disturbance vector is represented as:

[0039]

[0040] In the formula, , , , , and These represent the disturbance amounts of irradiance, spectral characteristics, ambient temperature, carbon dioxide purity, catalyst activity, and exhaust gas composition, respectively.

[0041] This invention maps key variables of the photoelectric hydrogen production end, buffer end, methanol synthesis end, and cycle separation end into a unified state vector. and perturbation vector It can provide a unified data interface for subsequent causal modeling, prediction and optimization, thereby achieving the beneficial effects of improving the consistency of cross-unit data coupling, reducing the impact of dimensional differences and sampling bias, and enhancing the stability of subsequent algorithms.

[0042] S2. Construction of cross-unit causal graphs and extraction of causal effects

[0043] Specifically, it includes:

[0044] Based on the material conservation relationships, energy conservation relationships, gas purity transfer relationships, and hydrogen-carbon ratio relationships among the photoelectric hydrogen production unit, hydrogen buffer unit, carbon dioxide supply unit, and methanol synthesis unit, a cross-unit causal graph model is constructed:

[0045]

[0046] In the formula, Indicates time The set of nodes is used to represent state variables, control variables, and disturbance variables. This represents a set of directed edges, used to characterize causal paths between variables. Let represent the set of edge weights, used to characterize the strength of each causal path. Indicates time Cross-unit causal graph.

[0047] Furthermore, the topology and edge weights of the causal graph are determined based on a joint learning approach combining structural priors and data-driven methods. The structural priors are derived from known interactions within the process, including the effects of irradiance on current density and hydrogen production flow rate, the effect of hydrogen production flow rate on buffer tank pressure, the effect of buffer tank pressure on hydrogen supply capacity, the combined effect of hydrogen supply capacity and carbon dioxide feed on methanol yield, and the effect of the recycle tail gas composition on the reactor's effective hydrogen-to-carbon ratio and methanol production rate. The data-driven part is obtained through joint optimization using sparse constraints and acyclic constraints, with the optimization objective being:

[0048]

[0049] In the formula, This represents the loss function for causal graph learning. Represents the fitting error term. This represents the norm sparse constraint term of the edge weight matrix. Indicates acyclic constraint terms. and This represents the corresponding weighting coefficient.

[0050] After obtaining the causal graph structure, the direct causal effect, indirect causal effect, and total causal effect of the perturbation variable on the target variable are extracted and expressed as follows:

[0051] ,

[0052] In the formula, Representing variables For variables Total causal effect Indicates a direct causal effect. This represents an indirect causal effect propagated through intermediate nodes; where the variable is... These can be disturbances or variables such as irradiance, raw material purity, catalyst activity, or exhaust gas composition. It can be a target quantity such as hydrogen production flow rate, hydrogen availability, effective feed state of the reactor, or methanol yield.

[0053] In this invention, the cross-unit causal graph model is trained using a combination of offline pre-training and online correction. In the offline phase, a sample set is constructed using historical runtime data and digital twin simulation data, and the initial graph structure and edge weight parameters are learned through a gradient optimization method with structural constraints. During the online phase, edge weights are recursively updated based on recent real data within the sliding time window, and the local subgraph is re-estimated when a sudden change in operating conditions is detected, so as to ensure that the causal graph can continuously reflect the real coupling relationship.

[0054] This invention constructs a cross-unit causal graph and explicitly extracts the direct and indirect impact paths of perturbations on the target quantity, thereby transforming the implicit coupling relationship between photoelectric hydrogen production and methanol synthesis into interpretable structured information. This achieves the beneficial effects of enhancing model interpretability, improving the relevance of subsequent predictions, and optimizing decision-making effectiveness.

[0055] S3. Mechanism-Constrained Temporal Prediction Network Construction and Multi-Step Prediction

[0056] Specifically, it includes:

[0057] The system state vector obtained in step S1 Perturbation vector The causal effect features extracted in step S2 are input together into the mechanism-constrained time-series prediction network to obtain future... Predicted output sequence within a step size:

[0058]

[0059] In the formula, Indicates from time At the time The predicted output sequence, Indicates the first The predicted output vector at time step [time]. This indicates the length of the prediction time domain.

