Minute-level intelligent scheduling method and system for renewable energy hydrogen-based energy

Abstract

The application provides a minute-level intelligent scheduling method and system of a renewable energy hydrogen production-based energy source, relates to the technical field of energy management and intelligent scheduling, and comprises the following steps: obtaining power generation, hydrogen production equipment and hydrogen storage system data, adaptively adjusting the length of a dynamic time window based on power generation fluctuation frequency and generating a fluctuation feature sequence, constructing a minute-level intelligent scheduling model, and optimizing hydrogen production power regulation instructions by using mutual feedback correction parameters; when an anomaly is detected, an optimal power transfer path is calculated and a stepwise load reduction shutdown is performed. The application can improve the operation stability of the hydrogen production system, realize efficient utilization of power generation power, and reduce system operation risks.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy management and intelligent scheduling, in particular to a minute-level intelligent scheduling method and system for renewable energy hydrogen-based energy. BACKGROUND

[0002] With the rapid development of renewable energy hydrogen production technology, how to effectively cope with the intermittency and volatility of renewable energy power generation and realize the stable operation of hydrogen energy system has become a key problem that needs to be solved in the industry. The current scheduling method of renewable energy hydrogen production system mainly adjusts power based on fixed time windows, which is difficult to adapt to the random fluctuation characteristics of power generation, and is prone to cause system response lag and insufficient adjustment accuracy.

[0003] The existing hydrogen production power regulation scheme often controls the water electrolysis hydrogen production equipment and the hydrogen storage system as independent units, lacks a system-level collaborative optimization mechanism, resulting in low equipment utilization, and it is difficult to achieve smooth transition when the system fails. At the same time, due to the complexity of the characteristics of renewable energy power generation, traditional power regulation methods are difficult to accurately capture the dynamic characteristics of power generation fluctuations, affecting the overall operation efficiency of the system.

[0004] The traditional scheduling control method uses a simple emergency shutdown strategy when the system fails, which can easily cause excessive stress on the equipment and fluctuations in hydrogen production, reducing the reliability and economy of the system. Therefore, it is urgent to develop a minute-level intelligent scheduling method that can adaptively adjust the time window, realize collaborative optimization of the hydrogen storage system and the hydrogen production equipment, and has a smooth transition function in the event of a fault. SUMMARY

[0005] The embodiments of the present application provide a minute-level intelligent scheduling method and system for renewable energy hydrogen-based energy, which can solve the problems in the prior art.

[0006] The first aspect of the embodiments of the present application is,

[0007] A minute-level intelligent scheduling method for renewable energy hydrogen-based energy is provided, comprising:

[0008] Obtaining real-time power generation data of renewable energy power generation equipment, operation parameter data of water electrolysis hydrogen production equipment and hydrogen storage amount data of hydrogen storage system;

[0009] According to the real-time power generation data, the power generation fluctuation frequency is calculated, the length of the dynamic time window is adaptively adjusted based on the power generation fluctuation frequency, the power generation data is segmented and processed using the dynamic time window, and the fluctuation characteristic sequence is generated;

[0010] The minute-level intelligent scheduling model is constructed based on the fluctuation characteristic sequence, the operation parameter data and the hydrogen storage amount data, initial hydrogen production power adjustment instructions are generated, the mutual feedback correction parameters between the hydrogen storage system and the water electrolysis hydrogen production equipment are calculated, the mutual feedback correction parameters are used for compensating and optimizing the initial hydrogen production power adjustment instructions, and the optimized hydrogen production power adjustment instructions are obtained;

[0011] The optimized hydrogen production power adjustment instructions are sent to the control system of the water electrolysis hydrogen production equipment, the water electrolysis hydrogen production equipment is controlled to operate, and the operation state of the hydrogen production system is monitored. When an abnormality is detected, the optimal power transfer path is calculated according to the fluctuation characteristic sequence, the power generation power is transferred to the power grid according to the optimal power transfer path, and the water electrolysis hydrogen production equipment is controlled to perform the stepwise load reduction shutdown.

[0012] In an alternative embodiment,

[0013] The power generation fluctuation frequency is calculated according to the real-time power generation data, the length of the dynamic time window is adaptively adjusted based on the power generation fluctuation frequency, the real-time power generation data is segmented by using the dynamic time window, and the fluctuation characteristic sequence is generated, including:

[0014] The frequency component energy spectrum is obtained by wavelet transform on the real-time power generation data, the energy values of each frequency interval are calculated based on the frequency component energy spectrum, and the frequency interval with an energy value greater than a preset energy threshold is determined as a dominant fluctuation frequency interval;

[0015] The frequency characteristic coefficient is calculated according to the energy distribution in the dominant fluctuation frequency interval, the frequency characteristic coefficient is input into a pre-constructed dynamic time window adaptive model, the model parameters of the dynamic time window adaptive model are optimized by using a multi-objective optimization algorithm, and the length of the dynamic time window is calculated based on the optimized model parameters;

[0016] The real-time power generation data is segmented by using a sliding window based on the length of the dynamic time window, and a time window power sequence is obtained. The power mean value sequence is calculated by calculating the mean value of the time window power sequence;

[0017] The power fluctuation amplitude feature is calculated by calculating the standard deviation of the power mean value sequence, the power fluctuation frequency feature is calculated by calculating the zero-crossing rate of the power mean value sequence, and the power fluctuation change feature is calculated by calculating the change rate of the power mean value sequence;

[0018] The power fluctuation amplitude feature, the power fluctuation frequency feature and the power fluctuation change feature are combined to construct the fluctuation characteristic sequence.

[0019] In an alternative embodiment,

[0020] Segmenting the real-time power generation data by using a sliding window based on the length of the dynamic time window includes:

[0021] The real-time power generation data is processed by sliding segmentation based on a dynamic time window length, a power change rate of adjacent sampling points is calculated, a data fluctuation evaluation function is constructed based on the power change rate, a data integrity of the time window is evaluated by the data fluctuation evaluation function, and a time window with a data integrity lower than a preset integrity threshold is determined as a to-be-supplemented window;

[0022] Feature extraction is performed on power generation data of the to-be-supplemented window, a feature vector including a power mean value and a fluctuation period is obtained, a multi-layer similarity matching rule is constructed based on the feature vector, and a candidate supplement sequence is screened from historical power generation data by using the multi-layer similarity matching rule, wherein the candidate supplement sequence has a similarity greater than a similarity threshold with power generation features of the to-be-supplemented window;

[0023] An initial weight coefficient of a dynamic time warping algorithm is set according to a power change rate in the to-be-supplemented window, a similarity between each data segment in the candidate supplement sequence and the to-be-supplemented window is calculated based on the initial weight coefficient, and a data segment with the highest similarity is selected as supplement data;

[0024] A weight decay function is constructed based on the power change rate, and weighted smoothing processing is performed on original power generation data in the to-be-supplemented window and the supplement data according to the weight decay function, to generate supplemented time window data;

[0025] A sequence continuity index and a dynamic consistency index of the supplemented time window data are calculated, a weighted result of the continuity index and the consistency index is compared with a preset evaluation threshold, when the weighted result is less than the preset evaluation threshold, parameters of the weight decay function are adjusted and the weighted smoothing processing step is executed again until the weighted result is greater than the preset evaluation threshold, and a time window power sequence meeting a requirement is output.

[0026] In an optional embodiment,

[0027] A mutual feedback correction parameter between the hydrogen storage system and the water electrolysis hydrogen production equipment is calculated, and the initial hydrogen production power regulation instruction is compensated and optimized by using the mutual feedback correction parameter to obtain an optimized hydrogen production power regulation instruction, including:

[0028] A mapping model of pressure and electrolysis efficiency is established based on a pressure value in the hydrogen storage amount data, a deviation of the current pressure relative to a rated pressure is calculated to obtain a pressure influence factor, a mapping model of temperature and hydrogen production efficiency is established based on a temperature value in the operation parameter data, a deviation of the current temperature relative to a rated temperature is calculated to obtain a temperature coupling factor, and the pressure influence factor and the temperature coupling factor are multiplied and combined with a device performance attenuation value to obtain the mutual feedback correction parameter;

[0029] Compensate and adjust the initial hydrogen production power adjustment instruction according to the mutual feedback correction parameter to obtain a corrected power instruction; meanwhile, compensate the maximum power limit of the electrolytic cell to obtain a corrected power upper limit, and compensate the power ramp rate limit of the electrolytic cell to obtain a corrected ramp rate limit;

[0030] Determine a smoothing factor based on the time variation rate of the mutual feedback correction parameter, weight average the corrected power instruction and the power adjustment instruction of the last control period according to the smoothing factor to generate a transition instruction; judge whether the transition instruction meets the corrected power upper limit and the corrected ramp rate limit, if yes, output the transition instruction as the optimized hydrogen production power adjustment instruction, if not, limit the transition instruction within the allowed range and output it as the optimized hydrogen production power adjustment instruction.

[0031] In an optional embodiment,

[0032] Monitor the running state of the hydrogen production system, and when an abnormality is detected, calculate an optimal power transfer path according to the fluctuation feature sequence, transfer the power generation power to the power grid according to the optimal power transfer path, and control the water electrolysis hydrogen production equipment to perform a stepwise load shedding shutdown including:

[0033] Collect the membrane voltage fluctuation rate, hydrogen production efficiency decay rate and system pressure mutation rate of the water electrolysis hydrogen production equipment, and calculate the first-order change rate and the second-order change rate; when the first-order change rate or the second-order change rate exceeds the corresponding early warning threshold, trigger the power transfer mechanism;

[0034] Obtain the power generation power sequence of the water electrolysis hydrogen production equipment, and extract the power fluctuation feature; combine the first-order change rate, the second-order change rate and the power fluctuation feature, and use a deep reinforcement learning algorithm to generate an initial power transfer path;

[0035] Collect stress monitoring data of the water electrolysis hydrogen production equipment, calculate a stress state index based on the stress monitoring data; divide the load shedding process into a buffer load shedding section, a rapid load shedding section and a protection load shedding section according to the stress state index; establish a dynamic mapping relationship between the stress state index and the load shedding rate of each load shedding section to generate a stepwise load shedding strategy;

[0036] Wavelet-decompose the pressure signal of the hydrogen storage system to extract disturbance components and trend components; identify pressure pulsation features based on the disturbance components, and predict pressure change trends based on the trend components; combine the pressure pulsation features and the pressure change trends with the stress state index to obtain system evolution features;

[0037] Use the system evolution features to real-time correct the initial power transfer path to obtain an optimal power transfer path; control the power generation power to be transferred to the power grid according to the optimal power transfer path, and control the water electrolysis hydrogen production equipment to perform load shedding based on the stepwise load shedding strategy;

[0038] The stress monitoring data and pressure signal feedback during the load shedding process are used to update the system evolution characteristics, the optimal power transfer path and the load shedding rate in the stepwise load shedding strategy are optimized in real time based on the updated system evolution characteristics, and the optimization process is repeated until the electrolytic water hydrogen production equipment is safely shut down.

