New energy electrolytic hydrogen production yield prediction method based on center radial clustering
By using a central radial clustering method, weather data clusters were divided and a hydrogen production model was constructed, which solved the problems of large errors and inaccuracies in the prediction of hydrogen production from new energy electrolysis, and achieved more efficient hydrogen production prediction and system optimization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SOUTHWEST JIAOTONG UNIV
- Filing Date
- 2026-03-05
- Publication Date
- 2026-06-12
AI Technical Summary
Existing technologies for predicting hydrogen production from new energy electrolysis suffer from problems such as overfitting of training datasets, lack of direct conversion relationships between wind power generation and solar irradiance to hydrogen production, and difficulty in distinguishing the volatility of weather data using traditional clustering methods, resulting in large prediction errors and insufficient accuracy.
A central radial clustering method was adopted to divide historical weather data into two categories: no sunshine and sunshine. Clusters were divided according to the central radial distance and linkage criteria. A hydrogen production model was constructed to directly establish a weather-hydrogen production mapping. Hydrogen production was predicted by fitting the relationship between historical data.
It reduces the prediction error of hydrogen production, improves prediction accuracy and robustness, and is suitable for new energy electrolysis hydrogen production systems that are multi-energy coupled and subject to weather fluctuations, enabling efficient and economical operation.
Smart Images
Figure CN122198224A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of new energy power systems and data prediction technology, specifically relating to a method for predicting the production of new energy electrolytic hydrogen based on central radial clustering. Background Technology
[0002] Energy shortages and environmental pollution have become major threats to the sustainable development of society today, and renewable energy-based water electrolysis for hydrogen production is an important direction for achieving energy transition. However, this field currently faces the following technological bottlenecks in production forecasting:
[0003] In the scenario of predicting the production of hydrogen by electrolysis in new energy based on wind speed, if a hybrid wind speed prediction technology of k-NN-CEEMDAN-bidlstm is adopted, that is, using k-nearest neighbors for data input, using complete ensemble empirical mode decomposition with adaptive noise to denoise the wind speed sequence, and using a bidirectional long short-term memory network for final prediction, this AI-based model can flexibly handle nonlinear datasets, but it will lead to overfitting of the training dataset.
[0004] In scenarios where hydrogen production from renewable energy sources via electrolysis is predicted based on solar irradiance, hybrid prediction algorithms, such as the CEEMDAN-BiDLSTM and CEEMD-SCM-GA-LGBM algorithms, are frequently used. These hybrid algorithms better reveal the fluctuations in solar irradiance over time, thus improving the accuracy of solar irradiance prediction. However, they lack consideration for hydrogen production from wind power generation and the direct conversion relationship between wind and solar irradiance to hydrogen production. Accurate prediction of hydrogen production from renewable energy sources via electrolysis is a crucial prerequisite for achieving automated scheduling of hydrogen energy systems, optimizing energy storage resource allocation, and reducing operating costs.
[0005] In existing technologies, hydrogen production forecasting typically employs a multi-stage indirect modeling approach: first predicting weather, then predicting renewable energy generation, and finally estimating hydrogen production. This method ignores the direct coupling effect of wind speed and irradiance on hydrogen production and suffers from significant cumulative errors. Furthermore, traditional clustering methods, such as K-means and hierarchical clustering, use distance metrics with fixed formulas, such as Euclidean distance, Manhattan distance, and Chebyshev distance. Existing connection methods, such as single links, average links, and centroid links, struggle to effectively distinguish between weather data under conditions of sunshine and lack sufficient sensitivity to fluctuations in wind speed and irradiance, resulting in an inaccurate mapping between clustering results and hydrogen production. Summary of the Invention
[0006] To address the aforementioned issues, this invention proposes a central radial clustering-based method for predicting hydrogen production from renewable energy electrolysis. By integrating the characteristics of hydrogen production prediction methods from wind power and photovoltaic power generation, a central radial clustering method is proposed. This method minimizes the error between predicted and actual hydrogen production in various scenarios, making it suitable for renewable energy electrolysis hydrogen production systems that are multi-energy coupled and highly susceptible to weather fluctuations. Based on traditional clustering methods, this invention proposes a central radial clustering method based on central radial distance and central radial link. This provides more accurate data clustering for different hydrogen production models, directly establishing a "weather-hydrogen production" mapping model, avoiding multi-stage prediction errors, and improving prediction accuracy and robustness. It is applicable to day-ahead scheduling and operation optimization of renewable energy electrolysis hydrogen production systems.