[0060] The predicted output vector is represented as follows:

[0061]

[0062] In the formula, This indicates the predicted hydrogen production flow rate. This indicates a prediction of hydrogen purity. This indicates the predicted methanol yield. This indicates the predicted total energy consumption per unit of methanol. This indicates the predicted carbon dioxide utilization rate. This indicates the uncertainty of the prediction.

[0063] To ensure that the prediction results satisfy the mechanistic constraints, a total loss function is constructed:

[0064]

[0065] In the formula, This represents the total loss of the prediction model. This represents the prediction error term. This represents the material conservation constraint. This represents the energy conservation constraint term. This indicates the hydrogen-to-carbon ratio constraint. Indicates the safety boundary penalty item. , , and This represents the corresponding weighting coefficient.

[0066] The hydrogen-carbon ratio is defined as follows:

[0067]

[0068] In the formula, Indicates time The hydrogen-carbon ratio, Indicates the hydrogen supply flow rate. This indicates the carbon dioxide feed flow rate.

[0069] The mechanism-constrained time-series prediction network can be composed of a graph structure encoding layer, a time-series dynamic modeling layer, and an uncertainty output layer. The graph structure encoding layer is used to fuse the topology and edge weight information of the causal graph; the time-series dynamic modeling layer is used to extract the temporal evolution of states and disturbances; and the uncertainty output layer is used to output the mean prediction result and the uncertainty. .

[0070] In this invention, the mechanism-constrained time-series prediction network employs a training method of "historical supervised sample training + digital twin sample augmentation + online incremental fine-tuning". In the offline phase, historical running data is used to form input-output supervised sample pairs to adjust the model parameters. Backpropagation training is performed; at the same time, extended samples under extreme radiation fluctuations, raw material disturbances and equipment degradation scenarios are generated using a digital twin environment to improve the model's generalization ability to rare working conditions; in the online stage, small-step incremental fine-tuning is performed based on newly collected real samples using a sliding window method, and an experience playback mechanism is used to maintain the model's memory of historical typical working conditions.

[0071] This invention embeds causal structure information and mechanistic constraints into a time-series prediction network, enabling the simultaneous acquisition of predicted values ​​and uncertainties for key performance indicators in multi-step prediction. This achieves the beneficial effects of improving prediction accuracy under complex dynamic conditions, enhancing the physical consistency of prediction results, and providing reliable priors for subsequent robust optimization.

[0072] S4. Construction of a multi-objective optimization model for split-bars

[0073] Specifically, it includes:

[0074] Based on the future time-domain prediction results obtained in step S3, a multi-objective optimization model with sub-Bruker bars is constructed to achieve high methanol yield, low overall energy consumption, high carbon dioxide utilization rate, and low operating risk. The objective function is expressed as:

[0075]

[0076] In the formula, This represents the objective function for optimizing the blue bar. Indicates from time to The sequence of control actions, This represents the probability distribution of the perturbation. This represents an uncertain set of perturbation distributions. to This represents the weight coefficient of each objective item.

[0077] The control action vector is defined as follows:

[0078]

[0079] In the formula, This indicates the applied bias setting value. This indicates the current density setpoint. This indicates the electrolyte flow rate setpoint. This indicates the setpoint for the hydrogen discharge flow rate from the buffer tank. This indicates the carbon dioxide feed setpoint. This indicates the cycle ratio setting value. This indicates the reactor temperature setpoint. This indicates the reactor pressure setpoint.

[0080] Introduce opportunity constraints during the optimization process:

[0081]

[0082] In the formula, Indicates the first A safety or quality constraint function, Indicates the first The probability of default allowed by each constraint.

[0083] The constraint functions include at least the photoelectrode temperature constraint, buffer tank pressure constraint, hydrogen-carbon ratio constraint, reactor temperature constraint, and reactor pressure constraint, for example:

[0084]

[0085] In the formula, This indicates the maximum allowable temperature of the photoelectrode. This indicates the maximum allowable pressure of the buffer tank. and These represent the lower and upper limits of the hydrogen-to-carbon ratio, respectively. and These represent the lower and upper limits of the reactor temperature, respectively. and These represent the lower and upper limits of the reactor pressure, respectively.

