Control method and device of hydrogen production system, electronic equipment, medium and program product
By combining a hybrid prediction model and a multi-level optimization framework with adaptive compensation control, the problems of low prediction accuracy and low operating efficiency of wind and solar power output in hydrogen production systems were solved, achieving global system optimization, extending equipment life and reducing costs.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-09
- Publication Date
- 2026-03-13
AI Technical Summary
Existing hydrogen production system control methods suffer from low accuracy in predicting wind and solar power output, low system operating efficiency, lack of self-learning and self-adaptive capabilities, and inability to achieve global optimization, resulting in frequent equipment start-ups and shutdowns and shortened lifespan.
A hybrid prediction model is used to generate predicted power output from wind and solar power, a multi-state model of the hydrogen production system is constructed, and the hydrogen production allocation of the electrolyzer is obtained through a multi-level optimization framework. Combined with adaptive compensation control and online learning, the optimization objectives at the second, hour, and day levels are coordinated to achieve global optimization of the system.
It improved the accuracy of wind and solar power output prediction, enhanced system operating efficiency, extended equipment lifespan, and reduced operation and maintenance costs.
Smart Images
Figure CN121653754A_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of hydrogen production technology, and in particular to a control method, apparatus, electronic equipment, medium, and program product for a hydrogen production system. Background Technology
[0002] With the accelerating pace of the global energy transition, renewable energy-based hydrogen production technology is receiving increasing attention. While wind and solar energy are clean and renewable, their intermittent, fluctuating, and uncertain power output poses significant challenges to the stable operation of hydrogen production systems. Currently, existing hydrogen production system control methods suffer from several shortcomings: First, most control strategies employ single-time-scale optimization methods, making it difficult to simultaneously address short-term fluctuation mitigation and long-term economic optimization, thus failing to achieve globally optimal system operation. Second, traditional control methods do not adequately consider the operating characteristics of electrolyzer equipment, lacking the dynamic adaptability to long-term factors such as equipment aging and efficiency degradation, leading to shortened equipment lifespan. Third, existing prediction methods have limited accuracy, failing to provide accurate wind and solar power output forecasts for the control system, thus affecting control effectiveness. Finally, existing systems lack self-learning and adaptive capabilities, unable to automatically adjust control parameters based on operational experience and environmental changes, requiring frequent manual intervention.
[0003] In practical engineering applications, wind and solar hydrogen production systems face multiple complex challenges. On the one hand, the randomness and volatility of wind and solar power output lead to frequent start-ups and shutdowns of electrolyzers and drastic load changes, which not only reduces system operating efficiency but also accelerates equipment aging. On the other hand, as the core equipment of the hydrogen production system, the operating characteristics of the electrolyzer change with usage time and operating conditions, while traditional control methods, often based on fixed equipment models, cannot adapt to these changes. Furthermore, there are conflicting optimization objectives at different time scales: short-term optimization focuses on fluctuation mitigation, medium-term optimization focuses on operating efficiency, and long-term optimization emphasizes equipment lifespan.
[0004] Some studies have attempted to address these issues, such as using stochastic programming to handle medium- to long-term optimization problems. However, these methods often operate independently, lacking effective coordination mechanisms and failing to achieve global system optimization. Other studies have attempted to use machine learning methods to improve prediction accuracy, but often neglect the effective connection between prediction results and the control system. Summary of the Invention
[0005] The technical problem to be solved by this disclosure is to overcome the shortcomings of existing hydrogen production system control methods, such as low accuracy of wind and solar power output prediction and low system operating efficiency, and to provide a control method, device, electronic equipment, medium and program product for a hydrogen production system.
[0006] This disclosure solves the above-mentioned technical problems through the following technical solution:
[0007] The first aspect of this disclosure provides a control method for a hydrogen production system, the control method comprising:
[0008] Predicted wind and solar power output based on a hybrid prediction model;
[0009] Constructing a multi-state model of a hydrogen production system;
[0010] Based on the predicted power output of wind and solar power and the multi-state model of the hydrogen production system, a multi-level optimization framework is constructed.
[0011] The hydrogen production allocation of each electrolyzer in the hydrogen production system is obtained based on the multi-level optimization framework.
[0012] Preferably, the step of obtaining the predicted wind and solar power output based on the hybrid prediction model includes:
[0013] Acquire historical and current wind and solar data for the hydrogen production system;
[0014] The historical landscape data is preprocessed to obtain preprocessed historical landscape data;
[0015] The hybrid prediction model is trained based on the preprocessed historical landscape data to obtain the trained hybrid prediction model.
[0016] The current wind and solar data are input into the trained hybrid prediction model to obtain the predicted wind and solar power output.
[0017] Preferably, the step of constructing a multi-state model of the hydrogen production system includes:
[0018] Obtain efficiency characteristic models, aging models, and dynamic response models;
[0019] By coupling state variables of the efficiency characteristic model, the aging model, and the dynamic response model, a multi-state model of the hydrogen production system is obtained.
[0020] Preferably, the control method further includes:
[0021] The hydrogen production allocation is compensated;
[0022] And / or,
[0023] The control method further includes:
[0024] The multi-level optimization framework is updated.
[0025] Preferably, the predicted wind and solar power output includes ultra-short-term predicted wind and solar power output, short-term predicted wind and solar power output, and medium- and long-term predicted wind and solar power output; the multi-level optimization framework includes a second-level optimization layer, an hour-level optimization layer, and a day-level optimization layer.
[0026] Preferably, the control method further includes:
[0027] The second-level optimization layer aims to minimize power fluctuations based on ultra-short-term wind and solar power output predictions, and uses a model predictive control algorithm to generate second-level control commands.
