Green hydrogen energy balance management method and system based on source-load interaction

Abstract

The invention provides a green hydrogen energy balance management method and system based on source-load interaction, and relates to the technical field of hydrogen energy management, and the method comprises the steps: building an index set containing equipment capacity matching, operation efficiency, energy loss and supply and demand balance degree, building a three-layer nested optimization structure of a day-ahead scheduling layer, an intra-day rolling layer and a real-time response layer, and carrying out the optimization of the three-layer nested optimization structure. Cooperative scheduling is carried out by adopting a layered progressive supply and demand matching method, so that efficient scheduling and source-load balance of the green hydrogen energy system are realized, the utilization rate of renewable energy sources is improved, the operation cost of the system is reduced, and the reliability of energy supply is enhanced.

Description

Technical Field

[0001] This invention relates to the field of hydrogen energy management technology, and in particular to a green hydrogen energy balance management method and system based on source-load interaction. Background Technology

[0002] With the global energy structure transformation and the introduction of carbon emission reduction targets, green hydrogen energy has received widespread attention as a clean energy carrier. Green hydrogen energy systems primarily rely on renewable energy power generation to produce hydrogen, forming a complete energy supply chain through hydrogen production, storage, transportation, and utilization. In green hydrogen energy systems, the volatility of renewable energy power generation, the operating characteristics of hydrogen production units, and the diversity of hydrogen demand present complex balance management challenges. Currently, the scheduling and management of green hydrogen energy systems mainly adopts a single time-scale optimization method, utilizing water electrolysis technology to convert renewable energy electricity into hydrogen energy, and achieving the spatiotemporal transfer of energy through hydrogen storage devices.

[0003] Existing technologies do not adequately consider the interaction characteristics between sources and loads. Most methods treat renewable energy power generation and hydrogen production load as relatively independent links, lacking in-depth analysis of the coupling relationships between the various components of the system, which makes it difficult to achieve efficient operation of the system under fluctuating renewable energy conditions.

[0004] Existing technologies generally employ optimization methods based on a single time scale, which are insufficient to address the uncertainties and fluctuations across different time scales in green hydrogen energy systems. The lack of an effective mechanism linking current planning with real-time operation prevents flexible adjustments based on the system's real-time status, reducing the system's adaptability and robustness. Summary of the Invention

[0005] The embodiments of the present invention provide a green hydrogen energy balance management method and system based on source-load interaction, which can solve the problems in the prior art.

[0006] A first aspect of this invention provides a green hydrogen energy balance management method based on source-load interaction, comprising:

[0007] Key features of green hydrogen energy system data are extracted to construct an indicator set, which includes equipment capacity matching indicators, equipment operating efficiency indicators, energy transmission loss indicators, and supply and demand balance indicators.

[0008] A three-layer nested optimization structure is constructed, comprising a day-ahead scheduling layer, an intraday rolling layer, and a real-time response layer: the day-ahead scheduling layer uses the equipment capacity matching index to formulate a baseline scheduling plan for system operation; the intraday rolling layer dynamically corrects the baseline scheduling plan based on predicted data; and the real-time response layer adjusts the plan according to system fluctuations. A time-scale decoupling algorithm is used to construct a multi-level linkage mechanism with time-dimensional correlation constraints, enabling coordination and complementarity between different layers.

[0009] Based on the output of the three-layer nested optimization structure, a hierarchical progressive supply and demand matching method is adopted to coordinate the scheduling of the green hydrogen energy system. At the source-end power generation prediction layer, the power generation of renewable energy power generation devices is predicted and modeled based on the equipment operating efficiency index. At the hydrogen production power allocation layer, the hydrogen production power allocation scheme of the hydrogen production device is obtained based on the power generation prediction results and the energy transmission loss index. At the hydrogen storage capacity adjustment layer, the hydrogen charging and discharging power of the hydrogen storage device is adjusted in real time based on the hydrogen production power allocation scheme and the supply and demand balance index. Progressive information transmission and optimization are carried out between each level, and a cross-level collaborative response mechanism is triggered when the system fluctuates.

[0010] Dynamically revising the baseline scheduling plan based on forecast data at the intraday rolling layer includes:

[0011] In the intraday rolling layer, time-series features of historical system operation data are extracted to construct a time-series feature matrix;

[0012] The time-series feature matrix is ​​subjected to nonlinear feature mapping, and combined with a long short-term memory neural network to perform rolling forecasts of renewable energy power generation and hydrogen load demand. The root mean square values ​​of the power generation forecast deviation rate and the hydrogen load forecast deviation rate are calculated to obtain the power forecast error.

[0013] Based on the temporal feature matrix, scene generation is performed and the temporal distance between scenes is calculated. A preset number of initial class centers are randomly selected, and each scene is assigned to the class center with the closest temporal distance. The scene with the highest average similarity to other scenes in each class is iteratively updated as the new class center. The sum of scene similarities within the class is calculated. When the change in the sum of scene similarities between two adjacent iterations is less than a preset convergence threshold, the iteration stops. A probabilistic scene tree is constructed with the current time as the root node and the class centers as leaf nodes. The conditional probability and conditional expectation of each node in the probabilistic scene tree are calculated and weighted to obtain the scene expectation value.

[0014] The power prediction error is weighted and averaged with the expected value of the scenario to obtain the power correction amount, which is then superimposed on the corresponding time period of the day-ahead baseline scheduling plan.

[0015] Adjustments at the real-time response layer based on system fluctuations include:

[0016] In the real-time response layer, the real-time power sampling value is obtained based on the system operation data, and the deviation between the real-time power sampling value and the reference value is calculated to obtain the real-time power deviation. When the real-time power deviation of two adjacent sampling points exceeds the preset fluctuation threshold, it is determined that the system has fluctuated.

[0017] When the system fluctuates, the rate of change and integral value of the real-time power deviation are normalized according to a preset quantization factor to obtain the input variable, and the input variable is fuzzified using a triangular membership function to obtain the membership degree value.

[0018] Based on the rate of change of the real-time power deviation, the system state is divided into multiple fuzzy subsets, and an IF-THEN fuzzy rule base is established. The membership values ​​are substituted into the fuzzy rule base to obtain the output domain interval corresponding to each fuzzy subset. Discrete points are obtained by sampling the output domain interval corresponding to each fuzzy subset at equal intervals. The membership value corresponding to each discrete point is calculated. The centroid position of the fuzzy subset is obtained by summing the products of the abscissas of all discrete points and their corresponding membership values. The centroid position and its corresponding membership values ​​are weighted and summed to obtain the comprehensive output value. The comprehensive output value is multiplied by a preset proportional coefficient and inversely normalized to obtain the power adjustment amount.

[0019] The distribution ratio of hydrogen production power and hydrogen storage power is adjusted in real time according to the power adjustment amount.

[0020] A multi-level linkage mechanism with temporal-dimensional correlation constraints is constructed using a time-scale decoupling algorithm, enabling coordination and complementarity between different levels, including:

[0021] Time series data is decomposed into multiple intrinsic mode components (IMCs) through empirical mode decomposition. The IMCs are then subjected to Hilbert transform to obtain the corresponding analytic signals. The instantaneous frequency is obtained by calculating the rate of change of the instantaneous phase of the analytic signals over time. The IMCs are then recombined based on the magnitude of the instantaneous frequency to obtain the wave hierarchy.

[0022] Calculate the covariance of the time series deviations between each pair of fluctuation levels and their mean deviations, and then divide it by the product of the standard deviations of the time series of the corresponding fluctuation levels to obtain the inter-level coupling degree; construct a state observer based on the frequency characteristics and dynamic response characteristics of each fluctuation level, and construct a feedback gain matrix using the observation results of the state observer; calculate the state deviation between the state of each fluctuation level and the desired state, and weight and superimpose the product of the state deviation and the feedback gain matrix with the inter-level coupling degree as the reference command quantity;

[0023] The performance index gradient is calculated in real time based on the system's operating status. The adaptive gain is updated based on the gradient descent algorithm. The rate of change of the coordination coefficient between each level is obtained by multiplying the negative value of the product of the hierarchical coupling degree and the state deviation by the adaptive gain. The rate of change of the coordination coefficient is integrated to obtain the inter-level coordination coefficient. The product of the inter-level coordination coefficient and the reference command quantity is used as the inter-level coordination control quantity.