[0039] In an alternative embodiment,

[0040] The power generation sequence of the electrolytic water hydrogen production equipment is obtained, and the power fluctuation characteristics are extracted; the first-order change rate, the second-order change rate, and the power fluctuation characteristics are combined, and a deep reinforcement learning algorithm is used to generate an initial power transfer path, which includes:

[0041] The power fluctuation characteristics are input into the strategy network and the value network;

[0042] The first hidden layer of the strategy network uses an exponential linear unit function to nonlinearly combine the power fluctuation characteristics to obtain a power feature vector, the second hidden layer processes the power feature vector through a residual connection to obtain a power adjustment action, and the output layer generates an action probability of power transfer based on the power adjustment action;

[0043] The long short-term memory unit of the value network stores the power fluctuation characteristics to form a historical power sequence, and the historical power sequence is importance filtered through a gating mechanism to obtain a state evaluation result;

[0044] The power change acceleration coefficient, the power fluctuation amplitude coefficient, and the power response time coefficient are calculated according to the power fluctuation characteristics, the power fluctuation constraint rule is constructed based on the power change acceleration coefficient, the power fluctuation amplitude coefficient, and the power response time coefficient, and the power fluctuation constraint rule and the action probability of power transfer are combined to form a multi-objective optimization function;

[0045] The reward value is calculated using the multi-objective optimization function and the state evaluation result, the time series difference error is constructed according to the reward value, the trust domain constraint is established based on the time series difference error, the parameters of the strategy network are optimized under the trust domain constraint, and the initial power transfer path is generated through the optimized strategy network parameters.

[0046] In an alternative embodiment,

[0047] According to the stress state index, the load shedding process is divided into a buffer load shedding section, a fast load shedding section, and a protection load shedding section; a dynamic mapping relationship between the stress state index and the load shedding rate of each load shedding section is established, and a stepwise load shedding strategy is generated, which includes:

[0048] According to the stress state index, a state characteristic matrix is established, and a characteristic root combination of the state characteristic matrix is calculated; according to the characteristic root combination, the system stable state is judged, and the stable margin is calculated based on the system stable state; according to the stable margin, the load shedding process is divided into a buffer load shedding section, a rapid load shedding section and a protection load shedding section;

[0049] For each load shedding section, a response function of the load shedding rate and the dynamic stress is established; according to the response function, the initial load shedding rate of the buffer load shedding section, the rapid load shedding section and the protection load shedding section is determined;

[0050] A dynamic equation of the stress state index and the load shedding rate is established, wherein a nonlinear term is introduced to represent the system bifurcation characteristic; the initial load shedding rate is substituted into the dynamic equation to solve the bifurcation parameter;

[0051] An adaptive adjustment equation of the stress deviation is established; a compensation amount is calculated according to the deviation of the stress state index and a preset stress threshold; the compensation amount is input into the adaptive adjustment equation to dynamically adjust the bifurcation parameter; based on the adjusted bifurcation parameter, the dynamic equation is solved to obtain the load shedding rate of each load shedding section;

[0052] A neural network model containing multiple hidden layers is constructed, historical load shedding data is input into the neural network model for training, a mapping relationship between the load shedding rate and the stress fluctuation is established by using the trained neural network model, a load shedding rate correction value is calculated based on the mapping relationship, and the load shedding rate of each load shedding section is corrected in combination with the load shedding rate correction value;

[0053] A load shedding time allocation function is established, and the switching time and the duration of each load shedding section are calculated according to the time allocation function; the switching time, the duration and the corrected load shedding rate are combined to obtain a stepwise load shedding strategy.

[0054] The second aspect of the embodiment of the application,

[0055] A system is provided, comprising:

[0056] The first unit is configured to acquire real-time power generation data of a renewable energy power generation device, operation parameter data of a water electrolysis hydrogen production device, and hydrogen storage amount data of a hydrogen storage system;

[0057] The second unit is configured to calculate a power generation fluctuation frequency according to the real-time power generation data, adaptively adjust the length of a dynamic time window based on the power generation fluctuation frequency, and perform segmented processing on the power generation data by using the dynamic time window to generate a fluctuation feature sequence;

[0058] The third unit is configured to construct a minute-level intelligent scheduling model based on the fluctuation feature sequence, the operation parameter data and the hydrogen storage amount data, generate an initial hydrogen production power adjustment instruction, calculate a mutual feedback correction parameter between the hydrogen storage system and the water electrolysis hydrogen production equipment, compensate and optimize the initial hydrogen production power adjustment instruction by using the mutual feedback correction parameter, and obtain an optimized hydrogen production power adjustment instruction.

[0059] The fourth unit is configured to send the optimized hydrogen production power adjustment instruction to a control system of the water electrolysis hydrogen production equipment, control the water electrolysis hydrogen production equipment to operate, and monitor the operation state of the hydrogen production system, and when an abnormality is detected, calculate an optimal power transfer path according to the fluctuation feature sequence, transfer power generation power to the power grid according to the optimal power transfer path, and control the water electrolysis hydrogen production equipment to perform stepwise load shedding and shutdown.

[0060] The third aspect of the embodiment of the present application,

[0061] An electronic device is provided, comprising:

[0062] a processor;

[0063] a memory for storing processor-executable instructions;

[0064] The processor is configured to invoke the instructions stored in the memory to perform the method described above.

[0065] The fourth aspect of the embodiment of the present application,

[0066] A computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0067] In the embodiment, by adaptively adjusting the dynamic time window length and generating the fluctuation feature sequence, the fluctuation characteristics of renewable energy power generation can be accurately captured, the identification and prediction ability of power generation fluctuation is improved, the hydrogen production system can more flexibly cope with power generation fluctuation, and the stability and reliability of the system are improved. The mutual feedback correction parameter is used to compensate and optimize the initial hydrogen production power adjustment instruction, the coordinated operation between the hydrogen storage system and the water electrolysis hydrogen production equipment is realized, the problem of energy waste caused by parameter mismatch in the traditional scheduling method is effectively solved, and the energy utilization efficiency and economy of the system are improved. An abnormality processing mechanism based on the optimal power transfer path is designed, when the system appears abnormal, the power generation power can be intelligently transferred to the power grid and the stepwise load shedding and shutdown are performed, the risk of equipment damage and system collapse is avoided, the energy loss is minimized, and the safety and toughness of the entire hydrogen production system are enhanced. BRIEF DESCRIPTION OF DRAWINGS

[0068] Figure 1A flowchart of a minute-level intelligent scheduling method of a renewable energy hydrogen-based energy source according to an embodiment of the present application is shown in

[0069] Figure 2 A reliability and stability evaluation analysis diagram of supplementary data according to an embodiment of the present application is shown in

[0070] Figure 3 A system energy efficiency comparison diagram according to an embodiment of the present application is shown in

[0071] Figure 4 A dynamic adjustment simulation diagram of the load shedding rate of each section in the stepwise load shedding strategy according to an embodiment of the present application is shown in DETAILED DESCRIPTION

[0072] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in connection with the accompanying drawings of the embodiments of the present application. Obviously, the described embodiments are only a part but not all of the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0073] The technical solutions of the present application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and some embodiments may not be described again for the same or similar concepts or processes.

[0074] Figure 1 A flowchart of a minute-level intelligent scheduling method of a renewable energy hydrogen-based energy source according to an embodiment of the present application is shown in Figure 1 The method comprises the following steps.

[0075] Real-time power generation data of a renewable energy power generation device, operation parameter data of a water electrolysis hydrogen production device, and hydrogen storage amount data of a hydrogen storage system are acquired.

[0076] A power generation fluctuation frequency is calculated according to the real-time power generation data, a length of a dynamic time window is adaptively adjusted based on the power generation fluctuation frequency, the power generation data is segmented and processed by using the dynamic time window, and a fluctuation feature sequence is generated.

[0077] A minute-level intelligent scheduling model is constructed based on the fluctuation feature sequence, the operation parameter data, and the hydrogen storage amount data, an initial hydrogen production power adjustment instruction is generated, a mutual feedback correction parameter between the hydrogen storage system and the water electrolysis hydrogen production device is calculated, the initial hydrogen production power adjustment instruction is compensated and optimized by using the mutual feedback correction parameter, and an optimized hydrogen production power adjustment instruction is obtained.

[0078] The optimized hydrogen production power adjustment instruction is sent to a control system of the water electrolysis hydrogen production equipment, the water electrolysis hydrogen production equipment is controlled to run, and the running state of the hydrogen production system is monitored; when an abnormality is detected, an optimal power transfer path is calculated according to the fluctuation characteristic sequence, power generation power is transferred to the power grid according to the optimal power transfer path, and the water electrolysis hydrogen production equipment is controlled to perform stepwise load shedding and shutdown.

[0079] In an alternative embodiment,

[0080] The power generation fluctuation frequency is calculated according to real-time power generation data, the length of the dynamic time window is adaptively adjusted based on the power generation fluctuation frequency, the power generation data is segmented using the dynamic time window, and the fluctuation characteristic sequence is generated, including:

[0081] The frequency component energy spectrum is obtained by wavelet transform of the real-time power generation data, the energy value of each frequency interval is calculated based on the frequency component energy spectrum, and the frequency interval with an energy value greater than a preset energy threshold is determined as a dominant fluctuation frequency interval;

[0082] The frequency characteristic coefficient is calculated according to the energy distribution in the dominant fluctuation frequency interval, the frequency characteristic coefficient is input into a pre-constructed dynamic time window adaptive model, the model parameters of the dynamic time window adaptive model are optimized by a multi-objective optimization algorithm, and the length of the dynamic time window is calculated based on the optimized model parameters;

[0083] The real-time power generation data is segmented by a sliding window based on the length of the dynamic time window, and a time window power sequence is obtained, and the mean value of the time window power sequence is calculated to obtain a power mean value sequence;

[0084] The standard deviation of the power mean value sequence is calculated to obtain a power fluctuation amplitude feature, the zero-crossing rate of the power mean value sequence is calculated to obtain a power fluctuation frequency feature, and the change rate of the power mean value sequence is calculated to obtain a power fluctuation change feature;

[0085] The power fluctuation amplitude feature, the power fluctuation frequency feature, and the power fluctuation change feature are combined to construct the fluctuation characteristic sequence.

[0086] For example, the real-time power generation data is wavelet transformed to obtain a frequency component energy spectrum. Specifically, a multi-resolution wavelet decomposition method is used, a db4 wavelet basis is selected, and the power generation data is decomposed by 5 layers to obtain wavelet coefficients in different frequency intervals. The energy values of the frequency intervals are obtained by calculating the square sum of the wavelet coefficients of each layer. For example, for power generation data with a sampling frequency of 1 Hz, after 5-layer wavelet decomposition, energy values of five frequency intervals of 0-0.03 Hz, 0.03-0.06 Hz, 0.06-0.12 Hz, 0.12-0.25 Hz, and 0.25-0.5 Hz can be obtained.