[0028] And / or,
[0029] The control method further includes:
[0030] The hourly optimization layer generates hourly control commands based on short-term wind and solar power output predictions, aiming to maximize operational efficiency and minimize energy consumption, using a mixed integer programming algorithm.
[0031] And / or,
[0032] The control method further includes:
[0033] Based on the medium- and long-term predicted power output of wind and solar power, the above-mentioned optimization layer uses a multi-objective optimization algorithm to generate equipment operation and maintenance strategies for hydrogen production systems, with the goal of extending equipment life and minimizing maintenance costs.
[0034] A second aspect of this disclosure provides a control device for a hydrogen production system, the control device comprising:
[0035] The first acquisition module is used to acquire the predicted power output of wind and solar power based on the hybrid prediction model.
[0036] The first building block is used to construct a multi-state model of the hydrogen production system;
[0037] The second construction module is used to construct a multi-level optimization framework based on the predicted power output of wind and solar power and the multi-state model of the hydrogen production system.
[0038] The second acquisition module is used to acquire the hydrogen production allocation of each electrolyzer in the hydrogen production system based on the multi-level optimization framework.
[0039] Preferably, the first acquisition module includes:
[0040] The first acquisition unit is used to acquire historical and current wind and solar data of the hydrogen production system.
[0041] The preprocessing unit is used to preprocess the historical landscape data to obtain preprocessed historical landscape data.
[0042] The training unit is used to train the hybrid prediction model based on the preprocessed historical landscape data to obtain the trained hybrid prediction model.
[0043] The second acquisition unit is used to input the current wind and solar data into the trained hybrid prediction model to obtain the predicted wind and solar power output.
[0044] Preferably, the first building module includes:
[0045] The third acquisition unit is used to acquire the efficiency characteristic model, aging model and dynamic response model;
[0046] The fourth acquisition unit is used to couple the efficiency characteristic model, the aging model and the dynamic response model with state variables to obtain a multi-state model of the hydrogen production system.
[0047] Preferably, the control device further includes:
[0048] A compensation module is used to compensate for the hydrogen production allocation.
[0049] And / or,
[0050] The control device further includes:
[0051] The update module is used to update the multi-level optimization framework.
[0052] Preferably, the predicted wind and solar power output includes ultra-short-term predicted wind and solar power output, short-term predicted wind and solar power output, and medium- and long-term predicted wind and solar power output; the multi-level optimization framework includes a second-level optimization layer, an hour-level optimization layer, and a day-level optimization layer.
[0053] Preferably, the control device further includes:
[0054] The first generation module is used to generate second-level control commands by using the second-level optimization layer to generate second-level control commands based on the ultra-short-term wind and solar power output prediction power with the goal of minimizing power fluctuations, and employing a model predictive control algorithm.
[0055] And / or,
[0056] The control device further includes:
[0057] The second generation module is used to generate hourly control commands by using a hybrid integer programming algorithm based on the short-term wind and solar power output prediction power of the hourly optimization layer with the goal of maximizing operating efficiency and minimizing energy consumption.
[0058] And / or,
[0059] The control device further includes:
[0060] The third generation module is used to generate equipment operation and maintenance strategies based on the medium- and long-term wind and solar power output prediction power of the day-level optimization layer, with the goal of extending the equipment life and minimizing maintenance costs in the hydrogen production system, and using a multi-objective optimization algorithm.
[0061] A third aspect of this disclosure provides an electronic device including a memory, a processor, and a computer program stored in the memory and for running on the processor, wherein the processor executes the computer program to implement the control method of the hydrogen production system described in the first aspect.
[0062] The fourth aspect of this disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the control method for the hydrogen production system described in the first aspect.
[0063] The fifth aspect of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the control method for a hydrogen production system as described in the first aspect.
[0064] Based on common knowledge in the field, the above-mentioned preferred conditions can be combined arbitrarily to obtain various preferred embodiments of this disclosure.
[0065] The positive and progressive effects of this disclosure are as follows:
[0066] This disclosure constructs a multi-level optimization framework based on the fusion of wind and solar power output prediction with a multi-state model of the hydrogen production system; and obtains the hydrogen production allocation of each electrolyzer in the hydrogen production system based on the multi-level optimization framework, thereby improving the accuracy of wind and solar power output prediction and the system operating efficiency, and reducing operation and maintenance costs. Attached Figure Description
[0067] Figure 1 A flowchart of the control method for the hydrogen production system provided in Embodiment 1 of this disclosure.
[0068] Figure 2 This is a schematic diagram of the control device of the hydrogen production system provided in Embodiment 2 of this disclosure.
[0069] Figure 3 This is a schematic diagram of the electronic device used to implement the control method of the hydrogen production system according to Embodiment 3 of this disclosure. Detailed Implementation
[0070] The present disclosure is further illustrated below by way of embodiments, but the present disclosure is not limited to the scope of the embodiments described herein.
[0071] The prefixes such as "first" and "second" used in this disclosure are merely for distinguishing different descriptive objects and do not limit the position, order, priority, quantity, or content of the described objects. The use of ordinal numbers and other prefixes used to distinguish descriptive objects in this disclosure does not constitute a limitation on the described objects. The description of the described objects is given in the claims or the context of the embodiments, and should not be construed as an unnecessary limitation. Furthermore, in the description of this embodiment, unless otherwise stated, "multiple" means two or more.
[0072] In this embodiment of the disclosure, the collection, storage, use, processing, transmission, provision, and disclosure of user personal information comply with relevant laws and regulations and do not violate public order and good morals.
[0073] Example 1
[0074] Figure 1 A flowchart of a control method for a hydrogen production system provided in Embodiment 1 of this disclosure is shown below. Figure 1 As shown, the control method includes:
[0075] S1. Obtain the predicted power output of wind and solar power based on the hybrid prediction model;
[0076] In this embodiment, a hybrid prediction model is used to generate wind and solar power output predictions for ultra-short term, short term, and medium-to-long term, respectively.