[0024] In the hydrogen production power allocation layer, based on the power generation prediction results and the energy transmission loss index, the hydrogen production power allocation scheme for the hydrogen production unit includes:

[0025] The input power and conversion efficiency of each stage of DC-DC conversion are collected to obtain the conversion loss, and the equivalent resistance and transmission current of the transmission line are collected to obtain the line loss. The sum of the conversion loss and the line loss is used as the energy transmission loss index.

[0026] The power margin is obtained based on the difference between the predicted power generation and the load power. The power margin is multiplied by the preset power reserve coefficient to obtain the safe power margin. The power margin is subtracted from the sum of the safe power margin and the energy transmission loss index to obtain the allocable power.

[0027] Based on the allocable power, the upper limit constraint of hydrogen production power is set, the lower limit of the rated power of the hydrogen production device is set as the lower limit constraint of hydrogen production power, and the hydrogen production power change rate constraint is set according to the dynamic response characteristics of the hydrogen production device to construct a set of constraint conditions.

[0028] The entire prediction time domain is divided into several consecutive control periods. At the beginning of the current control period, the system operating status is obtained and the constraint set is updated. Based on the updated constraint set, the hydrogen production power sequence within the time period is solved. The first control variable of the hydrogen production power sequence is selected as the hydrogen production power control output at the current moment. The time domain is rolled forward by one control cycle and the solution steps are repeated until the entire prediction time domain is traversed. Based on the final hydrogen production power control output, a hydrogen production power allocation scheme that satisfies multiple constraints is obtained.

[0029] In the hydrogen storage capacity regulation layer, based on the hydrogen production power allocation scheme and the supply-demand balance index, the real-time adjustment of the hydrogen charging and discharging power of the hydrogen storage device includes:

[0030] The difference between the target hydrogen production power and the load power in the hydrogen production power allocation scheme is divided by the system rated power to obtain the supply and demand balance index; the difference between the current hydrogen storage capacity and the system minimum hydrogen storage capacity is divided by the difference between the system maximum hydrogen storage capacity and the system minimum hydrogen storage capacity to obtain the hydrogen storage capacity status index.

[0031] Based on the dynamic response characteristics of the hydrogen storage capacity state index, the state gain coefficient and the differential gain coefficient are determined, and the state gain coefficient and the differential gain coefficient are optimized using the steepest gradient method; the supply and demand balance index is multiplied by the preset proportional coefficient and integral coefficient and superimposed to obtain the power compensation amount; the deviation between the hydrogen storage capacity state index and the hydrogen storage capacity target value is multiplied by the state gain coefficient, and the product of this product and the rate of change of the hydrogen storage capacity state index multiplied by the differential gain coefficient is added to obtain the state feedback amount; the weighted sum of the power compensation amount and the state feedback amount is used as the hydrogen charging and discharging regulation amount;

[0032] Based on the physical characteristics of the hydrogen storage system, a constraint set is constructed, which includes hydrogen storage capacity limit constraints, hydrogen charging and discharging power limit constraints, and hydrogen charging and discharging power change rate constraints. The hydrogen charging and discharging adjustment quantities that do not meet the constraint set are mapped to the constraint boundaries to obtain the hydrogen storage capacity adjustment command.

[0033] A second aspect of the present invention provides a green hydrogen energy balance management system based on source-load interaction, comprising:

[0034] The first unit is used to extract key features of green hydrogen energy system data to construct an indicator set, which includes equipment capacity matching indicators, equipment operating efficiency indicators, energy transmission loss indicators, and supply and demand balance indicators.

[0035] The second unit is used to construct a three-layer nested optimization structure, which includes a day-ahead scheduling layer, an intraday rolling layer, and a real-time response layer. In the day-ahead scheduling layer, the equipment capacity matching index is used to formulate a baseline scheduling plan for system operation. In the intraday rolling layer, the baseline scheduling plan is dynamically corrected based on the predicted data. In the real-time response layer, adjustments are made according to system fluctuations. A time-scale decoupling algorithm is used to construct a multi-level linkage mechanism with time-dimensional correlation constraints, which coordinates and complements each other at different levels.

[0036] The third unit is used to coordinate the scheduling of the green hydrogen energy system using a hierarchical progressive supply and demand matching method based on the output results of the three-layer nested optimization structure. At the source-end power generation prediction layer, the power generation of renewable energy power generation devices is predicted and modeled based on the equipment operating efficiency index. At the hydrogen production power allocation layer, a hydrogen production power allocation scheme for the hydrogen production device is obtained based on the power generation prediction results and the energy transmission loss index. At the hydrogen storage capacity adjustment layer, the hydrogen charging and discharging power of the hydrogen storage device is adjusted in real time based on the hydrogen production power allocation scheme and the supply and demand balance index. Progressive information transmission and optimization are performed between each level, and a cross-level collaborative response mechanism is triggered when system fluctuations occur.

[0037] A third aspect of the present invention,

[0038] An electronic device is provided, comprising:

[0039] processor;

[0040] Memory used to store processor-executable instructions;

[0041] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0042] Fourth aspect of the embodiments of the present invention,

[0043] A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0044] The beneficial effects of this application are as follows:

[0045] The green hydrogen energy balance management method based on source-load interaction provided by this invention comprehensively evaluates the operating status of the green hydrogen energy system by constructing a comprehensive index set that includes equipment capacity matching index, equipment operating efficiency index, energy transmission loss index, and supply-demand balance index. This enables an accurate grasp of the key characteristics of the system and provides a scientific basis for optimization decisions.

[0046] The three-layer nested optimization structure designed in this invention, combined with the time-scale decoupling algorithm, realizes full-time-domain collaborative scheduling from day-ahead planning to real-time response. It effectively solves the challenges posed by the large volatility and strong randomness of renewable energy to the stable operation of green hydrogen energy systems, improves the system's ability to cope with uncertainties, and enhances the flexibility and robustness of operation.

[0047] The hierarchical and progressive supply and demand matching method adopted in this invention forms a closed-loop energy balance management mechanism by organically combining three levels: source power generation prediction, hydrogen production power allocation, and hydrogen storage capacity adjustment. This optimizes the energy conversion efficiency of each link in the system, reduces energy transmission losses, and significantly improves the overall economy and reliability of the green hydrogen energy system, providing technical support for the efficient operation of large-scale renewable energy and hydrogen energy coupling systems. Attached Figure Description

[0048] Figure 1 This is a flowchart illustrating the green hydrogen energy balance management method based on source-load interaction, as described in an embodiment of the present invention.

[0049] Figure 2 A thermodynamic diagram of the coupling degree matrix between IMF components;

[0050] Figure 3 This is a schematic diagram illustrating the performance comparison and analysis of hydrogen storage systems. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0053] Figure 1 This is a flowchart illustrating the green hydrogen energy balance management method based on source-load interaction, as described in an embodiment of the present invention. Figure 1 As shown, the method includes:

[0054] Key features of green hydrogen energy system data are extracted to construct an indicator set, which includes equipment capacity matching indicators, equipment operating efficiency indicators, energy transmission loss indicators, and supply and demand balance indicators.

[0055] A three-layer nested optimization structure is constructed, comprising a day-ahead scheduling layer, an intraday rolling layer, and a real-time response layer: the day-ahead scheduling layer uses the equipment capacity matching index to formulate a baseline scheduling plan for system operation; the intraday rolling layer dynamically corrects the baseline scheduling plan based on predicted data; and the real-time response layer adjusts the plan according to system fluctuations. A time-scale decoupling algorithm is used to construct a multi-level linkage mechanism with time-dimensional correlation constraints, enabling coordination and complementarity between different layers.