[0087] The preset energy threshold is set as 15% of the total energy, and when the energy value of a certain frequency interval exceeds the threshold, it is identified as a dominant fluctuation frequency interval. For example, if the energy proportion of the 0.06-0.12 Hz frequency interval is 25%, and the energy proportion of the 0.12-0.25 Hz frequency interval is 20%, these two frequency intervals are determined as the dominant fluctuation frequency intervals.

[0088] According to the energy distribution in the dominant fluctuation frequency interval, the frequency characteristic coefficient is calculated. Specifically, the sum of the product of the center frequency of the dominant fluctuation frequency interval and its energy value is divided by the total energy to obtain the frequency characteristic coefficient. For example, if the dominant fluctuation frequency interval is 0.06-0.12 Hz and 0.12-0.25 Hz, the center frequencies are 0.09 Hz and 0.185 Hz, the energy values are 250 units and 200 units, and the total energy is 1000 units, then the frequency characteristic coefficient is (0.09*250+0.185*200) / 1000=0.06.

[0089] Based on the calculated frequency characteristic coefficient, the dynamic time window length is determined using a dynamic time window adaptive model. The model uses a BP neural network structure, the input layer is the frequency characteristic coefficient, the hidden layer contains 10 neurons, and the output layer is the time window length. Optionally, the model parameters are optimized by a particle swarm optimization algorithm, and the optimization objectives include the adaptability and computational efficiency of the window length. The particle swarm size is set to 50, the maximum iteration number is 100, the inertia weight is linearly decreased from 0.9 to 0.4, and the acceleration factors are both 2.0. For example, when the frequency characteristic coefficient is 0.06, the optimized model calculates the time window length of 180 seconds.

[0090] The real-time power generation data is segmented by sliding window using the calculated dynamic time window length. The window sliding step is set to 1 / 4 of the window length, and for a window length of 180 seconds, the sliding step is 45 seconds. The arithmetic mean of the power data in each time window is calculated to form a power mean sequence. For example, for the power generation data of a certain photovoltaic power station, if the power mean in the first time window (0-180 seconds) is 500 kW, the power mean in the second time window (45-225 seconds) is 520 kW, and so on, the power mean sequence [500, 520, 535, 510, 490,...] kW can be obtained.

[0091] Based on the power mean sequence, three kinds of fluctuation characteristics are calculated. First, the power fluctuation amplitude characteristic, i.e. the standard deviation of the power mean sequence, is calculated. The power mean sequence is subtracted from its average value, the sum of squared deviations is calculated, and the square root of the sequence length is divided to obtain the standard deviation. For example, for the power mean sequence [500, 520, 535, 510, 490,...] kW, if the standard deviation calculation result is 15 kW, then the power fluctuation amplitude characteristic is 15 kW.

[0092] Secondly, the power fluctuation frequency characteristic, i.e. the zero crossing rate of the power mean sequence, is calculated. The power mean sequence is subtracted from its average value, the number of times the sequence value changes from negative to positive or from positive to negative is counted, and the sequence length is divided to obtain the zero crossing rate. For example, for the power sequence [-11, 9, 24, -1, -21,...] kW after mean subtraction, if there are 12 zero crossings in a sequence of 100 points, then the zero crossing rate is 0.12.

[0093] Finally, the power fluctuation change characteristic, i.e. the change rate of the power mean sequence, is calculated. The difference between adjacent power means is calculated, and the average of the absolute values is obtained to obtain the average change rate. For example, for the power mean sequence [500, 520, 535, 510, 490,...] kW, the absolute value of the difference between adjacent points is [20, 15, 25, 20,...] kW, and if the average is 18 kW, then the power fluctuation change characteristic is 18 kW.

[0094] The above three characteristics are combined to construct a fluctuation characteristic sequence. For each time window, the power fluctuation amplitude characteristic, the power fluctuation frequency characteristic and the power fluctuation change characteristic are combined into a three-dimensional feature vector. If the power fluctuation amplitude characteristic of a certain time window is 15 kW, the power fluctuation frequency characteristic is 0.12, and the power fluctuation change characteristic is 18 kW, then the fluctuation characteristic vector of the window is [15, 0.12, 18]. With the sliding of the time window, a series of fluctuation characteristic vectors can be obtained to form a complete fluctuation characteristic sequence.

[0095] In this embodiment, efficient utilization of renewable energy generation can be achieved, the response capability of the water electrolysis hydrogen production system is improved, and it can adapt to the real-time fluctuation characteristics of renewable energy generation. By calculating the power generation fluctuation frequency and dynamically adjusting the time window, the short-term changes of the generated power can be accurately captured, making the scheduling decision more flexible and improving the adaptability of the system to fluctuating power. The minute-level intelligent scheduling model makes the hydrogen production power regulation more precise, reduces the energy waste and system instability problems caused by the lag of traditional scheduling methods. Through the mutual feedback correction parameter optimization scheduling instruction, the energy flow between the hydrogen storage system and the hydrogen production equipment can be effectively coordinated, avoiding the overloading of the hydrogen storage system or the unstable operation of the hydrogen production equipment, and improving the overall operation efficiency. When the system detects an anomaly, the optimal power transfer path is calculated based on the fluctuation feature sequence, so that the generated power can be timely and effectively distributed to the power grid or other loads, avoiding energy waste or power grid impact caused by sudden failure of the hydrogen production equipment. Through the step-by-step load shedding shutdown strategy, the hydrogen production load can be gradually reduced during power adjustment, reducing the loss caused by sudden shutdown or severe fluctuations of the equipment, prolonging the service life of the equipment, and ensuring the safe and stable operation of the system.

[0096] In an alternative embodiment,

[0097] The real-time power generation data is segmented based on the dynamic time window length, including:

[0098] The real-time power generation data is segmented based on the dynamic time window length, the power change rate of adjacent sampling points is calculated, the data fluctuation evaluation function is constructed based on the power change rate, the data integrity of the time window is evaluated through the data fluctuation evaluation function, and the time window with a data integrity lower than a preset integrity threshold is determined as a to-be-supplemented window;

[0099] The power generation data of the to-be-supplemented window is feature extracted, a feature vector including a power mean and a fluctuation period is obtained, a multi-layer similarity matching rule is constructed based on the feature vector, and the multi-layer similarity matching rule is used to filter data segments from historical power generation data, which have a similarity greater than a similarity threshold with the power generation characteristics of the to-be-supplemented window to form a candidate supplement sequence;

[0100] The initial weight coefficient of the dynamic time warping algorithm is set according to the power change rate in the to-be-supplemented window, the similarity between each data segment in the candidate supplement sequence and the to-be-supplemented window is calculated based on the initial weight coefficient, and the data segment with the highest similarity is selected as the supplement data;

[0101] A weight decay function is constructed based on the power change rate, and the original power generation data in the to-be-supplemented window and the supplement data are weighted and smoothed according to the weight decay function to generate supplemented time window data;

[0102] The sequence continuity index and dynamic consistency index of the supplemented time window data are calculated, a weighted result of the continuity index and the consistency index is compared with a preset evaluation threshold, when less than the preset evaluation threshold, the parameters of the weight decay function are adjusted and the weighted smoothing processing step is executed again until greater than the preset evaluation threshold, and a time window power sequence meeting the requirements is output.

[0103] Exemplarily, first, real-time power generation data is processed by sliding segmentation based on a dynamic time window length. Specifically, the initial time window length is set to 30 minutes, and the sliding step is 5 minutes. For the data in each time window, the power change rate of adjacent sampling points is calculated. Assuming that the power value at time t is P(t), and the power value at time t+1 is P(t+1), the power change rate can be expressed as the ratio of the power difference between the two times to the time interval. For example, if P(t) = 100 kW, P(t+1) = 105 kW, and the sampling interval is 1 minute, the power change rate is 5 kW / min.

[0104] A data fluctuation evaluation function is constructed based on the power change rate. The function comprehensively considers the average, maximum, and standard deviation of the power change rate in the window to obtain a data integrity index of the time window. For example, in a certain 30-minute window, the average power change rate is 3 kW / min, the maximum power change rate is 8 kW / min, and the standard deviation is 1.5 kW / min. The data integrity is 0.85 obtained by weighted calculation. If the preset integrity threshold is 0.9, the window is determined as a window to be supplemented.

[0105] The power generation data of the window to be supplemented is processed to extract features and obtain a feature vector. The feature vector includes: power mean, standard deviation, fluctuation period, peak value number, etc. For example, the feature vector of a certain window to be supplemented is: power mean 120 kW, standard deviation 15 kW, main fluctuation period 10 minutes, and peak value number 3.

[0106] A multi-layer similarity matching rule is constructed based on the feature vector. The rule first filters according to the power mean, retaining historical data segments with a power mean difference of no more than 20% from the power mean of the window to be supplemented; then filters according to the fluctuation period, retaining segments with a fluctuation period difference of no more than 30%; and finally considers the similarity of the standard deviation and the peak value number. Through this multi-layer filtering, data segments with a similarity greater than 0.85 (preset similarity threshold) are selected from the historical database to form a candidate supplement sequence.

[0107] The initial weight coefficient of the dynamic time warping algorithm is set according to the power change rate in the window to be supplemented. If the power change rate in the window is large, it indicates that the power generation fluctuates violently, and a higher time elasticity coefficient is set, for example, when the average power change rate is 5 kW / min, the initial weight coefficient is set to 0.7; if the power change rate is small, a lower time elasticity coefficient is set, for example, when the average power change rate is 1 kW / min, the initial weight coefficient is set to 0.3.

[0108] The similarity of each data segment in the candidate supplement sequence to the window to be supplemented is calculated based on the initial weight coefficient. For example, there are three data segments A, B and C in the candidate sequence, and the similarity of each data segment to the window to be supplemented is calculated to be 0.92, 0.88 and 0.95 respectively, then the data segment with the highest similarity, i.e. data segment C, is selected as the supplement data.

[0109] A weight decay function is constructed based on the power change rate. The function determines the weight distribution of the original data and the supplement data in the fusion process. At the boundary of the window to be supplemented, the weight of the original data is higher; in the middle part, the weight of the supplement data is higher. For example, for the part with a large power change rate, the weight of the original data may be 0.3 and the weight of the supplement data may be 0.7; for the part with a small power change rate, the weight of the original data may be 0.6 and the weight of the supplement data may be 0.4.

[0110] The original power generation data and the supplement data in the window to be supplemented are weighted and smoothed according to the weight decay function. For example, at a certain time, the original data is 95 kW, the supplement data is 105 kW, the weight of the original data is 0.4, and the weight of the supplement data is 0.6, then the smoothed data is 95*0.4+105*0.6=101 kW. All data points in the window are processed in this way to generate the supplemented time window data.