[0007] The present invention provides a new energy electrolysis hydrogen production prediction method based on central radial clustering. First, historical weather data is collected. Then, the presence of sunlight is determined, and the historical weather data is divided into two categories: data with no sunlight and data with sunlight. Second, according to the central radial clustering method, the radial distance between the weather data points in each cluster is calculated. Then, based on the central radial links and according to the national wind energy resource zoning standards and solar energy resource zoning standards, the data with no sunlight is divided into X clusters based on wind intensity, and the data with sunlight is divided into Y clusters based on wind and sunlight intensity. Next, by fitting the relationship between historical weather data and hydrogen production data, the most suitable prediction benchmark is selected to construct a hydrogen production model. Then, based on the weighted correlation between wind and sunlight intensity and the corresponding clusters, i.e., firstly, based on the cluster to which the predicted hour belongs in the historical N days, the weighting coefficient of each cluster is calculated. Then, based on the different weighting coefficients of the hydrogen production model, the hydrogen production amount for the predicted hour is obtained. Finally, the required daily hydrogen production can be predicted, enabling better participation in demand response projects and achieving efficient and economical operation of the electrolysis hydrogen production system.
[0008] The specific technical solution adopted in this invention is as follows:
[0009] A method for predicting hydrogen production from electrolysis in new energy sources based on centroradial clustering includes the following steps:
[0010] S1. Statistically analyze historical data, compile data on wind intensity, sunlight intensity and new energy electrolysis hydrogen production for 24 hours every day for N days past the local event date, and correct abnormal data to obtain the processed data;
[0011] S2. Cluster the processed data into two clusters based on the presence or absence of light intensity, and calculate the radial distance between the data points in each of the two clusters using the centroradial clustering method; specifically including:
[0012] S21, Let Indicates the first Heavenly Hourly wind intensity and sunlight intensity, Indicates the first Heavenly Hourly wind intensity and sunlight intensity, Indicates the first Heavenly Hourly wind intensity Indicates the first Heavenly Hourly light intensity, Indicates the first Heavenly Hourly wind intensity Indicates the first Heavenly The light intensity per hour, according to , Whether the value is 0, the weather data points are divided into two clusters: one with light intensity and one without light intensity. For the cluster without light intensity;
[0013] S22, In the cluster with no light intensity, that is, when and At that time, calculate the center point of all wind intensity data to and The distance between them is:
[0014] ,
[0015] ,
[0016] in and This represents the number of days and hours with zero light intensity.
[0017] S23. Calculate the data points in the cluster with no light intensity. and Distance between:
[0018] ,
[0019] in ;
[0020] S24. In clusters with light intensity, i.e. when and At that time, calculate the center point of all wind intensity data and and distance and :
[0021] ,
[0022] ,
[0023] in and The number of days and hours that light intensity exists;
[0024] S25. Calculate the data points within the cluster of light intensity. and The distance between them is:
[0025] ;
[0026] S3. Using central radial clustering, based on the wind energy resource zoning standards and solar energy resource zoning standards, weather data with no solar radiation intensity are divided into X clusters according to wind intensity, and weather data with solar radiation intensity are divided into Y clusters according to both wind intensity and solar radiation intensity; specifically:
[0027] S31. In the weather data with no sunlight intensity, define that there are M weather data points. First, divide the weather data with no sunlight intensity into M clusters, and use the method in S2 to calculate the distance between each cluster. The two clusters with the smallest distance are merged into one cluster, thus reducing the number of weather data clusters to M-1. This process continues until the number of weather data clusters is reduced to X.
[0028] S32. In weather data with available light intensity, define V weather data points as coexisting. Use the method in S2 to calculate the distance between each cluster. The two clusters with the smallest distance are merged into one cluster, thus reducing the number of weather data clusters to V-1. This process continues until the number of weather data clusters is reduced to Y.
[0029] S4. By fitting historical weather data and hydrogen production data, the most suitable prediction benchmark is selected to construct a hydrogen production model; specifically:
[0030] S41. For the predicted target points, the weather data without sunlight intensity are divided into two clusters: high wind intensity without sunlight intensity HW-W / OI and low wind intensity without sunlight intensity; the weather data with sunlight intensity are divided into four clusters: high wind intensity with high sunlight intensity HW-W / HI, low wind intensity with high sunlight intensity LW-W / HI, high wind intensity with low sunlight intensity HW-W / LI and low wind intensity with low sunlight intensity LW-W / LI. The method in S3 is used to perform central radial clustering on the weather data.