[0056] Based on the output of the three-layer nested optimization structure, a hierarchical progressive supply and demand matching method is adopted to coordinate the scheduling of the green hydrogen energy system. At the source-end power generation prediction layer, the power generation of renewable energy power generation devices is predicted and modeled based on the equipment operating efficiency index. At the hydrogen production power allocation layer, the hydrogen production power allocation scheme of the hydrogen production device is obtained based on the power generation prediction results and the energy transmission loss index. At the hydrogen storage capacity adjustment layer, the hydrogen charging and discharging power of the hydrogen storage device is adjusted in real time based on the hydrogen production power allocation scheme and the supply and demand balance index. Progressive information transmission and optimization are carried out between each level, and a cross-level collaborative response mechanism is triggered when the system fluctuates.

[0057] In one alternative implementation, dynamically adjusting the baseline scheduling plan based on forecast data at the intraday rolling layer includes:

[0058] In the intraday rolling layer, time-series features of historical system operation data are extracted to construct a time-series feature matrix;

[0059] The time-series feature matrix is ​​subjected to nonlinear feature mapping, and combined with a long short-term memory neural network to perform rolling forecasts of renewable energy power generation and hydrogen load demand. The root mean square values ​​of the power generation forecast deviation rate and the hydrogen load forecast deviation rate are calculated to obtain the power forecast error.

[0060] Based on the temporal feature matrix, scene generation is performed and the temporal distance between scenes is calculated. A preset number of initial class centers are randomly selected, and each scene is assigned to the class center with the closest temporal distance. The scene with the highest average similarity to other scenes in each class is iteratively updated as the new class center. The sum of scene similarities within the class is calculated. When the change in the sum of scene similarities between two adjacent iterations is less than a preset convergence threshold, the iteration stops. A probabilistic scene tree is constructed with the current time as the root node and the class centers as leaf nodes. The conditional probability and conditional expectation of each node in the probabilistic scene tree are calculated and weighted to obtain the scene expectation value.

[0061] The power prediction error is weighted and averaged with the expected value of the scenario to obtain the power correction amount, which is then superimposed on the corresponding time period of the day-ahead baseline scheduling plan.

[0062] In the intraday rolling layer, the time-series feature matrix contains several feature dimensions, such as the values ​​of wind power, photovoltaic power, and hydrogen load at different times. For example, for a wind farm, power generation data sampled every 15 minutes over the past 30 days can be extracted to form a time-series feature vector containing 2880 time points; similarly, multi-dimensional features such as photovoltaic power generation and hydrogen load can be extracted to form a time-series feature matrix.

[0063] When performing nonlinear feature mapping on the temporal feature matrix, a deep autoencoder is used for dimensionality reduction. The autoencoder consists of a 5-layer structure: the input layer has the same number of nodes as the original feature dimension, such as 96 dimensions (representing one data point every 15 minutes in 24 hours); the encoding layers have 64 nodes and 32 nodes respectively; the decoding layers have 64 nodes respectively; and the output layer has the same dimension as the input layer. The autoencoder is trained using the backpropagation algorithm to minimize the reconstruction error, thereby obtaining the nonlinearly dimensionality-reduced feature representation.

[0064] The Long Short-Term Memory (LSTM) neural network consists of one input layer, two LSTM hidden layers (128 units each), and a fully connected output layer. The model takes 24 hours of time-series data as input and outputs predicted values ​​for the next 4 hours. For example, inputting wind power data from 10:00 AM on June 1st to 10:00 AM on June 2nd (96 data points in total, one data point every 15 minutes) would output predicted values ​​from 10:15 AM to 2:15 PM on June 2nd (16 data points in total). The training set uses three months of historical data, and the validation set uses data from the last two weeks.

[0065] Taking a certain forecast as an example, the actual wind power output was 100MW, and the predicted value was 95MW, resulting in a deviation rate of 5%; the actual photovoltaic power output was 80MW, and the predicted value was 76MW, also with a deviation rate of 5%; the actual hydrogen load output was 60MW, and the predicted value was 63MW, with a deviation rate of 5%. Calculating the root mean square of these deviation rates yields a power forecast error of 5%.

[0066] Scene generation based on the time-series feature matrix employs Monte Carlo simulation, generating 500 power scenarios based on historical error distribution. The time-series distance is calculated using a dynamic time warping algorithm, which effectively measures the similarity between two time series. For example, two 4-hour (16 time points) wind power prediction scenarios, [90,92,95,98,100,102,105,106,105,103,100,98,95,93,90,88] MW and [88,90,93,96,99,101,104,105,106,104,101,99,96,94,91,89] MW, have a calculated time-series distance of 3.2.

[0067] A preset number of initial class centers are randomly selected, such as 20 randomly selected from 500 scenes. Each scene is assigned to the class of the class center with the closest temporal distance. For example, if scene S1 has a temporal distance of 2.5 from class center C3 and a distance greater than 2.5 from all other class centers, then S1 is assigned to the class of C3.

[0068] The process iteratively updates the scene with the highest average similarity to other scenes within each category as the new class center. For example, in category C3, there are scenes S1, S2, S3, S4, and S5. The average similarity of each scene to other scenes is calculated (based on the reciprocal of the temporal distance). Assuming that S2 has the highest average similarity, S2 is selected as the new class center. The sum of scene similarities within each class is calculated, and iteration stops when the change in the sum of scene similarities between two consecutive iterations is less than a preset convergence threshold (e.g., 0.01).

[0069] The current time is the root node. Five class centers are selected as leaf nodes. The probability of each leaf node is determined based on the proportion of scenarios included in each class to the total number of scenarios, such as 0.2, 0.15, 0.25, 0.3, and 0.1 respectively. The conditional probability and conditional expected value of each node in the probability scenario tree are calculated. Taking a certain leaf node as an example, the average wind power deviation among the included scenarios is +3MW, the average photovoltaic power deviation is -2MW, and the average hydrogen load deviation is +1MW. Then the conditional expected value of this node is [+3, -2, +1]MW. The conditional expected values ​​of each leaf node are weighted by probability to obtain the scenario expected value, such as [+2.5, -1.8, +0.7]MW.

[0070] The power correction is obtained by weighting the power prediction error and the scenario expectation value. For example, a power prediction error of 5% corresponds to a correction of 1.5MW; the power correction in the scenario expectation value is 2.5MW, with weights of 0.4 and 0.6 respectively, resulting in a final power correction of 2.1MW. This power correction is then added to the corresponding time period of the day-ahead baseline dispatch plan. For example, if the wind power output plan for the 14:00 time period in the day-ahead baseline dispatch plan is 90MW, adding the correction of 2.1MW will result in a revised plan value of 92.1MW.

[0071] Through the above steps, dynamic correction of the baseline scheduling plan based on forecast data at the intraday rolling layer is achieved, improving the accuracy and economy of coordinated scheduling of renewable energy and hydrogen energy systems.

[0072] In one alternative implementation, adjusting the real-time response layer based on system fluctuations includes:

[0073] In the real-time response layer, the real-time power sampling value is obtained based on the system operation data, and the deviation between the real-time power sampling value and the reference value is calculated to obtain the real-time power deviation. When the real-time power deviation of two adjacent sampling points exceeds the preset fluctuation threshold, it is determined that the system has fluctuated.

[0074] When the system fluctuates, the rate of change and integral value of the real-time power deviation are normalized according to a preset quantization factor to obtain the input variable, and the input variable is fuzzified using a triangular membership function to obtain the membership degree value.

[0075] Based on the rate of change of the real-time power deviation, the system state is divided into multiple fuzzy subsets, and an IF-THEN fuzzy rule base is established. The membership values ​​are substituted into the fuzzy rule base to obtain the output domain interval corresponding to each fuzzy subset. Discrete points are obtained by sampling the output domain interval corresponding to each fuzzy subset at equal intervals. The membership value corresponding to each discrete point is calculated. The centroid position of the fuzzy subset is obtained by summing the products of the abscissas of all discrete points and their corresponding membership values. The centroid position and its corresponding membership values ​​are weighted and summed to obtain the comprehensive output value. The comprehensive output value is multiplied by a preset proportional coefficient and inversely normalized to obtain the power adjustment amount.