[0111] The sequence continuity index and the dynamic consistency index of the supplemented time window data are calculated. The sequence continuity index evaluates the smoothing degree of the data before and after supplementation, for example, the continuity index is calculated to be 0.88; the dynamic consistency index evaluates the consistency of the change trend of the supplement data and the original data, for example, the consistency index is calculated to be 0.92. Assuming that the weight of the continuity index is 0.4 and the weight of the consistency index is 0.6, then the weighted result is 0.88*0.4+0.92*0.6=0.904.

[0112] The weighted result of the continuity and consistency indicators is compared with a preset evaluation threshold (e.g., 0.9). If the weighted result 0.904 is greater than the preset evaluation threshold 0.9, the current supplemented time window power sequence is output; if it is less than the preset evaluation threshold, the parameters of the weight decay function are adjusted. For example, the weights of the original data are increased by 0.1, and then the weighted smoothing process is repeated until the weighted result is greater than the preset evaluation threshold.

[0113] The above method can effectively identify incomplete windows in power generation data and intelligently supplement them using historical data, ensuring that the output power sequence meets the requirements of continuity and consistency, and providing reliable data support for subsequent power generation forecasting and grid dispatch.

[0114] Figure 2 The reliability and stability evaluation analysis diagrams for supplementary data in the embodiments of the present invention are as follows: Figure 2 As shown in the chart, this chart comprehensively illustrates the reliability and stability performance of different supplementation methods under long-term operating conditions. It includes a one-year reliability trend analysis of the supplementation data. From the long-term trend analysis, this technical solution (circled marker) maintained high supplementation data reliability throughout the year, consistently remaining between 88% and 91%, demonstrating excellent long-term stability. Particularly noteworthy is that even during the most challenging periods of power system operation—high temperature and high load in summer (June-August) and low temperature in winter (November-January of the following year)—its reliability index only fluctuated slightly, reaching a high level of 88% even at its lowest point in July. In contrast, the sliding window regression method (triangle marked) showed significant fluctuations in reliability throughout the year, decreasing from 81% in winter and spring to 72% in summer, a fluctuation of 9 percentage points; the multi-model integration method (square marked) performed moderately, with its reliability index fluctuating between 79% and 86%.

[0115] The two key time points marked in the chart—"equipment upgrade" in July and "algorithm optimization" in September—had different impacts on the various methods. After the equipment upgrade, the sliding window regression method showed a significant decline in performance, indicating its poor adaptability to equipment changes; while this technical solution demonstrated good adaptability, with its reliability indicators remaining stable. After algorithm optimization, the performance of all three methods improved, but the improvement of this technical solution was the most significant, indicating that it more effectively utilized the performance gains brought about by algorithm optimization.

[0116] The prior art uses a fixed time window to segment the power generation data, which is difficult to adapt to the fluctuation characteristics of renewable energy, resulting in insufficient data integrity and affecting scheduling accuracy. Traditional data completion methods mainly rely on linear interpolation or historical mean filling, ignoring the power change trend, which may introduce large errors and reduce data reliability. The application adjusts the time window length adaptively and constructs a data fluctuation evaluation function based on the power change rate to accurately identify windows with insufficient data integrity. Combined with feature extraction and multi-layer similarity matching rules, similar fragments with high similarity are selected from historical data for supplementation to avoid error accumulation caused by traditional interpolation methods. In addition, the dynamic time warping algorithm is introduced to optimize the matching results and improve the accuracy of the supplemented data. A weight decay function is used to weight and smooth the supplemented data and the original data to ensure a natural transition of the data, and the data quality is evaluated by continuity and consistency indicators to dynamically adjust the weight parameters to ensure that the final data meets the integrity and consistency requirements. Compared with traditional methods, this scheme more flexibly adapts to the dynamic changes of power generation data, provides more reliable input for intelligent scheduling, and improves the operation efficiency and stability of the hydrogen production system.

[0117] In an alternative embodiment,

[0118] The mutual feedback correction parameter between the hydrogen storage system and the water electrolysis hydrogen production equipment is calculated, and the initial hydrogen production power regulation instruction is compensated and optimized by using the mutual feedback correction parameter to obtain an optimized hydrogen production power regulation instruction.

[0119] Based on the pressure value in the hydrogen storage amount data, a mapping model of pressure and electrolysis efficiency is established, the deviation of the current pressure relative to the rated pressure is calculated, and a pressure influence factor is obtained. Based on the temperature value in the operating parameter data, a mapping model of temperature and hydrogen production efficiency is established, the deviation of the current temperature relative to the rated temperature is calculated, and a temperature coupling factor is obtained. The pressure influence factor and the temperature coupling factor are multiplied, and combined with the equipment performance attenuation value to obtain the mutual feedback correction parameter.

[0120] The initial hydrogen production power regulation instruction is compensated and adjusted according to the mutual feedback correction parameter to obtain a corrected power instruction; at the same time, the maximum power limit of the electrolytic cell is compensated to obtain a corrected power upper limit, and the power ramp rate limit of the electrolytic cell is compensated to obtain a corrected ramp rate limit.

[0121] Based on the time variation rate of the mutual feedback correction parameter, a smoothing factor is determined, the corrected power instruction and the power regulation instruction of the last control period are weighted and averaged according to the smoothing factor to generate a transition instruction; it is judged whether the transition instruction meets the corrected power upper limit and the corrected ramp rate limit, if it meets, the transition instruction is output as the optimized hydrogen production power regulation instruction, if it does not meet, the transition instruction is limited within the allowed range and then output as the optimized hydrogen production power regulation instruction.

[0122] The embodiment provides a method for calculating mutual feedback correction parameters between a hydrogen storage system and a water electrolysis hydrogen production device, and compensating and optimizing an initial hydrogen production power adjustment instruction by using the parameters.

[0123] Specifically, a mapping model of pressure and electrolysis efficiency is established based on pressure values in hydrogen storage amount data. In actual application, the electrolysis efficiency of the water electrolysis hydrogen production device changes with the change of the pressure of the hydrogen storage system. By collecting real-time pressure data of the hydrogen storage system, the deviation of the current pressure relative to the rated pressure can be calculated. For example, when the rated pressure of the hydrogen storage system is 30 MPa and the current measured pressure is 25 MPa, the pressure deviation is -5 MPa. According to the pre-established pressure-efficiency mapping relationship, the pressure influence factor corresponding to the pressure deviation can be determined. Specifically, it can be realized by table lookup, for example, when the pressure deviation is -5 MPa, the corresponding pressure influence factor is 1.05, indicating that the electrolysis efficiency is increased by 5% under the condition of lower than the rated pressure.

[0124] A mapping model of temperature and hydrogen production efficiency is established based on temperature values in operation parameter data. In the process of water electrolysis hydrogen production, temperature is a key factor affecting hydrogen production efficiency. By collecting real-time temperature data of the electrolytic cell, the deviation of the current temperature relative to the rated temperature can be calculated. For example, when the rated working temperature of the electrolytic cell is 75℃ and the current measured temperature is 65℃, the temperature deviation is -10℃. According to the pre-established temperature-efficiency mapping relationship, the temperature coupling factor corresponding to the temperature deviation can be determined. For example, when the temperature deviation is -10℃, the corresponding temperature coupling factor is 0.92, indicating that the hydrogen production efficiency is reduced by 8% under the condition of lower than the rated temperature.

[0125] The pressure influence factor and the temperature coupling factor are multiplied, and combined with the device performance attenuation value to obtain the mutual feedback correction parameter. The device performance attenuation value reflects the performance decline of the electrolytic cell with the increase of the use time. For example, it is assumed that the electrolytic cell has been operated for 5000 hours, and the device performance attenuation value obtained by analyzing the historical data is 0.98. Then, the mutual feedback correction parameter is calculated as: 1.05x0.92x0.98=0.947, indicating that the actual hydrogen production capacity is 94.7% of the theoretical value after considering the pressure, temperature and device attenuation.

[0126] The initial hydrogen production power adjustment instruction is compensated and adjusted according to the mutual feedback correction parameter to obtain a corrected power instruction. It is assumed that the initial power adjustment instruction is 1000 kW, then the corrected power instruction is 1000÷0.947=1056 kW, indicating that the input power needs to be increased to achieve the expected hydrogen production amount.

[0127] The maximum power limit of the electrolyzer is compensated to obtain a revised power upper limit. For example, if the rated maximum power of the electrolyzer is 1500 kW, the revised power upper limit is 1500 x 0.947 = 1420.5 kW, indicating that the actual maximum bearing power of the electrolyzer should be reduced under the current operating conditions. The power ramp rate limit of the electrolyzer is also compensated to obtain a revised ramp rate limit. For example, if the standard ramp rate limit of the electrolyzer is 10 kW / s, the revised ramp rate limit is 10 x 0.947 = 9.47 kW / s, indicating that the power change rate should be reduced under the current conditions to protect the equipment.

[0128] The time variation rate of the mutual feedback correction parameter is determined to determine the smoothing factor. The variation of the mutual feedback correction parameter reflects the fluctuation of the system operating state, and a smoothing mechanism is introduced to avoid frequent changes in the power command. Specifically, the variation rate of the current mutual feedback correction parameter and the last period mutual feedback correction parameter is calculated, if the variation rate is large, the smoothing factor takes a small value; if the variation rate is small, the smoothing factor takes a large value. For example, when the variation rate of the mutual feedback correction parameter is 2%, the corresponding smoothing factor is 0.7.

[0129] The revised power command and the power adjustment command of the last control period are weighted and averaged according to the smoothing factor to generate a transition command. Assuming that the power adjustment command of the last period is 980 kW, the current revised power command is 1056 kW, and the smoothing factor is 0.7, the transition command is calculated as: 1056 x 0.7 + 980 x 0.3 = 1033.2 kW.

[0130] Finally, it is determined whether the transition command meets the revised power upper limit and the revised ramp rate limit. If it meets, the transition command is output as the optimized hydrogen production power adjustment command; if it does not meet, the transition command is limited within the allowed range and then output. For example, if the actual executed power of the last period is 1000 kW, the current transition command is 1033.2 kW, the revised ramp rate limit is 9.47 kW / s, and the control period is 2 s, the maximum power change allowed in a single period is 9.47 x 2 = 18.94 kW. Since 1033.2 - 1000 = 33.2 kW > 18.94 kW, the ramp rate limit is exceeded, and the final output optimized power adjustment command should be 1000 + 18.94 = 1018.94 kW.