[0077] Among them, ultra-short-term forecasts cover a range of 0-4 hours with a time resolution of 5 minutes and are mainly used for real-time control; short-term forecasts cover a range of 4-72 hours with a time resolution of 1 hour and are used for scheduling plan formulation; medium- and long-term forecasts cover a range of more than 72 hours with a time resolution of 4 hours and are used for operation and maintenance planning.
[0078] It should be noted that the hybrid prediction model adopts a CNN-BiLSTM-ARIMA combined structure, in which CNN (convolutional neural network) is used for feature extraction (i.e., extracting spatial features of wind and light data), BiLSTM (bidirectional long short-term memory network) is used for time series prediction (i.e., capturing long-term dependencies of time series), and ARIMA (autoregressive integral moving average model) is used for residual correction (i.e., residual correction and uncertainty quantification).
[0079] S2. Construct a multi-state model of the hydrogen production system;
[0080] S3. Based on the multi-state model of the hydrogen production system fused with wind and solar power output prediction, a multi-level optimization framework is constructed.
[0081] In this embodiment, the multi-level optimization framework includes a second-level optimization layer, an hour-level optimization layer, and a day-level optimization layer.
[0082] Specifically, the second-level optimization layer accepts the settings of the hour-level optimization layer, and the hour-level optimization layer considers the maintenance plan of the day-level optimization layer, forming a closed-loop optimization structure.
[0083] S4. Obtain the hydrogen production allocation of each electrolyzer in the hydrogen production system based on a multi-level optimization framework.
[0084] This embodiment is based on the multi-state model of the hydrogen production system, which is fused with the predicted power output of wind and solar power, and constructs a multi-level optimization framework. Based on the multi-level optimization framework, the hydrogen production allocation of each electrolyzer in the hydrogen production system is obtained, which improves the accuracy of wind and solar power output prediction and the system operating efficiency, and reduces the operation and maintenance costs.
[0085] In an optional embodiment, S1 includes:
[0086] Acquire historical and current wind and solar data for the hydrogen production system;
[0087] Historical landscape data is preprocessed to obtain preprocessed historical landscape data;
[0088] The hybrid prediction model is trained based on the preprocessed historical landscape data to obtain the trained hybrid prediction model.
[0089] Input the current wind and solar data into the trained hybrid prediction model to obtain the predicted power output of wind and solar power.
[0090] In this embodiment, the historical landscape data is first imputed for outliers and missing values and then standardized. Then, a CNN network is used to extract features, and the feature map is fed into a BiLSTM network after passing through a flattening layer. The output of the BiLSTM network is then fed into an ARIMA model for residual correction. Finally, the prediction results of multiple models are fused through ensemble learning to generate probabilistic prediction outputs, including point prediction values and prediction interval estimates.
[0091] In an optional embodiment, S2 includes:
[0092] Obtain efficiency characteristic models, aging models, and dynamic response models;
[0093] By coupling state variables into the efficiency characteristic model, aging model, and dynamic response model, a multi-state model of the hydrogen production system is obtained.
[0094] In this embodiment, a multi-state model of the hydrogen production system is constructed, considering factors such as electrolyzer efficiency decay and equipment aging. The efficiency characteristic model in the multi-state model uses a polynomial function to characterize the relationship between efficiency and load rate, the aging model uses an exponential decay model based on operating time, and the dynamic response model uses a transfer function with a first-order inertial element and pure time delay. Specifically, the efficiency characteristic model is obtained by fitting experimental data, describing the efficiency characteristics of the electrolyzer under different load rates, temperatures, and pressures. It uses a polynomial function to characterize the relationship between efficiency and load rate, an exponential function to characterize the effect of temperature on efficiency, and a logarithmic function to characterize the effect of pressure on efficiency. The aging model uses an exponential decay model based on operating time, and the model parameters are determined through accelerated aging tests to characterize the law of electrolyzer performance decay with operating time. The dynamic response model uses a transfer function with a first-order inertial element and pure time delay to describe the dynamic response characteristics of the electrolyzer power change; the model parameters are determined through step response tests. These three sub-models (i.e., the efficiency characteristic model, the aging model, and the dynamic response model) are coupled through state variables to form a complete multi-state model of the hydrogen production system, which can accurately reflect the dynamic characteristics of the hydrogen production system under different operating conditions.
[0095] In an optional embodiment, the control method further includes:
[0096] Compensation for hydrogen production allocation;
[0097] In this embodiment, the hydrogen production allocation is adaptively and dynamically compensated based on the real-time monitored status data of the hydrogen production system. The adaptive dynamic compensation control includes efficiency compensation, aging compensation, and response compensation, wherein the efficiency compensation adopts an adaptive compensation algorithm based on fuzzy logic.
[0098] Specifically, efficiency compensation employs an adaptive compensation algorithm based on fuzzy logic. It dynamically adjusts the power setpoint according to the magnitude and trend of the deviation between the actual efficiency of the electrolyzer and the model's predicted value. The compensation coefficient is determined through a fuzzy inference system. The input variables are efficiency deviation and the rate of change of deviation, and the output variable is the compensation coefficient. Aging compensation adaptively adjusts the operating parameter range based on the degree of equipment aging. By monitoring the changing trends of parameters such as electrolyzer voltage, current, and temperature, it estimates the equipment's health status in real time and adjusts the power limit and rate of change limit accordingly. Response compensation optimizes the power change rate based on the dynamic response characteristics of the hydrogen production system. By identifying the system's dynamic characteristics online, it adaptively adjusts the controller parameters to improve the dynamic response performance of the hydrogen production system. These three compensation mechanisms work together to form a complete adaptive compensation control system.