[0031] S42. Based on the historical hydrogen production at the target point, and combined with the correlation between hydrogen production in each cluster and weather data, analyze the relevant data, compare the goodness of fit of different function forms, and select the optimal model, specifically:
[0032] In the no-light weather data cluster, a model of hydrogen production and weather data is constructed, resulting in hydrogen production as a univariate function of wind intensity, expressed in five functional forms: linear, logarithmic, exponential, power, and quadratic, as follows:
[0033] ,
[0034] in, The hydrogen production rate under no light intensity. Let represent the wind intensity under no-light conditions, where A, B, C, and D are linear coefficients, E is the quadratic coefficient, a, b, c, d, and f are constant terms, and e is the linear coefficient. As the base, As the base, The function is a power; these functions are tested on samples using the least squares method, and the optimal fitting function is determined by comparing the goodness of fit.
[0035] In a weather data cluster with available light intensity, a model of hydrogen production and weather data is constructed. Hydrogen production is obtained as a bivariate function of wind intensity and light intensity, expressed as a linear, logarithmic, exponential, power, and quadratic function, respectively, and then superimposed. The relationship between hydrogen production and wind intensity, which is affected by wind intensity, is expressed in the following five forms:
[0036] ,
[0037] in, Hydrogen production is affected by wind intensity. Let H, K, L, and M be the linear coefficients, N be the quadratic coefficient, h, k, l, m, and o be the constants, and n be the linear coefficient. As the base, As the base, It is an exponent;
[0038] The expression for hydrogen production affected by light intensity in relation to light intensity is:
[0039] ,
[0040] in, Hydrogen production is affected by light intensity. Light intensity, , , , The coefficients are linear. The coefficient of the quadratic term, , , , , For constant terms, The coefficient of the linear term, As the base, As the base, It is an exponent;
[0041] The relationship between hydrogen production and weather data under conditions of light intensity is as follows:
[0042] ,
[0043] in To determine the hydrogen production rate under illumination, the function was tested using the least squares method on a sample, and the best fitting function was selected as the hydrogen production model by comparing the goodness of fit.
[0044] S5. Based on the clusters to which the predicted time period belongs in the same hour over the historical N days, calculate the weighting coefficients for each cluster. Then, based on the hydrogen production model of each cluster and its weighting coefficients, obtain the hydrogen production for the predicted time period, specifically:
[0045] Based on the obtained hydrogen production model, the weighting coefficients for the HW-W / OI, LW-W / OI, HW-W / HI, LW-W / HI, HW-W / LI, and LW-W / LI clusters were calculated as follows: , , , , , ,in , , , , The percentage coefficients corresponding to HW-W / OI, LW-W / OI, HW-W / HI, LW-W / HI, HW-W / LI, and LW-W / LI are as follows:
[0046] ;
[0047] The hydrogen production for the predicted period is calculated as follows:
[0048] ,
[0049] in, The hydrogen production in hour j on day i. , , , , and These represent the average hydrogen production of each cluster (HW-W / OI, LW-W / OI, HW-W / HI, LW-W / HI, HW-W / LI, and LW-W / LI) during the historical days for the predicted hour.
[0050] Furthermore, the specific method of S1 is as follows:
[0051] S11. Based on the 24-hour data for each of the past N days from the event date, plot box plots of wind intensity, light intensity, and hydrogen production. Then, identify outliers in the data according to Tukey's 1.5×IQR rule, i.e., calculate the first quartile Q1, median Q2, and third quartile Q3 to obtain the interquartile range. The upper bound of outliers is The lower boundary of outliers is Data points that are less than the lower bound or greater than the upper bound are considered outliers.
[0052] S12. For the identified outliers, replace the outlier data for the corresponding hour with the average of the weather data for the two nearest normal hours: , , in, , and Correction data was provided to replace anomalies in wind intensity, light intensity, and hydrogen production. , , , , , These are adjacent normal data for wind intensity, light intensity, and hydrogen production, respectively, where subscripts i and j represent the j-th hour of the i-th day.