[0131] In this embodiment, by establishing a mapping model of pressure and electrolysis efficiency, temperature and hydrogen production efficiency, accurately calculating the pressure influence factor and temperature coupling factor, and combining the equipment performance attenuation value, the mutual feedback correction parameter is obtained, so that the power regulation instruction can dynamically adapt to the actual running state, and the energy efficiency decline caused by environmental changes is avoided. Through the compensation and adjustment of the initial hydrogen production power regulation instruction by the mutual feedback correction parameter, the accuracy of the power instruction can be optimized, and the maximum power limit and power ramp rate of the electrolytic cell are compensated, which ensures that the power adjustment process not only meets the safety requirements of equipment operation, but also fully utilizes renewable energy within a reasonable range and reduces equipment loss caused by power fluctuations. Based on the time variation rate of the mutual feedback correction parameter, a smoothing factor is introduced to make the power regulation more stable and avoid sudden adjustment impacting the equipment. Finally, by judging whether the transition instruction meets the correction power upper limit and the ramp rate limit, it is ensured that the optimized power regulation instruction not only meets the operation requirements of the hydrogen storage system and the hydrogen production equipment, but also can be dynamically adjusted within the safety range, thereby improving the safety, response speed and energy utilization efficiency of the overall system.

[0132] In an alternative embodiment,

[0133] Monitoring the running state of the hydrogen production system, when an abnormality is detected, calculating an optimal power transfer path according to the fluctuation feature sequence, transferring the power generation power to the power grid according to the optimal power transfer path, and controlling the water electrolysis hydrogen production equipment to execute a stepwise load shedding shutdown including:

[0134] Collecting the membrane voltage fluctuation rate, hydrogen production efficiency attenuation rate and system pressure mutation rate of the water electrolysis hydrogen production equipment, and calculating the first-order change rate and the second-order change rate; when the first-order change rate or the second-order change rate exceeds the corresponding warning threshold, triggering the power transfer mechanism;

[0135] Obtaining the power generation power sequence of the water electrolysis hydrogen production equipment, and extracting the power fluctuation feature; combining the first-order change rate, the second-order change rate and the power fluctuation feature, and using a deep reinforcement learning algorithm to generate an initial power transfer path;

[0136] Collecting stress monitoring data of the water electrolysis hydrogen production equipment, calculating a stress state index based on the stress monitoring data; dividing the load shedding process into a buffer load shedding section, a rapid load shedding section and a protection load shedding section according to the stress state index; establishing a dynamic mapping relationship between the stress state index and the load shedding rate of each load shedding section, and generating a stepwise load shedding strategy;

[0137] Wavelet decomposition is performed on the pressure signal of the hydrogen storage system to extract disturbance components and trend components; pressure pulsation features are identified based on the disturbance components, and pressure change trends are predicted based on the trend components; the pressure pulsation features and the pressure change trends are combined with the stress state index to obtain system evolution features;

[0138] The system evolution feature is used to correct the initial power transfer path in real time to obtain an optimal power transfer path; the power generation power is controlled to be transferred to the power grid according to the optimal power transfer path, and the water electrolysis hydrogen production equipment is controlled to perform load shedding based on the stepwise load shedding strategy;

[0139] The stress monitoring data and pressure signal feedback during load shedding are used to update the system evolution feature; the optimal power transfer path and the load shedding rate in the stepwise load shedding strategy are optimized in real time based on the updated system evolution feature; the optimization process is repeated until the water electrolysis hydrogen production equipment is safely shut down.

[0140] Exemplarily, the key operating parameters of the water electrolysis hydrogen production equipment are collected in real time by a sensor network, including membrane voltage, hydrogen production flow, system pressure and other data. Based on the collected data, the membrane voltage fluctuation rate, hydrogen production efficiency decay rate and system pressure mutation rate are calculated. Specifically, the membrane voltage fluctuation rate is obtained by dividing the difference between adjacent time points by the time interval; the hydrogen production efficiency decay rate is calculated by the change rate of the hydrogen production amount per unit power; and the system pressure mutation rate is obtained by the change rate of the pressure sensor data.

[0141] Taking the membrane voltage fluctuation rate as an example, the system collects membrane voltage data every 100 milliseconds, calculates the first-order change rate, i.e. the change rate of adjacent two sampling points; and calculates the second-order change rate, i.e. the change speed of the first-order change rate. When the first-order change rate of the membrane voltage exceeds 0.5V / s or the second-order change rate exceeds 0.8V / s 2 , the system determines that it is in an abnormal state, and triggers the power transfer mechanism.

[0142] After triggering the power transfer mechanism, the system obtains the power generation power sequence of the water electrolysis hydrogen production equipment, which is usually the power data points at intervals of 5 seconds within the last 30 minutes. The power fluctuation characteristics are extracted by time-frequency analysis method, including fluctuation amplitude, frequency distribution and fluctuation trend. For example, in a certain abnormal situation, the system records that the power fluctuation amplitude is 8% of the rated power, the main fluctuation frequency is concentrated in the range of 0.05-0.1 Hz, and it shows a gradually increasing trend.

[0143] The system takes the first-order change rate, the second-order change rate and the power fluctuation characteristics as inputs, and generates an initial power transfer path through a pre-trained deep reinforcement learning model. The model uses a double-layer LSTM network structure, the number of nodes in the input layer is the feature dimension, the number of nodes in the hidden layer is 128, and the output layer is the power transfer sequence. The model is trained using historical abnormal handling data, and the system stress in the power transfer process is minimized and the grid stability is maximized as the reward function. The generated initial power transfer path contains a time sequence from the current power to zero power, for example, in a certain abnormal handling, the initial path is to smoothly transfer 10MW power to the grid within 120 seconds.

[0144] Meanwhile, the system collects stress monitoring data of the water electrolysis hydrogen production equipment, including the temperature gradient of the stack, the stress distribution of the membrane, and the pressure difference of the hydrogen and oxygen separator, etc. Based on these data, the system calculates a comprehensive stress state index, which ranges from 0 to 100, where 0 represents a stress-free state and 100 represents a limit stress state. According to the stress state index, the load reduction process is divided into three stages: a buffer load reduction segment with a stress index of 0-30, a load reduction rate of 1-3% of the rated power per second; a fast load reduction segment with a stress index of 31-70, a load reduction rate of 3-8% of the rated power per second; and a protection load reduction segment with a stress index of 71-100, a load reduction rate of 0.5-1% of the rated power per second, and an auxiliary cooling system activated.

[0145] The system performs wavelet decomposition on the hydrogen storage system pressure signal, using db4 wavelet basis function for 5-layer decomposition, to extract high-frequency disturbance components and low-frequency trend components. Based on the disturbance components, the system identifies pressure pulsation characteristics such as pulsation frequency, amplitude, and phase; based on the trend components, the system predicts the pressure change trend in the next 30 seconds. For example, in one abnormal treatment, the system identifies a pressure pulsation frequency of 1.2 Hz, an amplitude of 0.3 MPa, and a trend component showing a pressure rise of 0.05 MPa / s.

[0146] The system combines the pressure pulsation characteristics, pressure change trend, and stress state index to form a system evolution feature vector. This feature vector is input into an adaptive correction module to real-time correct the initial power transfer path, obtaining the optimal power transfer path. The correction process takes into account the grid accommodation capacity, electrolyzer safety limits, and hydrogen storage system pressure state, and updates the transfer path every 5 seconds through a rolling time domain optimization method.

[0147] According to the optimal power transfer path, the system controls the power generation power transfer to the grid through power electronic conversion equipment, and controls the water electrolysis hydrogen production equipment to perform load reduction based on the stepwise load reduction strategy. In specific implementation, power transfer and load reduction are performed cooperatively, for example, in one treatment, the system first reduces 20% of the power at a rate of 2% per second (buffer load reduction segment), then reduces 50% of the power at a rate of 5% per second (fast load reduction segment), and finally reduces the remaining 30% of the power at a rate of 0.8% per second (protection load reduction segment).

[0148] During the load reduction process, the system continuously collects stress monitoring data and pressure signals, and calculates the updated system evolution features every 2 seconds. Based on the updated features, the system optimizes the power transfer path and load reduction rate in real time. For example, when a sudden increase in the temperature gradient of the stack is detected, the system adjusts the load reduction rate of the fast load reduction segment from 5% per second to 4% per second, and extends the protection load reduction segment time.

[0149] The system repeatedly performs the optimization process until the electrolytic water hydrogen production equipment is completely shut down. During the entire process, the system records the change curves of key parameters, including the power transfer curve, the stress state change curve, and the pressure change curve, for subsequent analysis and model optimization. In practical applications, this method successfully handles multiple abnormal situations, shortens the average abnormal handling time by 35%, and reduces the equipment stress peak by 42%, effectively improving the safety and reliability of the hydrogen production system.

[0150] In this embodiment, by collecting the membrane voltage fluctuation rate, hydrogen production efficiency decay rate and system pressure mutation rate, the first and second order change rates are calculated to accurately identify system abnormalities and trigger the power transfer mechanism. Combined with the deep reinforcement learning algorithm, the initial power transfer path is generated, making the power scheduling more adaptive. At the same time, based on the stress monitoring data, a dynamic load shedding mapping relationship is established, and the load shedding process is divided into buffer, fast and protection load shedding sections, realizing more refined load shedding control and avoiding equipment damage or system fluctuations. Wavelet decomposition technology is used to extract the pressure disturbance and trend components of the hydrogen storage system, combined with the stress state index, the system evolution characteristics are constructed, and based on real-time feedback, the power transfer path and load shedding strategy are continuously optimized to ensure that the load shedding rate adapts to the current system state. It can accurately respond to abnormal states, optimize power transfer, improve system safety and stability, and at the same time ensure the long-term operation reliability of the electrolytic water hydrogen production equipment.

[0151] In an alternative embodiment,

[0152] The power generation sequence of the electrolytic water hydrogen production equipment is obtained, and the power fluctuation feature is extracted; the initial power transfer path is generated by using a deep reinforcement learning algorithm in combination with the first order change rate, the second order change rate and the power fluctuation feature, including:

[0153] The power fluctuation feature is input into the policy network and the value network;

[0154] The first hidden layer of the policy network uses an exponential linear unit function to nonlinearly combine the power fluctuation feature to obtain a power feature vector, the second hidden layer processes the power feature vector through a residual connection to obtain a power adjustment action, and the output layer generates an action probability of power transfer based on the power adjustment action;

[0155] The long short-term memory unit of the value network stores the power fluctuation feature to form a historical power sequence, and the state evaluation result is obtained by importance screening the historical power sequence through a gating mechanism;

[0156] According to the power fluctuation characteristics, a power change acceleration coefficient, a power fluctuation amplitude coefficient and a power response time coefficient are calculated, and a power fluctuation constraint rule is constructed based on the power change acceleration coefficient, the power fluctuation amplitude coefficient and the power response time coefficient; the power fluctuation constraint rule and the action probability of the power transfer are combined to form a multi-objective optimization function;

[0157] A reward value is calculated using the multi-objective optimization function and the state evaluation result, a time series difference error is constructed according to the reward value, a trust domain constraint is established based on the time series difference error, and the parameters of the policy network are optimized under the trust domain constraint, and an initial power transfer path is generated through the optimized policy network parameters.