[0076] The distribution ratio of hydrogen production power and hydrogen storage power is adjusted in real time according to the power adjustment amount.

[0077] A data acquisition mechanism is established at the real-time response layer to collect system operation data, including parameters such as hydrogen production system power, hydrogen storage system power, and energy input power, through a distributed sensor network. The system controller acquires real-time power samples with a sampling period of 10 milliseconds and compares them with a preset benchmark value. The benchmark value can be obtained through historical data analysis; for example, under stable system operation, the benchmark value for hydrogen production power is 500 kW, and the benchmark value for hydrogen storage power is 200 kW. The controller calculates the difference between the real-time power sample value and the benchmark value to obtain the real-time power deviation. When the system's power sample value at time t is 520 kW and the benchmark value is 500 kW, the real-time power deviation is 20 kW.

[0078] The system fluctuation judgment mechanism is based on the deviation value of continuous sampling points. The preset fluctuation threshold can be set to 5% of the baseline value, i.e., 25 kW. When the real-time power deviations at time t and t+1 are 26 kW and 27 kW respectively, both exceeding the preset fluctuation threshold of 25 kW, a fluctuation is judged to have occurred, and the adjustment mechanism needs to be activated.

[0079] After fluctuations occur, the rate of change and integral value of the real-time power deviation are calculated. The rate of change represents the speed at which the power deviation changes per unit time, and the integral value represents the cumulative effect of the power deviation over time. Assuming that the power deviation changes from 24 kW to 26 kW between time t-1 and time t, with a sampling period of 10 milliseconds, the rate of change is 200 kW / s. The integral value is calculated by summing the power deviations of multiple consecutive sampling points. If the sum of the power deviations of the most recent 10 sampling points is 250 kW·ms, then the integral value is 0.25 kW·s.

[0080] The rate of change and integral value are normalized to map them to the interval [-1, 1]. The preset quantization factor for the rate of change is set to 500 kW / s, and the preset quantization factor for the integral value is set to 1 kW·s. According to the normalization formula, the normalized value of the rate of change of 200 kW / s is 0.4, and the normalized value of the integral value of 0.25 kW·s is 0.25.

[0081] For the normalized input variables, a triangular membership function is used for fuzzification, converting precise numerical values ​​into membership values ​​of fuzzy sets. The triangular membership function is defined by three parameters (a, b, c), where b is the peak point, and a and c are the left and right boundary points. For example, a rate of change of 0.4 belongs to both the "small positive rate of change" and "medium positive rate of change" fuzzy subsets. The membership degree to the "small positive rate of change" subset is 0.6, and the membership degree to the "medium positive rate of change" subset is 0.4.

[0082] The system state is divided into multiple fuzzy subsets based on the rate of change of real-time power deviation, such as "large negative rate of change", "medium negative rate of change", "small negative rate of change", "zero rate of change", "small positive rate of change", "medium positive rate of change", and "large positive rate of change". For each fuzzy subset, a corresponding IF-THEN form fuzzy rule is established to form a complete fuzzy rule base. For example, the rule "IF rate of change is small positive rate of change AND integral value is small positive integral THEN output is small positive adjustment" indicates that when the system power rises slowly and there is already a certain cumulative deviation, a small positive adjustment is required.

[0083] Based on the membership values ​​of the input variables and the fuzzy rule base, the output universe interval corresponding to each fuzzy subset is determined. For example, the output universe interval of the "small positive adjustment" subset is [0.1, 0.3]. The output universe interval is sampled at equal intervals, such as sampling at intervals of 0.05 to obtain points 0.1, 0.15, 0.2, 0.25, and 0.3, and the membership value corresponding to each sampling point is calculated.

[0084] For each fuzzy subset, its centroid position is calculated, which is the sum of the products of the x-coordinates of all discrete points and their corresponding membership values, divided by the sum of all membership values. For example, if the membership values ​​corresponding to sampling points 0.1, 0.15, 0.2, 0.25, and 0.3 are 0.2, 0.5, 0.8, 0.5, and 0.2 respectively, then the centroid position of this fuzzy subset is 0.2.

[0085] The comprehensive output value is obtained by weighting and summing the centroid positions of all activated fuzzy subsets according to their corresponding membership values. Assuming two fuzzy subsets are activated with centroid positions of 0.2 and 0.3, and corresponding membership values ​​of 0.6 and 0.4, the comprehensive output value is 0.2 × 0.6 + 0.3 × 0.4 = 0.24.

[0086] The overall output value is multiplied by a preset proportional coefficient and then inversely normalized to obtain the actual power adjustment. Assuming the preset proportional coefficient is 100 kW, the power adjustment is 0.24 × 100 = 24 kW. This adjustment is used to adjust the distribution ratio between hydrogen production and hydrogen storage power. A reduction of 24 kW in hydrogen production power corresponds to a corresponding increase of 24 kW in hydrogen storage power, restoring the system to a stable state.

[0087] By following the steps above, timely adjustments can be made when power fluctuations occur, ensuring the stable operation of the overall system. This method is characterized by rapid response and strong adaptability, effectively addressing the challenges posed by renewable energy fluctuations and improving the operational efficiency and reliability of hydrogen production-storage systems.

[0088] In one optional implementation, a time-scale decoupling algorithm is used to construct a multi-level linkage mechanism that includes time-dimensional correlation constraints, and coordination and complementarity between different levels include:

[0089] Time series data is decomposed into multiple intrinsic mode components (IMCs) through empirical mode decomposition. The IMCs are then subjected to Hilbert transform to obtain the corresponding analytic signals. The instantaneous frequency is obtained by calculating the rate of change of the instantaneous phase of the analytic signals over time. The IMCs are then recombined based on the magnitude of the instantaneous frequency to obtain the wave hierarchy.

[0090] Calculate the covariance of the time series deviations between each pair of fluctuation levels and their mean deviations, and then divide it by the product of the standard deviations of the time series of the corresponding fluctuation levels to obtain the inter-level coupling degree; construct a state observer based on the frequency characteristics and dynamic response characteristics of each fluctuation level, and construct a feedback gain matrix using the observation results of the state observer; calculate the state deviation between the state of each fluctuation level and the desired state, and weight and superimpose the product of the state deviation and the feedback gain matrix with the inter-level coupling degree as the reference command quantity;

[0091] The performance index gradient is calculated in real time based on the system's operating status. The adaptive gain is updated based on the gradient descent algorithm. The rate of change of the coordination coefficient between each level is obtained by multiplying the negative value of the product of the hierarchical coupling degree and the state deviation by the adaptive gain. The rate of change of the coordination coefficient is integrated to obtain the inter-level coordination coefficient. The product of the inter-level coordination coefficient and the reference command quantity is used as the inter-level coordination control quantity.

[0092] Acquire time-series data of the system, which may include grid frequency fluctuations, energy consumption curves, or equipment operating parameters. Perform empirical mode decomposition on the acquired time-series data, decomposing it into multiple intrinsic mode components (IMFs). For example, decomposing the load data of a power system yields eight IMF components and one residual term, with each IMF representing a fluctuation component with different frequency characteristics.

[0093] For each IMF, a Hilbert transform is performed to obtain the corresponding analytic signal. Specifically, a Fourier transform is performed on the IMF signal, the negative frequency components are set to zero, then multiplied by 2, and finally an inverse Fourier transform is performed to obtain the analytic signal. The analytic signal is a complex signal; its real part is the original IMF, and its imaginary part is the Hilbert transform result of the IMF. Taking the first IMF as an example, its analytic signal can be expressed as the original IMF plus its Hilbert transform result multiplied by the imaginary unit *i*.