[0053] The beneficial effects of this invention are as follows: This invention proposes a new energy electrolysis hydrogen production prediction method based on central radial clustering. It comprehensively adopts the characteristics of production prediction methods for hydrogen production from wind power generation and photovoltaic power generation, thereby reducing the error between the predicted hydrogen production value and the actual hydrogen production of the new energy electrolysis hydrogen production system, which is highly affected by weather factors. This allows the system to better participate in demand response projects and achieve efficient and economical operation of the electrolysis hydrogen production system. Specifically, it has the following advantages: (1) The new energy electrolysis hydrogen production prediction method based on central radial clustering comprehensively considers the characteristics of different hydrogen production system production prediction methods, enabling the electrolysis hydrogen production system to obtain a more accurate production prediction, thereby better participating in demand response projects and achieving efficient and economical operation of the electrolysis hydrogen production system. (2) This invention comprehensively considers the different scenarios faced by the prediction of new energy electrolysis hydrogen production, making the predicted hydrogen production more convincing. (3) Compared with the traditional hybrid prediction method, this invention adopts the characteristics of the hydrogen production prediction methods of wind power generation and photovoltaic power generation, and integrates them to form a new energy electrolysis hydrogen production prediction method based on central radial clustering, which is more adaptable and targeted in the production prediction scenario of new energy electrolysis hydrogen production. (4) This invention incorporates the central radial clustering method, which makes the hydrogen production prediction results more close to the actual value. Attached Figure Description
[0054] Figure 1 This is a flowchart of a new energy electrolysis hydrogen production prediction method based on central radial clustering. Detailed Implementation
[0055] The present invention will now be described in detail with reference to the accompanying drawings.
[0056] like Figure 1 As shown, the present invention generally includes the following steps:
[0057] Step 1: Statistically analyze historical data and compile data on wind intensity, sunlight intensity, and hydrogen production from new energy electrolysis for each of the 24 hours in the N days prior to the local event date, and remove outlier data;
[0058] Step 2: Cluster the processed data points into two clusters based on the presence or absence of light intensity. Calculate the radial distance between the data points in each of the two clusters using the centroradial clustering method.
[0059] Step 3: Subsequently, through central radial clustering, based on the national wind energy resource zoning standards and solar energy resource zoning standards, the weather data without sunlight intensity are divided into X clusters according to wind intensity, and the weather data with sunlight intensity are divided into Y clusters according to both wind intensity and sunlight intensity.
[0060] Step 4: By fitting historical weather data and hydrogen production data, select the most suitable prediction benchmark and construct a hydrogen production model;
[0061] Step 5: Based on the clusters to which the predicted period belongs in the same hour over the historical N days, calculate the weighting coefficients for each cluster, and then obtain the hydrogen production for the predicted period based on the hydrogen production model of each cluster and its weighting coefficients.
[0062] In the aforementioned method for predicting hydrogen production from electrolysis based on central radial clustering, step 1 involves statistically analyzing historical data to compile data on wind intensity, solar radiation intensity, and hydrogen production from electrolysis for each of the past N days on the event date at the local location, and then removing outlier data. This process includes the following steps:
[0063] Step 1.1: Based on the 24-hour data of each of the past N days from the event date, plot box plots of wind intensity, light intensity, and hydrogen production. Then, identify outliers in the data according to Tukey's 1.5×IQR rule, i.e., calculate the first quartile (Q1), median (Q2), and third quartile (Q3) to obtain the interquartile range. The upper bound of outliers is The lower boundary of outliers is Data points that are less than the lower bound or greater than the upper bound are considered outliers.
[0064] Step 1.2: Replace the outlier data for this hour with the average of the weather data from the two nearest normal hours: , , in, , and Correction data was provided to replace anomalies in wind intensity, light intensity, and hydrogen production. , , , , , These are adjacent normal data for wind intensity, light intensity, and hydrogen production, respectively, where subscripts i and j represent the j-th hour of the i-th day.
[0065] In the above-mentioned method for predicting hydrogen production from electrolysis in new energy sources based on centroradial clustering, step 2 involves clustering the processed data points into two clusters based on the presence or absence of light intensity, and then calculating the centroradial distance between data points in each of the two clusters using centroradial clustering. This process includes the following steps:
[0066] Step 2.1: First set Indicates the first Heavenly Hourly wind intensity and sunlight intensity, Indicates the first Heavenly Hourly wind intensity and sunlight intensity, according to Whether the value is 0 or not, the weather data points are divided into two clusters: one with light intensity and one without light intensity.