[0158] Exemplarily, first, the power generation sequence of the water electrolysis hydrogen production equipment is obtained. In actual application, power data can be collected every 5 seconds by the power sensor on the equipment, and 24 hours of continuous collection can obtain a power sequence P(t) composed of 17280 sampling points. For example, the power data (unit: kW) of a certain water electrolysis hydrogen production equipment during 8:00-8:05 am may be: [120.5, 121.2, 119.8, 122.3, 123.1, 121.7, 120.9, 122.5, 123.8, 124.2, 123.5, 122.8].

[0159] Next, the power fluctuation characteristics are extracted. The power fluctuation characteristics include power mean, standard deviation, maximum value, minimum value, peak factor and coefficient of variation. Taking the above power data as an example, the power mean is 122.19kW, the standard deviation is 1.37kW, the maximum value is 124.2kW, the minimum value is 119.8kW, the peak factor (ratio of maximum value to mean) is 1.016, and the coefficient of variation (ratio of standard deviation to mean) is 0.011.

[0160] Then, the first-order change rate and the second-order change rate of the power sequence are calculated. The first-order change rate represents the change speed of the power at adjacent time points, and the second-order change rate represents the change of the power change speed. For the above power data, the first-order change rate sequence (unit: kW / s) is: [0.14, -0.28, 0.5, 0.16, -0.28, -0.16, 0.32, 0.26, 0.08, -0.14, -0.14], and the second-order change rate sequence (unit: kW / s 2 ) is: [-0.42, 0.78, -0.34, -0.44, 0.12, 0.48, -0.06, -0.18, -0.22, 0].

[0161] The specific steps of generating an initial power transfer path using a deep reinforcement learning algorithm in combination with the first-order change rate, the second-order change rate and the power fluctuation characteristics are as follows:

[0162] The power fluctuation features are input into the policy network and the value network. The policy network and the value network are the core components of deep reinforcement learning, the policy network is responsible for generating actions (i.e. power adjustment decisions), and the value network is responsible for evaluating state values. The input power fluctuation feature vector contains the calculated feature values described above, with a dimension of 6.

[0163] The structure of the policy network is designed as follows: the first hidden layer contains 64 neurons, which use an exponential linear unit function to perform nonlinear combination on the power fluctuation features. The exponential linear unit function maintains a linear relationship when the input is positive and exponentially decays when the input is negative, which helps to solve the gradient vanishing problem. Through this layer of processing, the power feature vector is obtained, with a dimension of 64. The second hidden layer processes the power feature vector through residual connection, containing 32 neurons. Residual connection refers to directly adding the input to the output, which helps to solve the problem of difficulty in training deep networks. The output of this layer is the power adjustment action, with a dimension of 16. The output layer generates the action probability of power transfer based on the power adjustment action, with a dimension of 5, corresponding to 5 possible power adjustment strategies: large increase (10%), small increase (5%), keep unchanged, small decrease (5%), and large decrease (10%).

[0164] The value network uses long short-term memory units to store power fluctuation features to form a historical power sequence. The long short-term memory unit contains 32 memory units, which can store power data for the past 10 time steps. Through the gating mechanism (including the input gate, the forgetting gate, and the output gate), the historical power sequence is screened for importance, and the state evaluation result is obtained. The input gate controls the degree of new information entering the memory unit, the forgetting gate controls the degree of old information retention, and the output gate controls the degree of memory unit information output. After processing, the value network outputs a scalar value representing the value evaluation of the current state.

[0165] According to the power fluctuation features, the power change acceleration coefficient, the power fluctuation amplitude coefficient, and the power response time coefficient are calculated. The power change acceleration coefficient is proportional to the root mean square of the second-order change rate, and for the above data, it is calculated to be 0.38; the power fluctuation amplitude coefficient is proportional to the power standard deviation, and it is calculated to be 0.42; the power response time coefficient is inversely proportional to the mean of the first-order change rate, and it is calculated to be 0.29. Based on these three coefficients, the power fluctuation constraint rule is constructed to ensure that the power adjustment does not exceed the physical limits of the device.

[0166] The power fluctuation constraint rule and the action probability of power transfer are combined to form a multi-objective optimization function. The multi-objective optimization function considers the efficiency and safety of power adjustment, with an efficiency weight of 0.6 and a safety weight of 0.4.

[0167] The reward value is calculated by using the multi-objective optimization function and the state evaluation result. The reward value is composed of three parts: power tracking accuracy reward (weight 0.5), power stability reward (weight 0.3) and energy efficiency reward (weight 0.2). The timing difference error is constructed according to the reward value, and the timing difference error is the difference between the current state value estimation and the next state value estimation plus the instant reward.

[0168] The trust domain constraint is established based on the timing difference error, the step of policy updating is limited, and the training instability caused by the excessively large policy updating is prevented. The trust domain radius is set to 0.02, which means that the KL divergence of the new and old policies does not exceed 0.02. The parameters of the policy network are optimized under the trust domain constraint, the Adam optimizer is used, the learning rate is 0.001, and the training iteration number is 1000 times.

[0169] The initial power transfer path is generated by the optimized policy network parameters. For the above power data, the generated initial power transfer path is: [120.5, 121.2, 120.5, 121.5, 122.6, 122.0, 121.0, 122.0, 123.0, 123.5, 123.0, 122.5], compared with the original power sequence, the fluctuation amplitude is reduced by 18%, the power change acceleration is reduced by 25%, and the average level of power output is maintained.

[0170] Figure 3 The system energy efficiency comparison chart of the embodiment of the present application is as follows: Figure 3As shown in the figure, the performance of five different control methods in terms of electrolytic water hydrogen production equipment energy efficiency and process parameters is demonstrated. Under light load conditions (25% load), the energy efficiency of the present technical solution is 72.3%, which is 27.3% higher than the traditional PID control of 56.8%; under medium load conditions (50% load), the energy efficiency is 79.8%, which is increased by 16.8%; under heavy load conditions (85% load), the energy efficiency is 82.5%, which is increased by 14.4%. This shows that the present technical solution is particularly suitable for light load conditions and can effectively solve the problem of low energy efficiency of traditional electrolytic water hydrogen production equipment under low load. In terms of electrolytic cell temperature stability, the temperature fluctuation of the present technical solution is only ±1.4°C, which is 73.6% lower than the traditional control of ±5.3°C; the temperature rise time is shortened from 28.6 minutes to 16.2 minutes, which is increased by 43.4%; the maximum temperature rise is only 5.3°C, which is 58.6% lower than the traditional control of 12.8°C. This excellent temperature control capability is due to the precise control of power changes and the comprehensive consideration of temperature influencing factors by the deep reinforcement learning algorithm. In terms of hydrogen purity, the average purity of the present technical solution is 99.8%, and the minimum purity is not less than 99.3%, which is respectively increased by 1.1% and 2.1% compared with the traditional control. Although the purity improvement seems not big, considering that it is already in the high purity range, this improvement is of great significance to hydrogen energy application, especially fuel cell application. In summary, the present technical solution realizes fine control of electrolytic water hydrogen production under all working conditions through the long short-term memory unit of the value network and the nonlinear combination of the policy network, significantly improving the system energy efficiency and process indicators.

[0171] The prior art usually uses simple rules or fixed thresholds for power scheduling, but these methods perform poorly when faced with complex power fluctuations, often failing to respond to fluctuation changes in time, resulting in inaccurate regulation and affecting the stability and efficiency of the system. These traditional methods do not fully utilize historical data and power fluctuation characteristics to make more dynamic and accurate decisions.

[0172] The application significantly improves the generation capability of the power transfer path by adopting a deep reinforcement learning algorithm, combining a policy network and a value network to model the power fluctuation characteristics. The policy network non-linearly combines the power fluctuation characteristics through an exponential linear unit function, and optimizes the power adjustment action through a residual connection to make the adjustment more accurate. The value network uses a long short-term memory unit to process the historical power sequence, and combines a gating mechanism to filter important information, improving the accuracy and real-time performance of state evaluation. In addition, based on the acceleration coefficient, amplitude coefficient and response time coefficient of the power fluctuation, the application constructs a power fluctuation constraint rule, and combines it with a multi-objective optimization function to ensure that the generated power transfer path not only meets the performance requirements of the equipment, but also effectively suppresses the fluctuation. Through the optimization of the time difference error and the trust domain constraint, the stability of the policy network is enhanced, making the power adjustment process more stable and efficient. These improvements enable the application to intelligently adjust the power transfer path based on real-time power fluctuations and historical data, improving the scheduling capability of the electrolytic water hydrogen production system in complex dynamic environments, and ensuring that the system operates more smoothly and efficiently in fluctuations.

[0173] In an alternative embodiment,

[0174] According to the stress state index, the load shedding process is divided into a buffer load shedding section, a rapid load shedding section and a protection load shedding section; a dynamic mapping relationship between the stress state index and the load shedding rate of each load shedding section is established to generate a stepwise load shedding strategy, which includes:

[0175] According to the stress state index, a state feature matrix is established, and a characteristic root combination of the state feature matrix is calculated; the system stable state is judged according to the characteristic root combination, and the stable margin is calculated based on the system stable state; according to the stable margin, the load shedding process is divided into a buffer load shedding section, a rapid load shedding section and a protection load shedding section;

[0176] For each load shedding section, a load shedding rate and dynamic stress response function is established; the initial load shedding rate of the buffer load shedding section, the rapid load shedding section and the protection load shedding section is determined according to the response function;

[0177] A dynamic equation of the stress state index and the load shedding rate is established, in which a nonlinear term is introduced to represent the bifurcation characteristics of the system; the initial load shedding rate is substituted into the dynamic equation to solve the bifurcation parameter;

[0178] An adaptive adjustment equation of the stress deviation is established; a compensation amount is calculated according to the deviation between the stress state index and the preset stress threshold; the compensation amount is input into the adaptive adjustment equation to dynamically adjust the bifurcation parameter; based on the adjusted bifurcation parameter, the dynamic equation is solved to obtain the load shedding rate of each load shedding section;

[0179] The neural network model comprising multiple hidden layers is constructed, historical load shedding data is input into the neural network model for training, a mapping relationship between load shedding rate and stress fluctuation is established by using the trained neural network model, a load shedding rate correction value is calculated based on the mapping relationship, and the load shedding rate of each load shedding section is corrected in combination with the load shedding rate correction value;

[0180] A load shedding time allocation function is established, the switching time and duration of each load shedding section are calculated according to the time allocation function, and the stepped load shedding strategy is obtained by combining the switching time, duration and corrected load shedding rate.