[0086] This invention integrates prediction uncertainty, control variation magnitude, and safety constraints into a distributed bar optimization framework, enabling synergistic optimization of productivity, energy consumption, and carbon utilization even when the disturbance distribution is unknown or deviated. This achieves the beneficial effects of improving the robustness, economy, and safety of the control strategy.

[0087] S5, Two-layer rolling solution and cooperative control command generation, specifically includes:

[0088] A two-layer rolling solver structure is used to solve the sub-Bruker optimization problem in step S4. The upper solver is used to generate the hydrogen production-buffering-synthesis coordinated set trajectory on a longer time scale, while the lower solver is used to track the upper set trajectory according to the real-time status and output control actions on a shorter time scale.

[0089] The upper-level collaborative trajectory is represented as follows:

[0090]

[0091] In the formula, This indicates that the upper layer has set the trajectory. This indicates the target hydrogen supply trajectory within the future time domain. This represents the target buffer pressure trajectory in the future time domain. This represents the target methanol yield trajectory within the future time domain.

[0092] The lower-level real-time solver obtains the instantaneous optimal control action based on the reference trajectory and the current state:

[0093]

[0094] In the formula, Indicates time Candidate optimal control action, Indicates the currently feasible control domain. Indicates the reference state at the next moment. Represents the state error weight matrix. This represents the control increment weight matrix.

[0095] The upper-level solution cycle is longer than the lower-level solution cycle. The upper-level solution mainly focuses on global planning for matching energy and raw materials, while the lower-level solution mainly focuses on local closed-loop control for rapid adjustment of process variables.

[0096] This invention achieves a hierarchical approach by combining global collaborative planning with local rapid control, enabling the hydrogen production end, buffer end, and methanol synthesis end to operate collaboratively at different time scales. This results in a balance between global capacity matching, local real-time response, and control smoothness.

[0097] S6. Event Triggered Control Restructuring and Security Projection

[0098] Specifically, it includes:

[0099] Real-time monitoring is performed on the rate of change of irradiance, prediction uncertainty, magnitude of change of causal edge weight, rate of change of catalyst activity, reactor temperature gradient, and deviation of exhaust gas composition. A comprehensive event trigger index is constructed.

[0100] ,

[0101] In the formula, This represents a comprehensive event trigger indicator. Indicates the reference threshold for irradiance. This represents the threshold of prediction uncertainty. This represents the threshold for changes in causal edge weights. Indicates the threshold for changes in catalyst activity. Indicates the reactor temperature gradient threshold. Indicates the threshold for deviation in exhaust gas composition; when At that time, control reconfiguration is triggered.

[0102] After triggering control reconstruction, the parameters of the cause-effect graph are... Prediction model parameters and optimization cost parameters Perform synchronous updates and include candidate control actions. Projected onto the safe and feasible domain Within, the final action is obtained:

[0103]

[0104] In the formula, This indicates the action to be performed after a secure projection. Represents relative to the safe and feasible region The projection operator.

[0105] The secure feasible domain It is dynamically determined based on the current equipment safety margin, uncertainty level, and process boundaries, and is used to limit the risk of exceeding the limits of action amplitude, action change rate, and key process variables.

[0106] This invention achieves the beneficial effects of improving the system's adaptability to sudden operating conditions, reducing the probability of dangerous actions, and enhancing operational safety by triggering synchronous reconstruction of the model and control when sudden changes in illumination, accelerated equipment degradation, enhanced hotspot trends, or abnormal exhaust gas are detected, and by safely projecting the control actions.

[0107] S7. Counterfactual Sample Construction and Online Correction

[0108] Specifically, it includes:

[0109] In actual execution of control actions And obtain the actual output. Then, maintain the current state. With disturbance The actual action is replaced with at least one alternative action, while the actual action remains unchanged. And infer its corresponding output with the help of digital twin environment or structural causal model. Thus, counterfactual samples are constructed:

[0110]

[0111] In the formula, Indicates time Counterfactual samples, Indicates alternative actions. This indicates the counterfactual output corresponding to the alternative action under the same state and disturbance conditions.