[0099] In an optional embodiment, the control method further includes:
[0100] Update the multi-level optimization framework.
[0101] In this embodiment, a reinforcement learning algorithm is used to continuously update the hybrid prediction model and optimization parameters based on historical operating data. The Proximal Policy Optimization (PPO) algorithm is employed as the reinforcement learning framework. The state space includes 30 dimensions of features such as wind and solar power output prediction errors, hydrogen production system operating status, and equipment aging. The action space represents the adjustment of optimization parameters, including prediction model parameters, optimization target weights, and constraint limits. The reward function design considers multiple performance indicators, including wind and solar power absorption rate, hydrogen production efficiency, equipment lifespan, and operating costs. The learning process employs an offline training and online application approach. Offline pre-training is performed using historical wind and solar data, while online fine-tuning is conducted based on real-time operating data. The update mechanism combines periodic updates and triggered updates. The periodic update cycle is 7 days, and triggered updates are initiated when system performance degrades beyond a threshold. Through continuous learning, the system can adapt to environmental changes and equipment aging, maintaining optimal operating status.
[0102] In one optional embodiment, the predicted wind and solar power output includes ultra-short-term predicted wind and solar power output, short-term predicted wind and solar power output, and medium- and long-term predicted wind and solar power output; the multi-level optimization framework includes a second-level optimization layer, an hour-level optimization layer, and a day-level optimization layer.
[0103] In an optional embodiment, the control method further includes:
[0104] By using a second-level optimization layer based on ultra-short-term wind and solar power output prediction with the goal of minimizing power fluctuations, a model predictive control algorithm is used to generate second-level control commands.
[0105] In an optional embodiment, the control method further includes:
[0106] By using an hourly optimization layer to generate hourly control commands based on short-term wind and solar power output predictions, with the goal of maximizing operational efficiency and minimizing energy consumption, a mixed integer programming algorithm is employed to generate hourly control commands.
[0107] In an optional embodiment, the control method further includes:
[0108] By using a day-level optimization layer based on medium- to long-term wind and solar power output predictions, and with the goal of extending equipment lifespan and minimizing maintenance costs in hydrogen production systems, a multi-objective optimization algorithm is employed to generate equipment operation and maintenance strategies.
[0109] In this embodiment, a coordinated optimization framework is constructed, comprising three optimization layers: a second-level optimization layer, an hour-level optimization layer, and a day-level optimization layer. The second-level optimization layer, based on ultra-short-term wind and solar power output predictions, aims to minimize power fluctuations. It uses a model predictive control (MPC) algorithm to generate second-level control commands, with a control cycle of 1 second, a prediction time domain of 30 minutes, and a control time domain of 5 minutes. The optimization objective function includes power tracking error, control variable change rate, and constraint violation penalty terms. The hour-level optimization layer, based on short-term wind and solar power output predictions, aims to maximize operating efficiency and minimize energy consumption. It uses a mixed-integer programming algorithm to generate hourly scheduling plans, considering electrolyzer start-up and shutdown costs, operation and maintenance constraints, and grid interaction requirements. The optimization cycle is 1 hour, and the rolling optimization window is 24 hours. The day-level optimization layer, based on medium- and long-term wind and solar power output predictions, aims to extend equipment lifespan and minimize maintenance costs. It uses a multi-objective optimization algorithm to generate equipment operation and maintenance strategies, with an optimization cycle of 24 hours, considering equipment aging status, preventative maintenance plans, and spare parts inventory. The three-level optimization is coupled through coordination variables. The second-level optimization accepts the set values of the hour-level optimization, and the hour-level optimization takes into account the maintenance plan of the day-level optimization, forming a closed-loop optimization structure.
[0110] In specific implementation, for example, this embodiment is implemented in an 800MW wind-solar hydrogen production system, which includes 700MW of photovoltaic power, 100MW of wind power, 48 200 Nm³ PEM electrolyzers and 36 1000 Nm³ alkaline electrolyzers. The system is equipped with comprehensive monitoring equipment, including power sensors, temperature sensors, pressure sensors, hydrogen purity analyzers, etc., with a data acquisition frequency of 1Hz. The control system adopts a distributed architecture, including a field control layer, a process optimization layer, and a scheduling management layer.
[0111] The implementation process begins with data preparation and model training. Wind and solar power output data, meteorological data, and hydrogen production system operation data from the past year were collected. After data preprocessing, a CNN-BiLSTM-ARIMA hybrid prediction model was trained. Training utilized the TensorFlow framework, with the CNN portion employing three convolutional layers, the BiLSTM portion using two recurrent layers, and the ARIMA parameters determined through grid search. The trained hybrid prediction model can generate wind and solar power output forecasts across multiple time scales: ultra-short-term forecasts updated every 5 minutes, short-term forecasts updated hourly, and medium- to long-term forecasts updated every 4 hours.
[0112] Multi-state modeling of the hydrogen production system is based on performance data and historical operating data provided by the equipment manufacturer. The efficiency characteristic model is obtained through polynomial fitting, characterizing the relationship between electrolyzer efficiency and load rate; the aging model adopts an exponential decay form, and the decay coefficient is determined through accelerated aging tests; the dynamic response model is identified through step response tests, adopting a first-order inertial plus pure time-delay form. After integrating the three sub-models (i.e., the efficiency characteristic model, the aging model, and the dynamic response model), the performance of the hydrogen production system under different operating conditions can be accurately predicted.
[0113] The multi-timescale coordinated optimization implementation adopts a hierarchical structure. The second-level optimization layer is deployed on the field controller, using the MPC algorithm with a calculation cycle of 1 second; the hour-level optimization layer is deployed on the process optimization server, using the mixed integer programming algorithm with an optimization cycle of 1 hour; and the day-level optimization layer is deployed on the scheduling management platform, using a multi-objective optimization algorithm with an optimization cycle of 24 hours. The three levels of optimization exchange and coordinate data through the OPC protocol.