[0067] Step 2.2: In the cluster with no light intensity, i.e. when and At that time, calculate the center point of all wind intensity data to and The distance between them is
[0068]
[0069]
[0070] in , This represents the number of days and hours when the light intensity was 0.
[0071] Step 2.3: Calculate data points in the cluster with no light intensity. and The distance between them and For weather data points containing wind speed and light intensity for a given hour, ,in
[0072] Step 2.4: In clusters with light intensity, i.e. when and At that time, calculate the center point of all wind intensity data and and The distance will be expressed as
[0073]
[0074]
[0075] in , This refers to the number of days and hours that light intensity exists.
[0076] Step 2.5: Calculate data points within the clusters with illumination intensity. and The distance between them is
[0077]
[0078] in, and This refers to weather data points that include wind intensity and light intensity for a given hour.
[0079] In the aforementioned method for predicting hydrogen production from electrolysis in new energy sources based on central radial clustering, step 3 involves using central radial clustering to divide weather data without solar radiation into X clusters based on wind intensity, and weather data with solar radiation into Y clusters based on both wind and solar intensity, according to national wind energy resource zoning standards and solar energy resource zoning standards. Hierarchical clustering of the weather data is then performed using the central radial clustering method. This process includes the following steps:
[0080] Step 3.1: First, introduce the distance formula between data clusters in the central radial clustering method. The central radial clustering method divides data based on the maximum central radial distance between data points in each cluster. Let there exist any two data clusters C1 and C2, then... , where a and b are data points in two data clusters, and d(a,b) is the radial distance between the centers of a and b.
[0081] Step 3.2: In the no-sunlight intensity weather data, assuming there are M weather data points, first divide the no-sunlight intensity weather data into M clusters, and use the central radial clustering method to calculate the distance between each cluster. The two clusters with the smallest distance are merged into one cluster, thus reducing the number of weather data clusters to M-1. This process continues until the number of weather data clusters is reduced to X.
[0082] Step 3.3: In the weather data with sunshine intensity, suppose there are V weather data points. First, divide the weather data without sunshine intensity into V clusters. Then, use the central radial clustering method to calculate the distance between each cluster. The two clusters with the smallest distance are merged into one cluster, thus reducing the number of weather data clusters to V-1. This process continues until the number of weather data clusters is reduced to Y.
[0083] In the aforementioned method for predicting hydrogen production from electrolysis based on central radial clustering, step 4 involves fitting historical weather data and hydrogen production data to select the most suitable prediction benchmark and construct a hydrogen production model. This process includes the following steps.
[0084] Step 4.1: Assuming that the event location is in a wind energy available area and a solar energy rich zone, the weather data with no sunshine intensity can be divided into two clusters: HW-W / OI (high wind intensity with no sunshine intensity) and LW-W / OI (low wind intensity with no sunshine intensity). The weather data with sunshine intensity can be divided into four clusters: HW-W / HI (high wind intensity with high sunshine intensity), LW-W / HI (low wind intensity with high sunshine intensity), HW-W / LI (high wind intensity with low sunshine intensity) and LW-W / LI (low wind intensity with low sunshine intensity). Then, perform central radial clustering on the weather data according to Step 3.
[0085] Step 4.2: Based on the historical hydrogen production of the station and combined with the correlation between hydrogen production and weather data in each cluster in Step 4.1, analyze the relevant data and compare the goodness of fit of different function forms.
[0086] Step 4.3: In the no-irradiance weather data cluster, construct a model of hydrogen production and weather data, obtaining hydrogen production as a univariate function of wind intensity, which can be expressed in five functional forms: linear, logarithmic, exponential, power, or quadratic, as shown in the following equation:
[0087]
[0088] in, The hydrogen production rate under no light intensity. Let represent the wind intensity under no-light conditions, where A, B, C, and D are linear coefficients, E is the quadratic coefficient, a, b, c, d, and f are constant terms, and e is the linear coefficient. As the base, As the base, The function is a power. These functional forms are tested using the least squares method on the samples, and the best-fit function is determined by comparing the goodness of fits.
[0089] Step 4.4: In the weather data cluster with light intensity, construct a model of hydrogen production and weather data, obtaining hydrogen production as a bivariate function of wind intensity and light intensity. This model can be expressed as a superposition of five different functional forms: linear, logarithmic, exponential, power, or quadratic. The relationship between hydrogen production and wind intensity, which is affected by wind intensity, can be expressed in the following five forms:
[0090]
[0091] in, Hydrogen production is affected by wind intensity. Let H, K, L, and M be the linear coefficients, N be the quadratic coefficient, h, k, l, m, and o be the constants, and n be the linear coefficient. As the base, As the base, It is a power.