[0112] Online updates of prediction model parameters based on real and counterfactual samples:

[0113] ,

[0114] In the formula, and These represent the prediction model parameters before and after the update, respectively. Indicates time The learning rate Represents the loss term of the true sample. This represents the counterfactual sample loss term. This represents the counterfactual loss weight.

[0115] At the same time, the cost parameters are optimized based on the actual output and the counterfactual output. Recursive corrections are performed to make the optimization objective more closely match the actual benefits and risks under the current operating conditions.

[0116] In this invention, the online calibration employs a combination of experience replay, sliding window incremental training, and counterfactual sample augmentation. Specifically, recent real samples and counterfactual samples are jointly written into an online sample pool, and batches are extracted according to a set ratio for small-step updates. To prevent the model from forgetting typical historical operating conditions, a historical sample replay mechanism is introduced. To prevent counterfactual samples from deviating from the real system, the digital twin environment or structural causal inference module is periodically recalibrated using the latest measured data.

[0117] This invention introduces counterfactual samples to correct the model online without increasing the cost of real trial and error. This can improve the model's ability to adapt to changes in operating conditions, enhance the long-term effectiveness of the control strategy, and reduce the risk of control performance degradation caused by model drift.

[0118] S8, Closed-loop cyclic execution and integrated adaptive cooperative control

[0119] Specifically, it includes:

[0120] Steps S1 to S7 are executed cyclically according to the set control cycle. In each control cycle, multi-source data acquisition, causal relationship identification, mechanism constraint prediction, sub-Bruker optimization, two-layer rolling solution, event-triggered reconstruction and counterfactual online correction are completed in sequence. Finally, the safety control action is output to the photoelectric hydrogen production unit, hydrogen buffer unit, carbon dioxide supply unit, methanol synthesis unit and circulation separation unit to form a continuous closed-loop control.

[0121] The comprehensive performance evaluation function of the system during the statistical period is expressed as follows:

[0122]

[0123] In the formula, This represents the system's comprehensive operational evaluation function. This represents the average methanol yield during the statistical period. This represents the average comprehensive energy consumption per unit of methanol during the statistical period. This represents the average carbon dioxide utilization rate during the statistical period. This represents the average risk cost over the statistical period. to This represents the evaluation weighting coefficient.

[0124] When the comprehensive operation evaluation function meets the preset target, the current control parameters are maintained; when the comprehensive operation evaluation function is continuously lower than the preset threshold, the online update frequency is increased, the range of counterfactual sample generation is expanded, or some model parameters are reinitialized.

[0125] This invention unifies state perception, causal identification, mechanism prediction, robust optimization, event reconstruction, and online correction into a closed-loop framework, enabling integrated adaptive and coordinated control of the entire process of photoelectric hydrogen production and methanol synthesis. This achieves the beneficial effects of increasing methanol yield, improving carbon utilization, reducing unit energy consumption and operational risks, and enhancing the long-term stable operation capability of the system.

[0126] The above description is only a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention; any modifications, equivalent substitutions, improvements or combinations made to the technical solutions within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for photoelectric hydrogen production to methanol production, characterized in that, Executed by the processor, including: A system state vector is constructed based on the operating data of the photoelectric hydrogen production unit, hydrogen buffer unit, carbon dioxide supply unit, methanol synthesis unit, and circulation separation unit. With disturbance vector A cross-unit causal graph was constructed based on the laws of material conservation, energy conservation, and the hydrogen-carbon ratio. Extract the causal effect characteristics of the perturbation variables on hydrogen production flow rate, hydrogen supply capacity, and methanol yield; and convert the system state vector... Perturbation vector The causal effect features are input into the mechanism-constrained time-series prediction network to obtain future... Within a given step, the predicted values ​​for hydrogen production flow rate, hydrogen purity, methanol yield, unit methanol comprehensive energy consumption, carbon dioxide utilization rate, and prediction uncertainty are calculated. Based on these prediction results, a multi-objective optimization model with opportunity constraints is constructed, and a two-layer rolling solution is used to generate control actions. When the comprehensive event trigger index exceeds a threshold, the causal graph parameters, prediction model parameters, and optimization cost parameters are updated synchronously, and candidate control actions are projected to the safe and feasible region before execution. Based on the actual output after execution and the counterfactual output generated by the candidate actions under the same state and disturbance conditions, the prediction model and optimization cost parameters are corrected online. This process is repeated cyclically to achieve integrated adaptive and coordinated control of the photoelectric hydrogen production and methanol production processes.