[0094] The instantaneous phase is obtained by calculating the argument of the analytical signal, and then the phase data is time-differentiated and divided by the sampling time interval to obtain the instantaneous frequency. For example, the instantaneous frequency calculation results of a certain IMF show that its frequency range fluctuates between 0.1-0.3Hz.

[0095] By setting frequency threshold ranges, IMFs with similar frequency characteristics can be combined into different fluctuation levels. For example, system fluctuations can be divided into low-frequency levels (0-0.05Hz), mid-frequency levels (0.05-0.5Hz), and high-frequency levels (above 0.5Hz). In specific implementations, a system might combine IMF1-IMF2 into a high-frequency level, IMF3-IMF5 into a mid-frequency level, and IMF6-IMF8 and the residual term into a low-frequency level.

[0096] Take the time series of two fluctuation levels, subtract their respective means to obtain the deviation series, calculate the covariance of these two deviation series, and then divide by the product of the standard deviations of the two time series to obtain the inter-level coupling degree. For example, for the mid-frequency level and the high-frequency level, the calculated coupling degree is 0.65, indicating that there is a strong correlation between them.

[0097] In state observer design, the system state equations contain the relationships between state variables at each level and the system inputs, while the observation equations describe the relationships between the system outputs and the state variables. State estimation is achieved using a Kalman filter algorithm. The prediction step predicts the state at the next moment based on the current state and the system model, and the update step corrects the predicted state by incorporating actual measurements. In a certain system observer design, the state vector contains state variables at three levels, and the system matrix reflects the dynamic characteristics of each level and their mutual influences.

[0098] Using the observations from the state observer, a state feedback control law is designed based on the theory of linear quadratic regulators, and the feedback gain matrix is ​​obtained by solving the Riccati equation. This matrix determines the contribution weight of each state variable to the control input. For example, in the feedback gain matrix of a certain system, the gain values ​​corresponding to high-frequency states are relatively small (0.1-0.3), the gain values ​​corresponding to mid-frequency states are moderate (0.3-0.7), and the gain values ​​corresponding to low-frequency states are relatively large (0.7-1.0), reflecting the differentiated processing strategy for disturbances of different frequencies.

[0099] The state deviation between each fluctuation level and the desired state is calculated. The product of the state deviation and the feedback gain matrix is ​​weighted and superimposed with the inter-level coupling degree to obtain the baseline command quantity. For example, if the state deviations of each level at a certain moment are [0.05, 0.12, 0.08], and the corresponding feedback gains are [0.8, 0.5, 0.2], after considering the weighting of the inter-level coupling degree, the baseline command quantity is 0.134.

[0100] Performance metrics include control deviation, energy consumption, and other aspects. The sensitivity of these metrics to control parameters is calculated to obtain gradients. The adaptive gain is updated based on the gradient descent algorithm, with the gain update amount proportional to the gradient. The learning rate is determined according to system stability requirements. For example, the initial adaptive gain is set to 1.0, and after gradient updates, it is adjusted to 1.25.

[0101] The rate of change of coordination coefficients between each level is obtained by multiplying the negative of the product of the level coupling degree and the state deviation by the adaptive gain. This design makes the coordination between levels with high coupling degree and large state deviation more compact. Taking the high-frequency and mid-frequency levels as an example, if the coupling degree is 0.65, the state deviation is 0.12, and the adaptive gain is 1.25, then the rate of change of coordination coefficients is calculated to be -0.09.

[0102] The coordination coefficient between levels is obtained by integrating the rate of change of the coordination coefficient, using numerical integration methods such as the trapezoidal rule. The initial value of the integral can be set as a preset constant or determined based on historical data. For example, the initial value of the coordination coefficient is set to 1.0, and then updated to 0.91 after one control cycle.

[0103] The product of the inter-level coordination coefficient and the baseline command quantity is used as the inter-level coordination control quantity. This control quantity is issued to the actuator as the final command. For example, if the baseline command quantity is 0.134 and the coordination coefficient is 0.91 at a certain moment, the final coordination control quantity is 0.122. This control quantity is applied to the corresponding level to achieve coordinated and complementary control between multiple levels.

[0104] Figure 2 This is a thermal diagram of the coupling degree matrix between IMF components, clearly illustrating the coupling degree matrix between IMF components constructed based on the time-scale decoupling algorithm of this invention. By calculating the covariance of the time series of each IMF component and its mean deviation, and then dividing by the product of the standard deviations of the corresponding time series, this invention achieves precise quantification of the correlation between different frequency levels. The 1.00 value on the diagonal of the matrix shows that each IMF component is fully coupled to itself. Adjacent IMF components exhibit strong coupling relationships; for example, the coupling degree between IMF1 and IMF2 reaches 0.78, demonstrating the effective identification of frequency domain continuity by this invention. As the frequency difference increases, the coupling degree shows a decreasing trend; the coupling degree between IMF1 and IMF8 is only 0.08, verifying the effective decoupling of fluctuation characteristics at different time scales by this invention. This invention utilizes this coupling degree matrix to construct a multi-level linkage mechanism, achieving coordinated complementarity between different frequency levels. The high-frequency layers (IMF1-IMF2) primarily handle rapid system disturbances, the mid-frequency layers (IMF3-IMF6) are responsible for regulation on medium timescales, and the low-frequency layers (IMF7-IMF8) focus on long-term trend control. Through a weighted superposition mechanism of inter-layer coupling, this invention can dynamically adjust the control strategy based on the correlation strength between layers, significantly improving the stability and response accuracy of the green hydrogen energy system.

[0105] In one optional implementation, at the hydrogen production power allocation layer, based on the power generation prediction results and the energy transmission loss index, the hydrogen production power allocation scheme for the hydrogen production unit includes:

[0106] The input power and conversion efficiency of each stage of DC-DC conversion are collected to obtain the conversion loss, and the equivalent resistance and transmission current of the transmission line are collected to obtain the line loss. The sum of the conversion loss and the line loss is used as the energy transmission loss index.

[0107] The power margin is obtained based on the difference between the predicted power generation and the load power. The power margin is multiplied by the preset power reserve coefficient to obtain the safe power margin. The power margin is subtracted from the sum of the safe power margin and the energy transmission loss index to obtain the allocable power.

[0108] Based on the allocable power, the upper limit constraint of hydrogen production power is set, the lower limit of the rated power of the hydrogen production device is set as the lower limit constraint of hydrogen production power, and the hydrogen production power change rate constraint is set according to the dynamic response characteristics of the hydrogen production device to construct a set of constraint conditions.

[0109] The entire prediction time domain is divided into several consecutive control periods. At the beginning of the current control period, the system operating status is obtained and the constraint set is updated. Based on the updated constraint set, the hydrogen production power sequence within the time period is solved. The first control variable of the hydrogen production power sequence is selected as the hydrogen production power control output at the current moment. The time domain is rolled forward by one control cycle and the solution steps are repeated until the entire prediction time domain is traversed. Based on the final hydrogen production power control output, a hydrogen production power allocation scheme that satisfies multiple constraints is obtained.

[0110] Input power and conversion efficiency at each stage of the DC-DC converter are collected to obtain conversion loss information. For example, if the input power of a DC-DC converter is 500kW and the conversion efficiency is 95%, then the conversion loss of this stage is 500kW × (1 - 95%) = 25kW. Simultaneously, the equivalent resistance and transmission current of the transmission line are collected to calculate the line loss. For example, if the equivalent resistance of a transmission line is 0.5Ω and the transmission current is 100A, then the line loss is I. 2 R = 100 2 A × 0.5Ω = 5000W = 5kW. Adding the conversion losses of all the above conversion stages to the line losses of all transmission lines yields the system's energy transmission loss index; for example, the total loss is 25kW + 5kW = 30kW.

[0111] If the predicted power generation is 800kW and the load power is 500kW at a certain moment, then the power margin is 800kW - 500kW = 300kW. To ensure the safe operation of the system, a preset power reserve coefficient needs to be set, for example, a value of 0.1, then the safe power margin is 300kW × 0.1 = 30kW. The allocable power is calculated by subtracting the safe power margin and the energy transmission loss index from the power margin, i.e., 300kW - (30kW + 30kW) = 240kW.