[0092] The expression for hydrogen production affected by light intensity is as follows:
[0093]
[0094] in, Hydrogen production is affected by light intensity. Light intensity, , , , The coefficients are linear. The coefficient of the quadratic term, , , , , For constant terms, The coefficient of the linear term, As the base, As the base, It is a power.
[0095] Therefore, the relationship between hydrogen production and weather data under conditions of light intensity can be derived as follows: ,in The hydrogen production rate under illumination is given. These functional forms are tested using the least squares method on a sample, and the best-fit function is determined by comparing the goodness-of-fit.
[0096] In the aforementioned method for predicting hydrogen production from electrolysis based on central radial clustering, step 5 involves calculating the weighting coefficients of each cluster based on the clusters to which the predicted period belongs within the same hour over N historical days. Then, based on the hydrogen production model of each cluster and its weighting coefficients, the hydrogen production for the predicted period is obtained. This process includes the following steps:
[0097] Step 5.1: Based on the hydrogen production prediction model established in Step 4, calculate the weighting coefficients for the HW-W / OI, LW-W / OI, HW-W / HI, LW-W / HI, HW-W / LI, and LW-W / LI clusters. , , , , , ,in , , , , The percentage coefficients corresponding to HW-W / OI, LW-W / OI, HW-W / HI, LW-W / HI, HW-W / LI, and LW-W / LI are as follows:
[0098]
[0099] Step 5.2: Calculate the hydrogen production for the predicted period:
[0100]
[0101] in, Hydrogen production on day i, hour j , ,
[0102] , , and Let represent the average hydrogen production of each cluster (HW-W / OI, LW-W / OI, HW-W / HI, LW-W / HI, HW-W / LI, and LW-W / LI) during the historical days for the predicted hour.
Claims
1. A method for predicting hydrogen production from electrolysis in new energy sources based on centroradial clustering, characterized in that, Includes the following steps: S1. Statistically analyze historical data, compile data on wind intensity, sunlight intensity and new energy electrolysis hydrogen production for 24 hours every day for N days past the local event date, and correct abnormal data to obtain the processed data; S2. Cluster the processed data into two clusters based on the presence or absence of light intensity, and calculate the radial distance between the data points in each of the two clusters using the centroradial clustering method; specifically including: S21, Let Indicates the first Heavenly Hourly wind intensity and sunlight intensity, Indicates the first Heavenly Hourly wind intensity and sunlight intensity, Indicates the first Heavenly Hourly wind intensity Indicates the first Heavenly Hourly light intensity, Indicates the first Heavenly Hourly wind intensity Indicates the first Heavenly The light intensity per hour, according to , Whether the value is 0, the weather data points are divided into two clusters: one with light intensity and one without light intensity. For the cluster without light intensity; S22, In the cluster with no light intensity, that is, when and At that time, calculate the center point of all wind intensity data to and The distance between them is: ， ， in and This represents the number of days and hours with zero light intensity. S23. Calculate the data points in the cluster with no light intensity. and Distance between: ， in ; S24. In clusters with light intensity, i.e. when and At that time, calculate the center point of all wind intensity data and and distance and : ， ， in and The number of days and hours that light intensity exists; S25. Calculate the data points within the cluster of light intensity. and The distance between them is: ； S3. Using central radial clustering, based on the wind energy resource zoning standards and solar energy resource zoning standards, weather data with no solar radiation intensity are divided into X clusters according to wind intensity, and weather data with solar radiation intensity are divided into Y clusters according to both wind intensity and solar radiation intensity; specifically: S31. In the weather data with no sunlight intensity, define that there are M weather data points. First, divide the weather data with no sunlight intensity into M clusters, and use the method in S2 to calculate the distance between each cluster. The two clusters with the smallest distance are merged into one cluster, thus reducing the number of weather data clusters to M-1. This process continues until the number of weather data clusters is reduced to X. S32. In weather data with available light intensity, define V weather data points as coexisting. Use the method in S2 to calculate the distance between each cluster. The two clusters with the smallest distance are merged into one cluster, thus reducing the number of weather data clusters to V-1. This process continues until the number of weather data clusters is reduced to Y. S4. By fitting historical weather data and hydrogen production data, the most suitable prediction benchmark is selected to construct a hydrogen production model; specifically: S41. For the predicted