[0114] Adaptive compensation control operates in real-time at the process optimization layer. The efficiency compensation module calculates the compensation coefficient every 5 minutes and adjusts the power setpoint based on real-time efficiency data; the aging compensation module assesses the equipment health status daily and adjusts the operating parameter range; the response compensation module updates the controller parameters hourly to optimize the system's dynamic response. Compensation results are sent to field equipment via control commands.
[0115] Online learning and updates are implemented using a cloud computing platform. The PPO algorithm is deployed on a cloud server, collecting runtime data weekly for model updates and parameter optimization. The updated model parameters are then distributed to control systems at all levels via a secure channel. A manual verification step is also included to ensure the update process is secure and reliable.
[0116] The results show that the hydrogen production system has achieved significant technical benefits, with a 35% increase in hydrogen production efficiency stability under fluctuating wind and solar power conditions, a 20% extension in equipment lifespan, a 25% reduction in overall operating costs, and a 15% increase in wind and solar power absorption rate.
[0117] This embodiment relates to the operation control strategy of hydrogen production systems powered by renewable energy sources such as wind and solar power, specifically an adaptive control method that can effectively address the uncertainty of wind and solar power output, improve system operating efficiency, and extend equipment lifespan. This method is applicable to wind-solar hybrid hydrogen production systems, photovoltaic hydrogen production systems, or wind power hydrogen production systems. Specifically, firstly, a hybrid prediction model combining CNN, BiLSTM, and ARIMA is used to generate probabilistic prediction data (i.e., predicted wind and solar power output) for ultra-short-term (0-4 hours), short-term (4-72 hours), and medium-to-long-term (over 72 hours) wind and solar power output. Secondly, a multi-state model of the hydrogen production system considering efficiency characteristics, equipment aging, and dynamic response is established. Based on this, a multi-timescale coordinated optimization framework at the second, hour, and day levels is constructed, employing model predictive control, mixed integer programming, and multi-objective optimization algorithms to achieve objectives such as minimizing power fluctuations, maximizing operating efficiency, and extending equipment lifespan. Furthermore, an adaptive compensation control based on fuzzy logic is introduced to compensate for efficiency, aging, and dynamic response in real time. Finally, by using a near-end strategy optimization algorithm for online learning and updating, the system parameters are continuously optimized, which can effectively cope with the fluctuations of wind and solar power, and improve the operating efficiency, adaptability and equipment life of the hydrogen production system.
[0118] Example 2
[0119] Corresponding to the aforementioned embodiment of a control method for a hydrogen production system, this disclosure also provides an embodiment of a control device for a hydrogen production system.
[0120] Figure 2 This is a schematic diagram of the control device of a hydrogen production system provided in Embodiment 2 of this disclosure, as shown below. Figure 2 As shown, the control device includes:
[0121] The first acquisition module 21 is used to acquire the predicted power output of wind and solar power based on the hybrid prediction model;
[0122] In this embodiment, a hybrid prediction model is used to generate wind and solar power output predictions for ultra-short term, short term, and medium-to-long term, respectively.
[0123] Among them, ultra-short-term forecasts cover a range of 0-4 hours with a time resolution of 5 minutes and are mainly used for real-time control; short-term forecasts cover a range of 4-72 hours with a time resolution of 1 hour and are used for scheduling plan formulation; medium- and long-term forecasts cover a range of more than 72 hours with a time resolution of 4 hours and are used for operation and maintenance planning.
[0124] It should be noted that the hybrid prediction model adopts a CNN-BiLSTM-ARIMA combined structure, in which CNN (convolutional neural network) is used for feature extraction (i.e., extracting spatial features of wind and light data), BiLSTM (bidirectional long short-term memory network) is used for time series prediction (i.e., capturing long-term dependencies of time series), and ARIMA (autoregressive integral moving average model) is used for residual correction (i.e., residual correction and uncertainty quantification).
[0125] The first building module 22 is used to build a multi-state model of the hydrogen production system;
[0126] The second building module 23 is used to build a multi-level optimization framework based on the multi-state model of the hydrogen production system fused with wind and solar power output prediction power.
[0127] In this embodiment, the multi-level optimization framework includes a second-level optimization layer, an hour-level optimization layer, and a day-level optimization layer.
[0128] Specifically, the second-level optimization layer accepts the settings of the hour-level optimization layer, and the hour-level optimization layer considers the maintenance plan of the day-level optimization layer, forming a closed-loop optimization structure.
[0129] The second acquisition module 24 is used to acquire the hydrogen production allocation of each electrolyzer in the hydrogen production system based on a multi-level optimization framework.
[0130] This embodiment is based on the multi-state model of the hydrogen production system, which is fused with the predicted power output of wind and solar power, and constructs a multi-level optimization framework. Based on the multi-level optimization framework, the hydrogen production allocation of each electrolyzer in the hydrogen production system is obtained, which improves the accuracy of wind and solar power output prediction and the system operating efficiency, and reduces the operation and maintenance costs.
[0131] In an optional embodiment, the first acquisition module includes:
[0132] The first acquisition unit is used to acquire historical and current wind and solar data of the hydrogen production system.
[0133] The preprocessing unit is used to preprocess historical landscape data to obtain preprocessed historical landscape data.
[0134] The training unit is used to train the hybrid prediction model based on preprocessed historical landscape data to obtain the trained hybrid prediction model.
[0135] The second acquisition unit is used to input the current wind and solar data into the trained hybrid prediction model to obtain the predicted wind and solar power output.