[0112] Based on the calculated allocable power, an upper limit constraint is set for hydrogen production power. For example, if the allocable power is 240kW, then the upper limit constraint for hydrogen production power is also 240kW. Simultaneously, a lower limit constraint is set for the rated power of the hydrogen production unit. For instance, if the lower limit of the rated power of the hydrogen production unit is 50kW, then the lower limit constraint for hydrogen production power is also 50kW. Considering the dynamic response characteristics of the hydrogen production unit, a constraint is set for the rate of change of hydrogen production power. For example, the change in hydrogen production power per minute should not exceed 5% of the unit's rated power. If the unit's rated power is 300kW, then the rate of change constraint for hydrogen production power is 15kW / min. These constraints together constitute a set of constraints.

[0113] In actual control, the entire prediction time domain is divided into several consecutive control periods. For example, a 4-hour prediction time domain is divided into 24 control periods of 10 minutes each. At the beginning of the current control period, such as 8:00, the system operating status is obtained and the constraint set is updated. Assuming that the predicted power generation is 720kW, the load power is 420kW, the power margin is 300kW, the safety power reserve is 30kW, and the energy transmission loss index is 28kW, then the allocable power is 300kW - (30kW + 28kW) = 242kW. The upper limit constraint for hydrogen production power is updated to 242kW, the lower limit constraint remains at 50kW, and the rate of change constraint remains at 15kW / min.

[0114] Based on the updated set of constraints, the hydrogen production power sequence for this time period is solved. For example, the hydrogen production power value per minute for the 10 minutes from 8:00 to 8:10 is solved as follows: [200kW, 215kW, 230kW, 230kW, 225kW, 220kW, 220kW, 210kW, 195kW, 180kW]. The first control variable of 200kW is selected from this sequence as the hydrogen production power control output at 8:00.

[0115] The time domain is rolled forward one control cycle to 8:01, the system operating status is reacquired, and the constraint set is updated. Assuming the predicted power generation is 725kW and the load power is 422kW, the recalculated power margin is 303kW, the safe power reserve is 30.3kW, and the energy transmission loss index is 28.5kW. Therefore, the allocable power is 303kW - (30.3kW + 28.5kW) = 244.2kW. The upper limit constraint for hydrogen production power is updated to 244.2kW, while the lower limit constraint remains at 50kW. However, the rate of change constraint needs to be considered. If the hydrogen production power at the previous moment was 200kW, then the current hydrogen production power cannot be lower than 185kW or higher than 215kW.

[0116] Based on the updated set of constraints, the hydrogen production power value per minute during the 10 minutes from 8:01 to 8:11 is calculated, and the first control variable is selected as the hydrogen production power control output at 8:01. This calculation process is repeated until the entire prediction time domain is traversed.

[0117] Based on the final hydrogen production power control output, a hydrogen production power allocation scheme that satisfies multiple constraints is obtained. This scheme considers multiple factors such as energy transmission losses, system safety margins, and equipment operating constraints, effectively utilizing renewable energy power generation to ensure the safe and efficient operation of the hydrogen production system. For example, the final hydrogen production power allocation scheme for a 4-hour period is as follows: average hydrogen production power of 210kW in the first hour, 230kW in the second hour, 190kW in the third hour, and 150kW in the fourth hour, for a total hydrogen production energy of 780kWh.

[0118] In one optional implementation, in the hydrogen storage capacity regulation layer, based on the hydrogen production power allocation scheme and the supply-demand balance index, the real-time adjustment of the hydrogen charging and discharging power of the hydrogen storage device includes:

[0119] The difference between the target hydrogen production power and the load power in the hydrogen production power allocation scheme is divided by the system rated power to obtain the supply and demand balance index; the difference between the current hydrogen storage capacity and the system minimum hydrogen storage capacity is divided by the difference between the system maximum hydrogen storage capacity and the system minimum hydrogen storage capacity to obtain the hydrogen storage capacity status index.

[0120] Based on the dynamic response characteristics of the hydrogen storage capacity state index, the state gain coefficient and the differential gain coefficient are determined, and the state gain coefficient and the differential gain coefficient are optimized using the steepest gradient method; the supply and demand balance index is multiplied by the preset proportional coefficient and integral coefficient and superimposed to obtain the power compensation amount; the deviation between the hydrogen storage capacity state index and the hydrogen storage capacity target value is multiplied by the state gain coefficient, and the product of this product and the rate of change of the hydrogen storage capacity state index multiplied by the differential gain coefficient is added to obtain the state feedback amount; the weighted sum of the power compensation amount and the state feedback amount is used as the hydrogen charging and discharging regulation amount;

[0121] Based on the physical characteristics of the hydrogen storage system, a constraint set is constructed, which includes hydrogen storage capacity limit constraints, hydrogen charging and discharging power limit constraints, and hydrogen charging and discharging power change rate constraints. The hydrogen charging and discharging adjustment quantities that do not meet the constraint set are mapped to the constraint boundaries to obtain the hydrogen storage capacity adjustment command.

[0122] The supply-demand balance index is calculated by dividing the difference between the target hydrogen production power and the load power by the system's rated power. For example, when the target hydrogen production power is 80 kW, the load power is 60 kW, and the system's rated power is 100 kW, the supply-demand balance index is (80-60) / 100 = 0.2, indicating that the system currently has a 20% capacity surplus.

[0123] The calculation process for the hydrogen storage capacity status index is as follows: subtract the system's minimum hydrogen storage capacity from the current hydrogen storage amount, and then divide by the difference between the system's maximum and minimum hydrogen storage capacities. For example, if the current hydrogen storage amount is 150 kg, the system's minimum hydrogen storage capacity is 50 kg, and the maximum hydrogen storage capacity is 250 kg, the hydrogen storage capacity status index is (150-50) / (250-50) = 0.5, indicating that the current hydrogen storage amount is in the middle of the available capacity range.

[0124] In determining the state gain coefficient and differential gain coefficient based on the dynamic response characteristics of the hydrogen storage capacity state index, the steepest gradient method is used to optimize these two coefficients. The initial state gain coefficient is set to 0.8, the differential gain coefficient to 0.3, and the learning rate to 0.05. According to the changing trend of the hydrogen storage capacity state index, the gradient descent method is used to optimize the coefficients. For example, when the hydrogen storage capacity state index increases from 0.5 to 0.55, the new state gain coefficient calculated by the steepest gradient method is 0.76, and the differential gain coefficient is 0.32.

[0125] The calculation process for the power compensation is as follows: the supply-demand balance index is multiplied by a preset proportional coefficient and an integral coefficient, and then summed. In this embodiment, the proportional coefficient is set to 0.6 and the integral coefficient is set to 0.4. When the supply-demand balance index is 0.2, the proportional term is 0.6 × 0.2 = 0.12. Assuming the cumulative value of the integral term is 0.08, the power compensation is 0.12 + 0.08 = 0.2, indicating that it is recommended to use 20% of the system's rated power for hydrogen charging.

[0126] The calculation of the state feedback quantity consists of two parts: one part is multiplying the deviation between the hydrogen storage capacity state index and the target value by the state gain coefficient; the other part is multiplying the rate of change of the hydrogen storage capacity state index by the differential gain coefficient. Assuming the target hydrogen storage capacity is 0.6, the current hydrogen storage capacity state index is 0.5, and the state gain coefficient is 0.76, then the first part is (0.6-0.5)×0.76=0.076. If the rate of change of the hydrogen storage capacity state index is 0.01 per second, and the differential gain coefficient is 0.32, then the second part is 0.01×0.32=0.0032. The state feedback quantity is 0.076+0.0032=0.0792.