target points, the weather data without sunlight intensity are divided into two clusters: high wind intensity without sunlight intensity HW-W / OI and low wind intensity without sunlight intensity; the weather data with sunlight intensity are divided into four clusters: high wind intensity with high sunlight intensity HW-W / HI, low wind intensity with high sunlight intensity LW-W / HI, high wind intensity with low sunlight intensity HW-W / LI and low wind intensity with low sunlight intensity LW-W / LI. The method in S3 is used to perform central radial clustering on the weather data. S42. Based on the historical hydrogen production at the target point, and combined with the correlation between hydrogen production in each cluster and weather data, analyze the relevant data, compare the goodness of fit of different function forms, and select the optimal model, specifically: In the no-light weather data cluster, a model of hydrogen production and weather data is constructed, resulting in hydrogen production as a univariate function of wind intensity, expressed in five functional forms: linear, logarithmic, exponential, power, and quadratic, as follows: ， in, The hydrogen production rate under no light intensity. Let represent the wind intensity under no-light conditions, where A, B, C, and D are linear coefficients, E is the quadratic coefficient, a, b, c, d, and f are constant terms, and e is the linear coefficient. As the base, As the base, The function is a power; these functions are tested on samples using the least squares method, and the optimal fitting function is determined by comparing the goodness of fit. In a weather data cluster with available light intensity, a model of hydrogen production and weather data is constructed. Hydrogen production is obtained as a bivariate function of wind intensity and light intensity, expressed as a linear, logarithmic, exponential, power, and quadratic function, respectively, and then superimposed. The relationship between hydrogen production and wind intensity, which is affected by wind intensity, is expressed in the following five forms: ， in, Hydrogen production is affected by wind intensity. Let H, K, L, and M be the linear coefficients, N be the quadratic coefficient, h, k, l, m, and o be the constants, and n be the linear coefficient. As the base, As the base, It is an exponent; The expression for hydrogen production affected by light intensity in relation to light intensity is: ， in, Hydrogen production is affected by light intensity. Light intensity, , , , The coefficients are linear. The coefficient of the quadratic term, , , , , For constant terms, The coefficient of the linear term, As the base, As the base, It is an exponent; The relationship between hydrogen production and weather data under conditions of light intensity is as follows: ， in To determine the hydrogen production rate under illumination, the function was tested using the least squares method on a sample, and the best fitting function was selected as the hydrogen production model by comparing the goodness of fit. S5. Based on the clusters to which the predicted time period belongs in the same hour over the historical N days, calculate the weighting coefficients for each cluster. Then, based on the hydrogen production model of each cluster and its weighting coefficients, obtain the hydrogen production for the predicted time period, specifically: Based on the obtained hydrogen production model, the weighting coefficients for the HW-W / OI, LW-W / OI, HW-W / HI, LW-W / HI, HW-W / LI, and LW-W / LI clusters were calculated as follows: , , , , , ,in , , , , The percentage coefficients corresponding to HW-W / OI, LW-W / OI, HW-W / HI, LW-W / HI, HW-W / LI, and LW-W / LI are as follows: ； The hydrogen production for the predicted period is calculated as follows: ， in, The hydrogen production in hour j on day i. , , , , and These represent the average hydrogen production of each cluster (HW-W / OI, LW-W / OI, HW-W / HI, LW-W / HI, HW-W / LI, and LW-W / LI) during the historical days for the predicted hour.
2. The method for predicting hydrogen production from electrolysis in new energy sources based on centroradial clustering according to claim 1, characterized in that, The specific method for S1 is as follows: S11. Based on the 24-hour data for each of the past N days from the event date, plot box plots of wind intensity, light intensity, and hydrogen production. Then, identify outliers in the data according to Tukey's 1.5×IQR rule, i.e., calculate the first quartile Q1, median Q2, and third quartile Q3 to obtain the interquartile range. The upper bound of outliers is The lower boundary of outliers is Data points that are less than the lower bound or greater than the upper bound are considered outliers. S12. For the identified outliers, replace the outlier data for the corresponding hour with the average of the weather data for the two nearest normal hours: , , in, , and Correction data was provided to replace anomalies in wind intensity, light intensity, and hydrogen production. , , , , , These are adjacent normal data for wind intensity, light intensity, and hydrogen production, respectively, where subscripts i and j represent the j-th hour of the i-th day.