[0136] In this embodiment, the historical landscape data is first imputed for outliers and missing values and then standardized. Then, a CNN network is used to extract features, and the feature map is fed into a BiLSTM network after passing through a flattening layer. The output of the BiLSTM network is then fed into an ARIMA model for residual correction. Finally, the prediction results of multiple models are fused through ensemble learning to generate probabilistic prediction outputs, including point prediction values and prediction interval estimates.
[0137] In an optional embodiment, the first building module includes:
[0138] The third acquisition unit is used to acquire the efficiency characteristic model, aging model and dynamic response model;
[0139] The fourth acquisition unit is used to couple state variables of the efficiency characteristic model, aging model and dynamic response model to obtain a multi-state model of the hydrogen production system.
[0140] In this embodiment, a multi-state model of the hydrogen production system is constructed, considering factors such as electrolyzer efficiency decay and equipment aging. The efficiency characteristic model in the multi-state model uses a polynomial function to characterize the relationship between efficiency and load rate, the aging model uses an exponential decay model based on operating time, and the dynamic response model uses a transfer function with a first-order inertial element and pure time delay. Specifically, the efficiency characteristic model is obtained by fitting experimental data, describing the efficiency characteristics of the electrolyzer under different load rates, temperatures, and pressures. It uses a polynomial function to characterize the relationship between efficiency and load rate, an exponential function to characterize the effect of temperature on efficiency, and a logarithmic function to characterize the effect of pressure on efficiency. The aging model uses an exponential decay model based on operating time, and the model parameters are determined through accelerated aging tests to characterize the law of electrolyzer performance decay with operating time. The dynamic response model uses a transfer function with a first-order inertial element and pure time delay to describe the dynamic response characteristics of the electrolyzer power change; the model parameters are determined through step response tests. These three sub-models (i.e., the efficiency characteristic model, the aging model, and the dynamic response model) are coupled through state variables to form a complete multi-state model of the hydrogen production system, which can accurately reflect the dynamic characteristics of the hydrogen production system under different operating conditions.
[0141] In an optional embodiment, the control device further includes:
[0142] The compensation module is used to compensate for the hydrogen production allocation.
[0143] In this embodiment, the hydrogen production allocation is adaptively and dynamically compensated based on the real-time monitored status data of the hydrogen production system. The adaptive dynamic compensation control includes efficiency compensation, aging compensation, and response compensation, wherein the efficiency compensation adopts an adaptive compensation algorithm based on fuzzy logic.
[0144] Specifically, efficiency compensation employs an adaptive compensation algorithm based on fuzzy logic. It dynamically adjusts the power setpoint according to the magnitude and trend of the deviation between the actual efficiency of the electrolyzer and the model's predicted value. The compensation coefficient is determined through a fuzzy inference system. The input variables are efficiency deviation and the rate of change of deviation, and the output variable is the compensation coefficient. Aging compensation adaptively adjusts the operating parameter range based on the degree of equipment aging. By monitoring the changing trends of parameters such as electrolyzer voltage, current, and temperature, it estimates the equipment's health status in real time and adjusts the power limit and rate of change limit accordingly. Response compensation optimizes the power change rate based on the dynamic response characteristics of the hydrogen production system. By identifying the system's dynamic characteristics online, it adaptively adjusts the controller parameters to improve the dynamic response performance of the hydrogen production system. These three compensation mechanisms work together to form a complete adaptive compensation control system.
[0145] In an optional embodiment, the control device further includes:
[0146] The update module is used to update the multi-level optimization framework.
[0147] In this embodiment, a reinforcement learning algorithm is used to continuously update the hybrid prediction model and optimization parameters based on historical operating data. The Proximal Policy Optimization (PPO) algorithm is employed as the reinforcement learning framework. The state space includes 30 dimensions of features such as wind and solar power output prediction errors, hydrogen production system operating status, and equipment aging. The action space represents the adjustment of optimization parameters, including prediction model parameters, optimization target weights, and constraint limits. The reward function design considers multiple performance indicators, including wind and solar power absorption rate, hydrogen production efficiency, equipment lifespan, and operating costs. The learning process employs an offline training and online application approach. Offline pre-training is performed using historical wind and solar data, while online fine-tuning is conducted based on real-time operating data. The update mechanism combines periodic updates and triggered updates. The periodic update cycle is 7 days, and triggered updates are initiated when system performance degrades beyond a threshold. Through continuous learning, the system can adapt to environmental changes and equipment aging, maintaining optimal operating status.
[0148] In one optional embodiment, the predicted wind and solar power output includes ultra-short-term predicted wind and solar power output, short-term predicted wind and solar power output, and medium- and long-term predicted wind and solar power output; the multi-level optimization framework includes a second-level optimization layer, an hour-level optimization layer, and a day-level optimization layer.
[0149] In an optional embodiment, the control device further includes:
[0150] The first generation module is used to generate second-level control commands by using a model predictive control algorithm to minimize power fluctuations based on ultra-short-term wind and solar power output prediction power through a second-level optimization layer;
[0151] In an optional embodiment, the control device further includes:
[0152] The second generation module is used to generate hourly control commands by using a hybrid integer programming algorithm to maximize operating efficiency and minimize energy consumption based on short-term wind and solar power output prediction through the hourly optimization layer;
[0153] In an optional embodiment, the control device further includes:
[0154] The third generation module is used to generate equipment operation and maintenance strategies based on medium- and long-term wind and solar power output predictions through a day-level optimization layer, with the goal of extending equipment life and minimizing maintenance costs in the hydrogen production system. It employs a multi-objective optimization algorithm.