[0181] Exemplarily, first, a state characteristic matrix is established according to a stress state index and load shedding sections are divided. The system collects real-time stress state indexes, including power grid frequency, voltage, power and other parameters. Taking a certain power system as an example, 10 key indexes such as frequency deviation, voltage deviation and power margin are collected to construct a 10x10 state characteristic matrix. By calculating the eigenroot combination of the matrix, λ1=0.82, λ2=0.76, λ3=0.65 and other eigenvalues are obtained. When the real parts of all eigenroots are negative and the absolute value of the real part of the largest eigenroot is greater than 0.5, it is determined that the system is in a stable state. The stability margin is calculated to be 0.68 based on the eigenroot distribution, and the load shedding process is divided into three sections according to the preset threshold value: the buffer load shedding section when the stability margin is greater than 0.7; the fast load shedding section when the stability margin is between 0.4 and 0.7; and the protection load shedding section when the stability margin is less than 0.4.

[0182] For each load shedding section, a response function of load shedding rate and dynamic stress is established. Through historical data analysis, it is determined that the initial load shedding rate of the buffer load shedding section is 5 MW / min, the fast load shedding section is 15 MW / min, and the protection load shedding section is 3 MW / min. Taking a certain 500 MW unit as an example, when the system frequency fluctuation is maintained within ±0.02 Hz per minute at a rate of 5 MW / min in the buffer load shedding section, the frequency fluctuation is controlled within ±0.05 Hz at a rate of 15 MW / min in the fast load shedding section, and the frequency fluctuation is controlled within ±0.01 Hz at a rate of 3 MW / min in the protection load shedding section.

[0183] Next, a dynamic equation of stress state index and load shedding rate is established. The equation includes linear and nonlinear terms, where the nonlinear term is used to represent the bifurcation characteristics of the system near the critical point. The initial load shedding rate is substituted into the dynamic equation, and the bifurcation parameters α=0.62, β=0.85 and γ=0.41 are obtained by solving. These parameters describe the critical conditions for the system to transition from one stable state to another.

[0184] To achieve dynamic adjustment of the load shedding rate, an adaptive adjustment equation of stress deviation is established. Real-time monitoring of the stress state index and the preset stress threshold deviation, when the frequency deviation exceeds 0.1 Hz, the compensation amount is 0.15; when the voltage deviation exceeds 5%, the compensation amount is 0.2. The compensation amount is input into the adaptive adjustment equation, and the bifurcation parameters are dynamically adjusted. The adjusted parameters are α' = 0.71, β' = 0.92, and γ' = 0.38. Based on the adjusted bifurcation parameters, the dynamic equation is solved, and the load shedding rate of the buffer load shedding segment is adjusted to 6.2 MW / min, the fast load shedding segment is adjusted to 17.5 MW / min, and the protection load shedding segment is adjusted to 2.8 MW / min.

[0185] To further optimize the load shedding rate, a neural network model containing 3 hidden layers is constructed, each layer contains 20, 15 and 10 neurons respectively. The input layer receives the stress state index, and the output layer outputs the load shedding rate correction value. The model is trained using historical load shedding data, including 100 groups of normal working condition data and 50 groups of fault working condition data. The accuracy of the model after training reaches 92.5%. Through the model, the mapping relationship between the load shedding rate and the stress fluctuation is established, and the load shedding rate correction value is calculated: the buffer load shedding segment is +0.8 MW / min, the fast load shedding segment is -1.2 MW / min, and the protection load shedding segment is +0.5 MW / min. Combined with the correction value, the load shedding rate of each load shedding segment is corrected, and the final load shedding rate of the buffer load shedding segment is 7.0 MW / min, the fast load shedding segment is 16.3 MW / min, and the protection load shedding segment is 3.3 MW / min.

[0186] Finally, the load shedding time allocation function is established, and the switching time and duration of each load shedding segment are calculated according to the initial load and target load of the system. Taking a system from 500 MW to 300 MW as an example, the calculation results are as follows: the buffer load shedding segment lasts for 15 minutes, starting from 0 minutes to 15 minutes; the fast load shedding segment lasts for 10 minutes, starting from 15 minutes to 25 minutes; the protection load shedding segment lasts for 12 minutes, starting from 25 minutes to 37 minutes. Combining the switching time, duration and corrected load shedding rate, the complete stepwise load shedding strategy is obtained: 0-15 minutes at a rate of 7.0 MW / min, a total of 105 MW is reduced; 15-25 minutes at a rate of 16.3 MW / min, a total of 163 MW is reduced; 25-37 minutes at a rate of 3.3 MW / min, a total of 39.6 MW is reduced, a total of 307.6 MW is reduced, close to the target load of 300 MW, the error is within the acceptable range.

[0187] The stepwise load shedding strategy can effectively reduce system oscillation, reduce frequency fluctuation amplitude by about 40%, shorten system recovery time by about 25%, and significantly improve the safety and stability of the power system.

[0188] Figure 4 The simulation diagram of dynamic adjustment of the load shedding rate in each section of the stepwise load shedding strategy of the embodiment of the present application is shown in FIG. 3. Figure 4 As shown in the diagram, the diagram shows in detail the dynamic adjustment process of the rate in the three load shedding stages in the technical solution. The load shedding rate in the buffer load shedding section (0-60 seconds) is averagely 0.42 MPa / s, and the rate presents a slow downward trend, which provides sufficient adaptation time for the system. The load shedding rate in the fast load shedding section (60-120 seconds) is greatly increased to 1.57 MPa / s, reaches a peak at 80 seconds, and then is slightly adjusted according to the system response. The load shedding rate in the protection load shedding section (120-180 seconds) is rapidly reduced to 0.25 MPa / s and continues to decrease, ensuring smooth transition of the system to the final state. As can be seen from the diagram, the load shedding rate is smoothly transitioned between sections, avoiding the sudden change problem commonly seen in traditional methods. In particular, the load shedding rate changes smoothly at the two turning points of 60 seconds and 120 seconds, which is due to the application of the nonlinear bifurcation characteristic analysis and the adaptive adjustment algorithm in the present solution, effectively preventing the system from producing oscillation due to sudden change of the rate.

[0189] Based on the above technical solution, a more precise load shedding process control can be achieved, the load shedding rate is dynamically adjusted through the stress state index, and the system stability and safety of the load shedding process are improved. The existing technology usually performs load shedding through a fixed rate or a simple preset mode, but often ignores the real-time changes and dynamic response of the system stress, resulting in inflexible load shedding process and difficulty in coping with complex operating environments. The present application establishes a dynamic mapping relationship between the stress state index and the load shedding rate, trains historical data using a multi-level neural network model, and thus realizes precise correction of the load shedding rate. By establishing a dynamic equation containing a nonlinear term and introducing an adaptive adjustment mechanism, the load shedding rate can be dynamically adjusted, the load shedding process is optimized, and in particular in the case of large stress fluctuations, the system instability can be effectively reduced. In addition, by segmenting the load shedding process and calculating the stability margin and bifurcation parameters, the present solution ensures that the load shedding process can be reasonably adjusted according to different stress states, avoiding potential risks caused by single load shedding strategy in the existing technology. The improvement point is to solve the problem of ignoring the actual stress and dynamic changes of the system in the traditional method, making the load shedding process more flexible and safe. The improved effect significantly improves the response capability and adaptability of the load shedding process, while avoiding the negative impact of too rough load shedding scheme on the system, thereby improving the safety and stability of the system.

[0190] The second aspect of the embodiment of the present application is,

[0191] A minute-level intelligent scheduling system of a renewable energy hydrogen-based energy source is provided, and the system comprises:

[0192] The first unit is configured to acquire real-time power generation data of a renewable energy power generation device, operation parameter data of a water electrolysis hydrogen production device, and hydrogen storage amount data of a hydrogen storage system.

[0193] The second unit is configured to calculate a power generation fluctuation frequency according to the real-time power generation data, adaptively adjust a length of a dynamic time window based on the power generation fluctuation frequency, segment the power generation data using the dynamic time window, and generate a fluctuation feature sequence.

[0194] The third unit is configured to construct a minute-level intelligent scheduling model based on the fluctuation feature sequence, the operation parameter data, and the hydrogen storage amount data, generate an initial hydrogen production power adjustment instruction, calculate a mutual feedback correction parameter between the hydrogen storage system and the water electrolysis hydrogen production device, compensate and optimize the initial hydrogen production power adjustment instruction using the mutual feedback correction parameter, and obtain an optimized hydrogen production power adjustment instruction.

[0195] The fourth unit is configured to send the optimized hydrogen production power adjustment instruction to a control system of the water electrolysis hydrogen production device, control the water electrolysis hydrogen production device to operate, and monitor an operation state of the hydrogen production system; when an abnormality is detected, calculate an optimal power transfer path according to the fluctuation feature sequence, transfer power generation power to a power grid according to the optimal power transfer path, and control the water electrolysis hydrogen production device to perform stepwise load reduction and shutdown.

[0196] A third aspect of the embodiments of the present application,

[0197] An electronic device is provided.

[0198] A processor;

[0199] A memory for storing processor-executable instructions;

[0200] The processor is configured to invoke the instructions stored in the memory to perform the method described above.

[0201] A fourth aspect of the embodiments of the present application,

[0202] A computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0203] The present application can be a method, device, system and / or computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions loaded thereon for performing various aspects of the present application.

[0204] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions recorded in the above embodiments can be modified, or some or all of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A minute-level intelligent scheduling method for renewable energy hydrogen-based energy sources, characterized by, The method comprises the following steps: acquiring real-time power generation data of a renewable energy power generation device, operation parameter data of a water electrolysis hydrogen production device, and hydrogen storage amount data of a hydrogen storage system; calculating a power generation fluctuation frequency according to the real-time power generation data, adaptively adjusting the length of a dynamic time window based on the power generation fluctuation frequency, and using the dynamic time window to segment the power generation data to generate a fluctuation feature sequence; constructing a minute-level intelligent scheduling model based on the fluctuation feature sequence, the operation parameter data, and the hydrogen storage amount data to generate an initial hydrogen production power adjustment instruction, establishing a mapping model of pressure and electrolysis efficiency based on the pressure value in the hydrogen storage amount data, calculating the deviation of the current pressure relative to the rated pressure to obtain a pressure influence factor, establishing a mapping model of temperature and hydrogen production efficiency based on the temperature value in the operation parameter data, calculating the deviation of the current temperature relative to the rated temperature to obtain a temperature coupling factor, multiplying the pressure influence factor and the temperature coupling factor, and combining with the equipment performance attenuation value to obtain a mutual feedback correction parameter; compensating and adjusting the initial hydrogen production power adjustment instruction according to the mutual feedback correction parameter to obtain a corrected power instruction; at the same time, compensating the maximum power limit of the electrolytic cell to obtain a corrected power upper limit, and compensating the power ramp rate limit of the electrolytic cell to obtain a corrected ramp rate limit; determining a smoothing factor based on the time variation rate of the mutual feedback correction parameter, weighting and averaging the corrected power instruction and the power adjustment instruction of the last control cycle according to the smoothing factor to generate a transition instruction; judging whether the transition instruction meets the corrected power upper limit and the corrected ramp rate limit, if yes, outputting the transition instruction as the optimized hydrogen production power adjustment instruction, if not, limiting the transition instruction within the allowable range and outputting it as the optimized hydrogen production power adjustment instruction; sending the optimized hydrogen production power adjustment instruction to the control system of the water electrolysis hydrogen production device to control the operation of the water electrolysis hydrogen production device, and monitoring the operation state of the hydrogen production system; when an abnormality is detected, calculating an optimal power transfer path according to the fluctuation feature sequence, transferring the power generation power to the power grid according to the optimal power transfer path, and controlling the water electrolysis hydrogen production device to perform stepwise load shedding and shutdown.