[0127] The calculation process for the hydrogen charging / discharging adjustment amount is as follows: the weighted sum of the power compensation amount and the state feedback amount is used as the hydrogen charging / discharging adjustment amount. In this embodiment, the weight of the power compensation amount is set to 0.7, and the weight of the state feedback amount is set to 0.3. When the power compensation amount is 0.2 and the state feedback amount is 0.0792, the hydrogen charging / discharging adjustment amount is 0.7×0.2+0.3×0.0792=0.16376, indicating that it is recommended to charge hydrogen at 16.376% of the system's rated power.

[0128] Based on the physical characteristics of the hydrogen storage system, the process of constructing the constraint set includes: setting hydrogen storage capacity limits, hydrogen charging / discharging power limits, and hydrogen charging / discharging power change rate constraints. In this embodiment, the hydrogen storage capacity limit is 50 to 250 kg; the hydrogen charging / discharging power limit is -40% to 40% of the system's rated power, i.e., -40 kW to 40 kW; and the hydrogen charging / discharging power change rate constraint is no more than 10% of the system's rated power per minute, i.e., 10 kW per minute.

[0129] The process of mapping hydrogen charging / discharging adjustments that do not meet the constraint set to the constraint boundary is as follows: check whether the calculated hydrogen charging / discharging adjustment satisfies all the conditions of the constraint set. If not, adjust it to the nearest constraint boundary value. For example, when the calculated hydrogen charging / discharging adjustment is 0.16376, the corresponding actual power is 16.376 kW, which meets the hydrogen charging / discharging power limit constraint. Assuming that the hydrogen charging / discharging power at the previous moment was 5 kW, and it increased by 11.376 kW in one minute, exceeding the hydrogen charging / discharging power change rate constraint of 10 kW per minute, then the hydrogen charging / discharging adjustment needs to be adjusted to 0.15, corresponding to an actual power of 15 kW.

[0130] Finally, the hydrogen charging / discharging adjustment amount, after constraint processing, is sent to the hydrogen storage system as a hydrogen storage capacity adjustment command for execution. In this embodiment, the hydrogen storage capacity adjustment command is 0.15, instructing the hydrogen storage system to charge hydrogen at a power of 15 kilowatts. In this way, real-time optimization and adjustment of the hydrogen charging / discharging power of the hydrogen storage device can be achieved while meeting physical constraints, improving energy utilization efficiency and extending equipment life.

[0131] Figure 3This diagram illustrates a comparative analysis of hydrogen storage system performance, clearly demonstrating the significant advantages of this invention over traditional control methods. By comparing the performance of traditional PID control, fuzzy control, and this invention across five key indicators—response speed, control accuracy, energy efficiency ratio, system stability, and adaptability—it's evident that this invention achieves substantial improvements across all performance dimensions. The hydrogen storage capacity regulation method based on a time-scale decoupling algorithm and a multi-level linkage mechanism achieves a response speed of 94.2%, a 21.9 percentage point improvement over traditional PID control's 72.3% and a 14.4 percentage point improvement over fuzzy control's 79.8%. In terms of control accuracy, this invention achieves a high level of 95.8%, significantly surpassing the 68.5% of traditional methods and 82.4% of fuzzy control. Particularly noteworthy is the invention's outstanding performance in adaptability, reaching 97.3%. This is attributed to its adaptive mechanism using the steepest gradient method to optimize the state gain coefficient and differential gain coefficient, as well as its intelligent boundary mapping strategy based on constraint sets. This invention combines the supply and demand balance index with the hydrogen storage capacity status index to achieve coordinated control of power compensation and status feedback. This not only improves the system's response performance and control accuracy, but also significantly enhances the green hydrogen energy system's stable operation and adaptive adjustment capabilities under complex operating conditions.

[0132] This invention relates to a source-load interaction-based green hydrogen energy balance management system, the system comprising:

[0133] The first unit is used to extract key features of green hydrogen energy system data to construct an indicator set, which includes equipment capacity matching indicators, equipment operating efficiency indicators, energy transmission loss indicators, and supply and demand balance indicators.

[0134] The second unit is used to construct a three-layer nested optimization structure, which includes a day-ahead scheduling layer, an intraday rolling layer, and a real-time response layer. In the day-ahead scheduling layer, the equipment capacity matching index is used to formulate a baseline scheduling plan for system operation. In the intraday rolling layer, the baseline scheduling plan is dynamically corrected based on the predicted data. In the real-time response layer, adjustments are made according to system fluctuations. A time-scale decoupling algorithm is used to construct a multi-level linkage mechanism with time-dimensional correlation constraints, which coordinates and complements each other at different levels.

[0135] The third unit is used to coordinate the scheduling of the green hydrogen energy system using a hierarchical progressive supply and demand matching method based on the output results of the three-layer nested optimization structure. At the source-end power generation prediction layer, the power generation of renewable energy power generation devices is predicted and modeled based on the equipment operating efficiency index. At the hydrogen production power allocation layer, a hydrogen production power allocation scheme for the hydrogen production device is obtained based on the power generation prediction results and the energy transmission loss index. At the hydrogen storage capacity adjustment layer, the hydrogen charging and discharging power of the hydrogen storage device is adjusted in real time based on the hydrogen production power allocation scheme and the supply and demand balance index. Progressive information transmission and optimization are performed between each level, and a cross-level collaborative response mechanism is triggered when system fluctuations occur.

[0136] A third aspect of the present invention provides an electronic device, comprising:

[0137] processor;

[0138] Memory used to store processor-executable instructions;

[0139] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0140] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0141] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0142] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A green hydrogen energy balance management method based on source-load interaction, characterized in that, include: Key features of green hydrogen energy system data are extracted to construct an indicator set, which includes equipment capacity matching indicators, equipment operating efficiency indicators, energy transmission loss indicators, and supply and demand balance indicators. A three-layer nested optimization structure is constructed, comprising a day-ahead scheduling layer, an intraday rolling layer, and a real-time response layer: the day-ahead scheduling layer uses the equipment capacity matching index to formulate a baseline scheduling plan for system operation; the intraday rolling layer dynamically corrects the baseline scheduling plan based on predicted data; and the real-time response layer adjusts the plan according to system fluctuations. A time-scale decoupling algorithm is used to construct a multi-level linkage mechanism with time-dimensional correlation constraints, enabling coordination and complementarity between different layers. Based on the output of the three-layer nested optimization structure, a hierarchical progressive supply and demand matching method is adopted to coordinate the scheduling of the green hydrogen energy system. At the source-end power generation prediction layer, the power generation of renewable energy power generation devices is predicted and modeled based on the equipment operating efficiency index. At the hydrogen production power allocation layer, the hydrogen production power allocation scheme of the hydrogen production device is obtained based on the power generation prediction results and the energy transmission loss index. At the hydrogen storage capacity adjustment layer, the hydrogen charging and discharging power of the hydrogen storage device is adjusted in real time based on the hydrogen production power allocation scheme and the supply and demand balance index. Progressive information transmission and optimization are carried out between each level, and a cross-level collaborative response mechanism is triggered when the system fluctuates.

2. The method according to claim 1, characterized in that, Dynamically revising the baseline scheduling plan based on forecast data at the intraday rolling layer includes: In the intraday rolling layer, time-series features of historical system operation data are extracted to construct a time-series feature matrix; The time-series feature matrix is ​​subjected to nonlinear feature mapping, and combined with a long short-term memory neural network to perform rolling forecasts of renewable energy power generation and hydrogen load demand. The root mean square values ​​of the power generation forecast deviation rate and the hydrogen load forecast deviation rate are calculated to obtain the power forecast error. Based on the temporal feature matrix, scene generation is performed and the temporal distance between scenes is calculated. A preset number of initial class centers are randomly selected, and each scene is assigned to the class center with the closest temporal distance. The scene with the highest average similarity to other scenes in each class is iteratively updated as the new class center. The sum of scene similarities within the class is calculated. When the change in the sum of scene similarities between two adjacent iterations is less than a preset convergence threshold, the iteration stops. A probabilistic scene tree is constructed with the current time as the root node and the class centers as leaf nodes. The conditional probability and conditional expectation of each node in the probabilistic scene tree are calculated and weighted to obtain the scene expectation value. The power prediction error is weighted and averaged with the expected value of the scenario to obtain the power correction amount, which is then superimposed on the corresponding time period of the day-ahead baseline scheduling plan.