[0155] In this embodiment, a coordinated optimization framework is constructed, comprising three optimization layers: a second-level optimization layer, an hour-level optimization layer, and a day-level optimization layer. The second-level optimization layer, based on ultra-short-term wind and solar power output predictions, aims to minimize power fluctuations. It uses a model predictive control (MPC) algorithm to generate second-level control commands, with a control cycle of 1 second, a prediction time domain of 30 minutes, and a control time domain of 5 minutes. The optimization objective function includes power tracking error, control variable change rate, and constraint violation penalty terms. The hour-level optimization layer, based on short-term wind and solar power output predictions, aims to maximize operating efficiency and minimize energy consumption. It uses a mixed-integer programming algorithm to generate hourly scheduling plans, considering electrolyzer start-up and shutdown costs, operation and maintenance constraints, and grid interaction requirements. The optimization cycle is 1 hour, and the rolling optimization window is 24 hours. The day-level optimization layer, based on medium- and long-term wind and solar power output predictions, aims to extend equipment lifespan and minimize maintenance costs. It uses a multi-objective optimization algorithm to generate equipment operation and maintenance strategies, with an optimization cycle of 24 hours, considering equipment aging status, preventative maintenance plans, and spare parts inventory. The three-level optimization is coupled through coordination variables. The second-level optimization accepts the set values of the hour-level optimization, and the hour-level optimization takes into account the maintenance plan of the day-level optimization, forming a closed-loop optimization structure.
[0156] In specific implementation, for example, this embodiment is implemented in an 800MW wind-solar hydrogen production system, which includes 700MW of photovoltaic power, 100MW of wind power, 48 200 Nm³ PEM electrolyzers and 36 1000 Nm³ alkaline electrolyzers. The system is equipped with comprehensive monitoring equipment, including power sensors, temperature sensors, pressure sensors, hydrogen purity analyzers, etc., with a data acquisition frequency of 1Hz. The control system adopts a distributed architecture, including a field control layer, a process optimization layer, and a scheduling management layer.
[0157] The implementation process begins with data preparation and model training. Wind and solar power output data, meteorological data, and hydrogen production system operation data from the past year were collected. After data preprocessing, a CNN-BiLSTM-ARIMA hybrid prediction model was trained. Training utilized the TensorFlow framework, with the CNN portion employing three convolutional layers, the BiLSTM portion using two recurrent layers, and the ARIMA parameters determined through grid search. The trained hybrid prediction model can generate wind and solar power output forecasts across multiple time scales: ultra-short-term forecasts updated every 5 minutes, short-term forecasts updated hourly, and medium- to long-term forecasts updated every 4 hours.
[0158] Multi-state modeling of the hydrogen production system is based on performance data and historical operating data provided by the equipment manufacturer. The efficiency characteristic model is obtained through polynomial fitting, characterizing the relationship between electrolyzer efficiency and load rate; the aging model adopts an exponential decay form, and the decay coefficient is determined through accelerated aging tests; the dynamic response model is identified through step response tests, adopting a first-order inertial plus pure time-delay form. After integrating the three sub-models (i.e., the efficiency characteristic model, the aging model, and the dynamic response model), the performance of the hydrogen production system under different operating conditions can be accurately predicted.
[0159] The multi-timescale coordinated optimization implementation adopts a hierarchical structure. The second-level optimization layer is deployed on the field controller, using the MPC algorithm with a calculation cycle of 1 second; the hour-level optimization layer is deployed on the process optimization server, using the mixed integer programming algorithm with an optimization cycle of 1 hour; and the day-level optimization layer is deployed on the scheduling management platform, using a multi-objective optimization algorithm with an optimization cycle of 24 hours. The three levels of optimization exchange and coordinate data through the OPC protocol.
[0160] Adaptive compensation control operates in real-time at the process optimization layer. The efficiency compensation module calculates the compensation coefficient every 5 minutes and adjusts the power setpoint based on real-time efficiency data; the aging compensation module assesses the equipment health status daily and adjusts the operating parameter range; the response compensation module updates the controller parameters hourly to optimize the system's dynamic response. Compensation results are sent to field equipment via control commands.
[0161] Online learning and updates are implemented using a cloud computing platform. The PPO algorithm is deployed on a cloud server, collecting runtime data weekly for model updates and parameter optimization. The updated model parameters are then distributed to control systems at all levels via a secure channel. A manual verification step is also included to ensure the update process is secure and reliable.
[0162] The results show that the hydrogen production system has achieved significant technical benefits, with a 35% increase in hydrogen production efficiency stability under fluctuating wind and solar power conditions, a 20% extension in equipment lifespan, a 25% reduction in overall operating costs, and a 15% increase in wind and solar power absorption rate.
[0163] This embodiment relates to the operation control strategy of hydrogen production systems powered by renewable energy sources such as wind and solar power, specifically an adaptive control method that can effectively address the uncertainty of wind and solar power output, improve system operating efficiency, and extend equipment lifespan. This method is applicable to wind-solar hybrid hydrogen production systems, photovoltaic hydrogen production systems, or wind power hydrogen production systems. Specifically, firstly, a hybrid prediction model combining CNN, BiLSTM, and ARIMA is used to generate probabilistic prediction data (i.e., predicted wind and solar power output) for ultra-short-term (0-4 hours), short-term (4-72 hours), and medium-to-long-term (over 72 hours) wind and solar power output. Secondly, a multi-state model of the hydrogen production system considering efficiency characteristics, equipment aging, and dynamic response is established. Based on this, a multi-timescale coordinated optimization framework at the second, hour, and day levels is constructed, employing model predictive control, mixed integer programming, and multi-objective optimization algorithms to achieve objectives such as minimizing power fluctuations, maximizing operating efficiency, and extending equipment lifespan. Furthermore, an adaptive compensation control based on fuzzy logic is introduced to compensate for efficiency, aging, and dynamic response in real time. Finally, by using a near-end strategy optimization algorithm for online learning and updating, the system parameters are continuously optimized, which can effectively cope with the fluctuations of wind and solar power, and improve the operating efficiency, adaptability and equipment life of the hydrogen production system.