2. The method according to claim 1, characterized in that, The system state vector is represented as follows: in, Solar irradiance, For spectral characteristics, For ambient temperature, The temperature of the photoelectrode. For output voltage, For current density, For hydrogen production flow rate, For hydrogen purity, For the pressure of the buffer tank, This represents the carbon dioxide feed flow rate. For carbon dioxide purity, The temperature of the methanol synthesis reactor. The pressure of the methanol synthesis reactor. The cycle ratio is... For methanol yield, Characteristic of exhaust gas composition, This is an indicator of catalyst activity.

3. The method according to claim 1, characterized in that, The disturbance vector is represented as: in, , , , , and These are the disturbances in irradiance, spectral characteristics, ambient temperature, carbon dioxide purity, catalyst activity, and exhaust gas composition, respectively.

4. The method according to claim 1, characterized in that, The learning objectives described satisfy: in, Learning loss function for cause-effect graphs For the fitting error term, For the norm 1 sparse constraint term of the edge weight matrix, For acyclic constraint terms, and The weighting coefficients are used; and based on the cross-unit causal graph, the total causal effect, direct causal effect, and indirect causal effect are extracted, satisfying: in, For variables For variables Total causal effect For direct causal effect, This is an indirect causal effect.

5. The method according to claim 1, characterized in that, The mechanism constrains the output of the time-series prediction network in the future. The predicted output vector within each step. in, To predict hydrogen production flow rate, To predict hydrogen purity, To predict methanol yield, To predict the comprehensive energy consumption per unit of methanol, To predict carbon dioxide utilization rate, To predict uncertainty; the loss function of the mechanism-constrained time-series prediction network satisfies: in, For the prediction error term, This is a material conservation constraint. For energy conservation constraints, This is a hydrogen-to-carbon ratio constraint term. This is a penalty item for safety boundaries.

6. The method according to claim 5, characterized in that, The hydrogen-carbon ratio is defined as follows: in, For the hydrogen-carbon ratio, To supply hydrogen flow rate, This represents the carbon dioxide feed flow rate.

7. The method according to claim 1, characterized in that, The objective function of the multi-objective optimization model of the sub-bars satisfies: in, To control the sequence of actions, For the perturbation probability distribution, For a perturbation distribution of an uncertain set, to These are the weighting coefficients.

8. The method according to claim 7, characterized in that, The multi-objective optimization model of the sub-Bruker bar satisfies the chance constraint: in, For the first A safety or quality constraint function, For the first The allowed probability of default under each constraint; the safety or quality constraint function includes at least the photoelectrode temperature constraint, the buffer tank pressure constraint, the hydrogen-carbon ratio constraint, the methanol synthesis reactor temperature constraint, and the methanol synthesis reactor pressure constraint.

9. The method according to claim 1, characterized in that, The dual-layer rolling solution includes: the upper solver generating a hydrogen production-buffering-synthesis co-defined trajectory on a longer timescale. in, For the target hydrogen supply trajectory, To buffer the pressure trajectory of the target, The target methanol yield trajectory is determined; the lower-level solver solves for candidate control actions based on the current state and reference state on a shorter time scale. in, This is a feasible control domain. For reference only. Here is the state error weight matrix. To control the incremental weight matrix.

10. The method according to claim 1, characterized in that, The comprehensive event triggering index satisfies: in, This is the reference threshold for irradiance. To predict the uncertainty threshold, The threshold for causal edge weight changes. The threshold for the change in catalyst activity. The reactor temperature gradient threshold. The exhaust gas composition deviation threshold; when When this occurs, control reconfiguration is triggered, and candidate control actions are projected into the safe and feasible domain. Within, the action to be performed is obtained: in, For relative to the safe and feasible domain The projection operator.

Images