3. The method according to claim 1, characterized in that, Adjustments at the real-time response layer based on system fluctuations include: In the real-time response layer, the real-time power sampling value is obtained based on the system operation data, and the deviation between the real-time power sampling value and the reference value is calculated to obtain the real-time power deviation. When the real-time power deviation of two adjacent sampling points exceeds the preset fluctuation threshold, it is determined that the system has fluctuated. When the system fluctuates, the rate of change and integral value of the real-time power deviation are normalized according to a preset quantization factor to obtain the input variable, and the input variable is fuzzified using a triangular membership function to obtain the membership degree value. Based on the rate of change of the real-time power deviation, the system state is divided into multiple fuzzy subsets, and an IF-THEN fuzzy rule base is established. The membership values ​​are substituted into the fuzzy rule base to obtain the output domain interval corresponding to each fuzzy subset. Discrete points are obtained by sampling the output domain interval corresponding to each fuzzy subset at equal intervals. The membership value corresponding to each discrete point is calculated. The centroid position of the fuzzy subset is obtained by summing the products of the abscissas of all discrete points and their corresponding membership values. The centroid position and its corresponding membership values ​​are weighted and summed to obtain the comprehensive output value. The comprehensive output value is multiplied by a preset proportional coefficient and inversely normalized to obtain the power adjustment amount. The distribution ratio of hydrogen production power and hydrogen storage power is adjusted in real time according to the power adjustment amount.

4. The method according to claim 1, characterized in that, A multi-level linkage mechanism with temporal-dimensional correlation constraints is constructed using a time-scale decoupling algorithm, enabling coordination and complementarity between different levels, including: Time series data is decomposed into multiple intrinsic mode components (IMCs) through empirical mode decomposition. The IMCs are then subjected to Hilbert transform to obtain the corresponding analytic signals. The instantaneous frequency is obtained by calculating the rate of change of the instantaneous phase of the analytic signals over time. The IMCs are then recombined based on the magnitude of the instantaneous frequency to obtain the wave hierarchy. Calculate the covariance of the time series deviations between each pair of fluctuation levels and their mean deviations, and then divide it by the product of the standard deviations of the time series of the corresponding fluctuation levels to obtain the inter-level coupling degree; construct a state observer based on the frequency characteristics and dynamic response characteristics of each fluctuation level, and construct a feedback gain matrix using the observation results of the state observer; calculate the state deviation between the state of each fluctuation level and the desired state, and weight and superimpose the product of the state deviation and the feedback gain matrix with the inter-level coupling degree as the reference command quantity; The performance index gradient is calculated in real time based on the system's operating status. The adaptive gain is updated based on the gradient descent algorithm. The rate of change of the coordination coefficient between each level is obtained by multiplying the negative value of the product of the hierarchical coupling degree and the state deviation by the adaptive gain. The rate of change of the coordination coefficient is integrated to obtain the inter-level coordination coefficient. The product of the inter-level coordination coefficient and the reference command quantity is used as the inter-level coordination control quantity.

5. The method according to claim 1, characterized in that, In the hydrogen production power allocation layer, based on the power generation prediction results and the energy transmission loss index, the hydrogen production power allocation scheme for the hydrogen production unit includes: The input power and conversion efficiency of each stage of DC-DC conversion are collected to obtain the conversion loss, and the equivalent resistance and transmission current of the transmission line are collected to obtain the line loss. The sum of the conversion loss and the line loss is used as the energy transmission loss index. The power margin is obtained based on the difference between the predicted power generation and the load power. The power margin is multiplied by the preset power reserve coefficient to obtain the safe power margin. The power margin is subtracted from the sum of the safe power margin and the energy transmission loss index to obtain the allocable power. Based on the allocable power, the upper limit constraint of hydrogen production power is set, the lower limit of the rated power of the hydrogen production device is set as the lower limit constraint of hydrogen production power, and the hydrogen production power change rate constraint is set according to the dynamic response characteristics of the hydrogen production device to construct a set of constraint conditions. The entire prediction time domain is divided into several consecutive control periods. At the beginning of the current control period, the system operating status is obtained and the constraint set is updated. Based on the updated constraint set, the hydrogen production power sequence within the time period is solved. The first control variable of the hydrogen production power sequence is selected as the hydrogen production power control output at the current moment. The time domain is rolled forward by one control cycle and the solution steps are repeated until the entire prediction time domain is traversed. Based on the final hydrogen production power control output, a hydrogen production power allocation scheme that satisfies multiple constraints is obtained.

6. The method according to claim 1, characterized in that, In the hydrogen storage capacity regulation layer, based on the hydrogen production power allocation scheme and the supply-demand balance index, the real-time adjustment of the hydrogen charging and discharging power of the hydrogen storage device includes: The difference between the target hydrogen production power and the load power in the hydrogen production power allocation scheme is divided by the system rated power to obtain the supply and demand balance index; the difference between the current hydrogen storage capacity and the system minimum hydrogen storage capacity is divided by the difference between the system maximum hydrogen storage capacity and the system minimum hydrogen storage capacity to obtain the hydrogen storage capacity status index. Based on the dynamic response characteristics of the hydrogen storage capacity state index, the state gain coefficient and the differential gain coefficient are determined, and the state gain coefficient and the differential gain coefficient are optimized using the steepest gradient method; the supply and demand balance index is multiplied by the preset proportional coefficient and integral coefficient and superimposed to obtain the power compensation amount; the deviation between the hydrogen storage capacity state index and the hydrogen storage capacity target value is multiplied by the state gain coefficient, and the product of this product and the rate of change of the hydrogen storage capacity state index multiplied by the differential gain coefficient is added to obtain the state feedback amount; the weighted sum of the power compensation amount and the state feedback amount is used as the hydrogen charging and discharging regulation amount; Based on the physical characteristics of the hydrogen storage system, a constraint set is constructed, which includes hydrogen storage capacity limit constraints, hydrogen charging and discharging power limit constraints, and hydrogen charging and discharging power change rate constraints. The hydrogen charging and discharging adjustment quantities that do not meet the constraint set are mapped to the constraint boundaries to obtain the hydrogen storage capacity adjustment command.

7. A green hydrogen energy balance management system based on source-load interaction, used to implement the method as described in any one of claims 1-6, characterized in that, include: The first unit is used to extract key features of green hydrogen energy system data to construct an indicator set, which includes equipment capacity matching indicators, equipment operating efficiency indicators, energy transmission loss indicators, and supply and demand balance indicators. The second unit is used to construct a three-layer nested optimization structure, which includes a day-ahead scheduling layer, an intraday rolling layer, and a real-time response layer. In the day-ahead scheduling layer, the equipment capacity matching index is used to formulate a baseline scheduling plan for system operation. In the intraday rolling layer, the baseline scheduling plan is dynamically corrected based on the predicted data. In the real-time response layer, adjustments are made according to system fluctuations. A time-scale decoupling algorithm is used to construct a multi-level linkage mechanism with time-dimensional correlation constraints, which coordinates and complements each other at different levels. The third unit is used to coordinate the scheduling of the green hydrogen energy system using a hierarchical progressive supply and demand matching method based on the output results of the three-layer nested optimization structure. At the source-end power generation prediction layer, the power generation of renewable energy power generation devices is predicted and modeled based on the equipment operating efficiency index. At the hydrogen production power allocation layer, a hydrogen production power allocation scheme for the hydrogen production device is obtained based on the power generation prediction results and the energy transmission loss index. At the hydrogen storage capacity adjustment layer, the hydrogen charging and discharging power of the hydrogen storage device is adjusted in real time based on the hydrogen production power allocation scheme and the supply and demand balance index. Progressive information transmission and optimization are performed between each level, and a cross-level collaborative response mechanism is triggered when system fluctuations occur.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images