[0164] For the system embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs.
[0165] Example 3
[0166] Figure 3 This is a schematic diagram of the structure of an electronic device according to Embodiment 3 of this disclosure. The electronic device includes a memory, a processor, and a computer program stored in the memory and used to run on the processor. When the processor executes the computer program, it implements the control method of the hydrogen production system described in any of the above embodiments. Figure 3 The electronic device 90 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.
[0167] like Figure 3 As shown, the electronic device 90 can be manifested as a general-purpose computing device, such as a server device. The components of the electronic device 90 may include, but are not limited to: at least one processor 91, at least one memory 92, and a bus 93 connecting different system components (including memory 92 and processor 91).
[0168] Bus 93 includes a data bus, an address bus, and a control bus.
[0169] The memory 92 may include volatile memory, such as random access memory (RAM) 921 and / or cache memory 922, and may further include read-only memory (ROM) 923.
[0170] The memory 92 may also include a program tool 925 (or utility) having a set (at least one) program module 924, such program module 924 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.
[0171] The processor 91 executes various functional applications and data processing by running computer programs stored in the memory 92, such as the control method of the hydrogen production system provided in any of the above embodiments.
[0172] Electronic device 90 can also communicate with one or more external devices 94 (e.g., keyboard, pointing device, etc.). This communication can be performed via input / output (I / O) interface 95. Furthermore, electronic device 90 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 96. Figure 3 As shown, network adapter 96 communicates with other modules of electronic device 90 via bus 93. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with electronic device 90, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems.
[0173] It should be noted that although several units / modules or sub-units / modules of the electronic device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.
[0174] Example 4
[0175] Embodiment 4 of this disclosure also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the control method of the hydrogen production system provided in any of the above embodiments.
[0176] The readable storage medium may be more specifically adopted, including but not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.
[0177] Example 5
[0178] Embodiment 5 of this disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the control method of the hydrogen production system described in any of the above claims.
[0179] The program code for executing the computer program product of this disclosure can be written in any combination of one or more programming languages, and the program code can be executed entirely on a user device, partially on a user device, as a stand-alone software package, partially on a user device and partially on a remote device, or entirely on a remote device.
[0180] While specific embodiments of this disclosure have been described above, those skilled in the art should understand that these are merely illustrative examples, and the scope of protection of this disclosure is defined by the appended claims. Those skilled in the art can make various changes or modifications to these embodiments without departing from the principles and essence of this disclosure, but all such changes and modifications fall within the scope of protection of this disclosure.
Claims
1. A control method for a hydrogen production system, characterized in that, The control method includes: Predicted wind and solar power output based on a hybrid prediction model; Constructing a multi-state model of a hydrogen production system; Based on the predicted power output of wind and solar power and the multi-state model of the hydrogen production system, a multi-level optimization framework is constructed. The hydrogen production allocation of each electrolyzer in the hydrogen production system is obtained based on the multi-level optimization framework.
2. The control method for the hydrogen production system as described in claim 1, characterized in that, The steps for obtaining the predicted wind and solar power output based on the hybrid prediction model include: Acquire historical and current wind and solar data for the hydrogen production system; The historical landscape data is preprocessed to obtain preprocessed historical landscape data; The hybrid prediction model is trained based on the preprocessed historical landscape data to obtain the trained hybrid prediction model. The current wind and solar data are input into the trained hybrid prediction model to obtain the predicted wind and solar power output.
3. The control method for the hydrogen production system as described in claim 1, characterized in that, The steps for constructing a multi-state model of the hydrogen production system include: Obtain efficiency characteristic models, aging models, and dynamic response models; By coupling state variables of the efficiency characteristic model, the aging model, and the dynamic response model, a multi-state model of the hydrogen production system is obtained.
4. The control method for the hydrogen production system as described in claim 1, characterized in that, The control method further includes: The hydrogen production allocation is compensated; And / or, The control method further includes: The multi-level optimization framework is updated.
5. The control method for the hydrogen production system as described in claim 1, characterized in that, The predicted wind and solar power output includes ultra-short-term predicted wind and solar power output, short-term predicted wind and solar power output, and medium- and long-term predicted wind and solar power output; the multi-level optimization framework includes a second-level optimization layer, an hour-level optimization layer, and a day-level optimization layer.
6. The control method for the hydrogen production system as described in claim 5, characterized in that, The control method further includes: The second-level optimization layer aims to minimize power fluctuations based on ultra-short-term wind and solar power output predictions, and uses a model predictive control algorithm to generate second-level control commands. And / or, The control method further includes: The hourly optimization layer generates hourly control commands based on short-term wind and solar power output predictions, aiming to maximize operational efficiency and minimize energy consumption, using a mixed integer programming algorithm. And / or, The control method further includes: Based on the medium- and long-term predicted power output of wind and solar power, the above-mentioned optimization layer uses a multi-objective optimization algorithm to generate equipment operation and maintenance strategies for hydrogen production systems, with the goal of extending equipment life and minimizing maintenance costs.
7. A control device for a hydrogen production system, characterized in that, The control device includes: The first acquisition module is used to acquire the predicted power output of wind and solar power based on the hybrid prediction model. The first building block is used to construct a multi-state model of the hydrogen production system; The second construction module is used to construct a multi-level optimization framework based on the predicted power output of wind and solar power and the multi-state model of the hydrogen production system. The second acquisition module is used to acquire the hydrogen production allocation of each electrolyzer in the hydrogen production system based on the multi-level optimization framework.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and for running on the processor, characterized in that, When the processor executes the computer program, it implements the control method of the hydrogen production system according to any one of claims 1 to 6.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the control method of the hydrogen production system according to any one of claims 1 to 6.
10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the control method of the hydrogen production system as described in any one of claims 1 to 6.