2. The method of claim 1, wherein, The method comprises the following steps: performing wavelet transform on the real-time power generation data to obtain a frequency component energy spectrum, calculating the energy value of each frequency interval based on the frequency component energy spectrum, and determining the frequency interval with an energy value greater than a preset energy threshold as a dominant fluctuation frequency interval; calculating a frequency feature coefficient according to the energy distribution in the dominant fluctuation frequency interval, inputting the frequency feature coefficient into a pre-constructed dynamic time window adaptive model, optimizing the model parameters of the dynamic time window adaptive model through a multi-objective optimization algorithm, and calculating the length of the dynamic time window based on the optimized model parameters; performing sliding window segmentation on the real-time power generation data based on the length of the dynamic time window to obtain a time window power sequence, and calculating the mean value of the time window power sequence to obtain a power mean value sequence; A standard deviation of the power mean sequence is calculated to obtain a power fluctuation amplitude feature, a zero-crossing rate of the power mean sequence is calculated to obtain a power fluctuation frequency feature, and a change rate of the power mean sequence is calculated to obtain a power fluctuation change feature; The power fluctuation amplitude feature, the power fluctuation frequency feature, and the power fluctuation change feature are combined to construct a fluctuation feature sequence.

3. The method of claim 2, wherein, The real-time power generation data is segmented by a sliding window based on the dynamic time window length, including: The real-time power generation data is segmented by a sliding window based on the dynamic time window length, the power change rate of adjacent sampling points is calculated, a data fluctuation evaluation function is constructed based on the power change rate, the data integrity of the time window is evaluated by the data fluctuation evaluation function, and the time window with a data integrity lower than a preset integrity threshold is determined as a to-be-supplemented window; Features of the power generation data of the to-be-supplemented window are extracted to obtain a feature vector including a power mean and a fluctuation period, a multi-layer similarity matching rule is constructed based on the feature vector, and a data segment with a similarity greater than a similarity threshold between the to-be-supplemented window and the power generation features is selected from historical power generation data to form a candidate supplement sequence; An initial weight coefficient of a dynamic time warping algorithm is set according to the power change rate in the to-be-supplemented window, the similarity between each data segment in the candidate supplement sequence and the to-be-supplemented window is calculated based on the initial weight coefficient, and the data segment with the highest similarity is selected as the supplement data; A weight decay function is constructed based on the power change rate, and the original power generation data in the to-be-supplemented window and the supplement data are weighted and smoothed according to the weight decay function to generate supplemented time window data; A sequence continuity index and a dynamic consistency index of the supplemented time window data are calculated, a weighted result of the continuity index and the consistency index is compared with a preset evaluation threshold, when the weighted result is less than the preset evaluation threshold, the parameters of the weight decay function are adjusted and the weighted smoothing processing step is executed again until the weighted result is greater than the preset evaluation threshold, and a time window power sequence meeting the requirements is output.

4. The method of claim 1, wherein, When an abnormality is detected, an optimal power transfer path is calculated according to the fluctuation feature sequence, the power generation power is transferred to the power grid according to the optimal power transfer path, and the water electrolysis hydrogen production equipment is controlled to perform a stepwise load reduction shutdown, including: The membrane voltage fluctuation rate, hydrogen production efficiency decay rate, and system pressure mutation rate of the water electrolysis hydrogen production equipment are collected, and a first-order change rate and a second-order change rate are calculated; when the first-order change rate or the second-order change rate exceeds the corresponding early warning threshold, a power transfer mechanism is triggered; The power generation power sequence of the water electrolysis hydrogen production equipment is obtained, and power fluctuation features are extracted; the first-order change rate, the second-order change rate, and the power fluctuation features are combined, and a deep reinforcement learning algorithm is used to generate an initial power transfer path; Stress monitoring data of the water electrolysis hydrogen production equipment is collected, and a stress state index is calculated based on the stress monitoring data; a load shedding process is divided into a buffer load shedding section, a rapid load shedding section and a protection load shedding section according to the stress state index; a dynamic mapping relationship between the stress state index and load shedding rates of the load shedding sections is established, and a stepwise load shedding strategy is generated; The pressure signal of the hydrogen storage system is wavelet decomposed to extract a disturbance component and a trend component; a pressure pulsation feature is identified based on the disturbance component, and a pressure change trend is predicted based on the trend component; the pressure pulsation feature and the pressure change trend are combined with the stress state index to obtain a system evolution feature; The system evolution feature is used to correct an initial power transfer path in real time to obtain an optimal power transfer path; power generation power is controlled to be transferred to a power grid according to the optimal power transfer path, and the water electrolysis hydrogen production equipment is controlled to perform load shedding based on the stepwise load shedding strategy; The stress monitoring data and the pressure signal in the load shedding process are fed back to update the system evolution feature; the optimal power transfer path and the load shedding rate in the stepwise load shedding strategy are optimized in real time based on the updated system evolution feature; the optimization process is repeatedly performed until the water electrolysis hydrogen production equipment is safely shut down.

5. The method of claim 4, wherein, A power generation power sequence of the water electrolysis hydrogen production equipment is obtained, and a power fluctuation feature is extracted; the first-order change rate, the second-order change rate and the power fluctuation feature are combined, and a deep reinforcement learning algorithm is used to generate an initial power transfer path, including: The power fluctuation feature is input into a policy network and a value network; An exponential linear unit function is used in a first hidden layer of the policy network to perform nonlinear combination on the power fluctuation feature to obtain a power feature vector, a second hidden layer processes the power feature vector through a residual connection to obtain a power adjustment action, and an output layer generates an action probability of power transfer based on the power adjustment action; A long short-term memory unit of the value network stores the power fluctuation feature to form a historical power sequence, and a gate mechanism is used to screen the historical power sequence to obtain a state evaluation result; A power change acceleration coefficient, a power fluctuation amplitude coefficient and a power response time coefficient are calculated based on the power fluctuation feature, a power fluctuation constraint rule is constructed based on the power change acceleration coefficient, the power fluctuation amplitude coefficient and the power response time coefficient, and the power fluctuation constraint rule and the action probability of power transfer are combined to form a multi-objective optimization function; A reward value is calculated using the multi-objective optimization function and the state evaluation result, a time series difference error is constructed according to the reward value, a trust domain constraint is established based on the time series difference error, parameters of the policy network are optimized under the trust domain constraint, and an initial power transfer path is generated through the optimized policy network parameters.

6. The method of claim 4, wherein, The load shedding process is divided into a buffer load shedding section, a rapid load shedding section and a protection load shedding section according to the stress state index; The dynamic mapping relationship between the stress state index and load shedding rates of the load shedding sections is established, and a stepwise load shedding strategy is generated, including: According to the stress state index, a state characteristic matrix is established, a characteristic root combination of the state characteristic matrix is calculated, the system stable state is judged according to the characteristic root combination, and a stable margin is calculated based on the system stable state; according to the stable margin, a load shedding process is divided into a buffer load shedding section, a rapid load shedding section and a protection load shedding section; For each load shedding section, a response function of the load shedding rate and the dynamic stress is established; according to the response function, the initial load shedding rate of the buffer load shedding section, the rapid load shedding section and the protection load shedding section is determined; A kinetic equation of the stress state index and the load shedding rate is established, wherein a nonlinear term is introduced to represent the system bifurcation characteristics; the initial load shedding rate is substituted into the kinetic equation to solve the bifurcation parameter; An adaptive adjustment equation of the stress deviation is established; a compensation amount is calculated according to the deviation of the stress state index and a preset stress threshold; the compensation amount is input into the adaptive adjustment equation to dynamically adjust the bifurcation parameter; based on the adjusted bifurcation parameter, the kinetic equation is solved to obtain the load shedding rate of each load shedding section; A neural network model containing multiple hidden layers is constructed, historical load shedding data is input into the neural network model for training, a mapping relationship between the load shedding rate and the stress fluctuation is established by using the trained neural network model, a load shedding rate correction value is calculated based on the mapping relationship, and the load shedding rate of each load shedding section is corrected in combination with the load shedding rate correction value; A load shedding time allocation function is established, the switching time and the duration of each load shedding section are calculated according to the time allocation function, and the switching time, the duration and the corrected load shedding rate are combined to obtain a stepwise load shedding strategy.

7. A minute-level intelligent scheduling system for renewable energy hydrogen-based energy sources for implementing the method of any of the preceding claims 1-6, characterized by, It comprises: A first unit is configured to acquire real-time power generation data of a renewable energy power generation device, operation parameter data of a water electrolysis hydrogen production device, and hydrogen storage amount data of a hydrogen storage system; A second unit is configured to calculate a power generation fluctuation frequency based on the real-time power generation data, adaptively adjust the length of a dynamic time window based on the power generation fluctuation frequency, and segment the power generation data using the dynamic time window to generate a fluctuation feature sequence; A third unit is configured to construct a minute-level intelligent scheduling model based on the fluctuation feature sequence, the operation parameter data, and the hydrogen storage amount data, generate an initial hydrogen production power adjustment instruction, calculate a mutual feedback correction parameter between the hydrogen storage system and the water electrolysis hydrogen production device, compensate and optimize the initial hydrogen production power adjustment instruction using the mutual feedback correction parameter, and obtain an optimized hydrogen production power adjustment instruction; A fourth unit is configured to send the optimized hydrogen production power adjustment instruction to a control system of the water electrolysis hydrogen production device, control the water electrolysis hydrogen production device to operate, and monitor the operation state of the hydrogen production system. When an abnormality is detected, an optimal power transfer path is calculated based on the fluctuation feature sequence, the power generation power is transferred to the power grid according to the optimal power transfer path, and the water electrolysis hydrogen production device is controlled to perform a stepwise load shedding shutdown.

8. An electronic device, comprising: It comprises: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to invoke the instructions stored in the memory to execute the method of any one of claims 1 to 6.

9. A computer-readable storage medium having stored thereon computer program instructions, wherein, The computer program instructions, when executed by the processor, implement the method of any one of claims 1 to 6. The computer program instructions, when executed by the processor, implement the method of any one of claims 1 to 6.

Images