Typical operating condition generation method and device for PEM electrolysis stack in wind power hydrogen production system

Abstract

The present invention provides a method and device for generating typical operating conditions of a PEM electrolysis stack in a wind power hydrogen production system. A high-dimensional output characteristic quantity matrix is ​​formed by extracting characteristic quantities of a wind farm station output curve; the obtained high-dimensional output characteristic quantity matrix is ​​simplified by principal component analysis; the extracted principal components are clustered and evaluated by K-Means clustering, fuzzy C-means clustering, hierarchical agglomerative clustering, Gaussian mixture model clustering and clustering evaluation system; based on the proportion probability and curve characteristics of each typical output curve after clustering, a typical operating condition curve in a wind power electrolysis hydrogen production scenario is obtained by combination; the typical operating condition curve is simplified according to the power interval simplification rule, and the load change process between power interval segments is generated according to the weighted average load change rate of each typical curve, and a typical operating condition curve suitable for attenuation test of a PEM electrolysis stack in a wind power hydrogen production system is obtained by combination. The typical operating condition curve provided by the present invention is more conducive to the attenuation test of a laboratory PEM electrolysis stack.

Description

Technical Field

[0001] The present invention relates to the field of typical operating condition generation of hydrogen production by electrolysis, and in particular to a typical operating condition generation method and device of a PEM electrolytic stack in a wind power hydrogen production system. Background Art

[0002] In the power system, energy storage devices are usually used to smooth out fluctuations in the output of new energy, participate in peak load regulation, etc. As a green and pollution-free clean energy storage, proton exchange membrane (PEM) electrolysis hydrogen production has a very broad application prospect. Compared with alkaline water electrolysis hydrogen production, PEM electrolysis hydrogen production has the advantages of fast response speed and wide operating range. It is usually used in wind power and photovoltaic stations to smooth out fluctuations in the output of new energy. The hourly volatility and randomness of wind turbines are stronger than other renewable energy sources such as photovoltaics. The PEM electrolysis hydrogen production devices equipped in wind farms usually need to undertake more fluctuation smoothing work.

[0003] The operating conditions can reflect the operating characteristics of the PEM electrolyzer under high volatility inputs, and the effectiveness of the operating condition curve generation will directly affect the test results of the PEM electrolyzer durability. Common energy storage system operating condition generation usually uses heuristic scenario reduction methods, cluster analysis and other scenario reduction methods to reduce the output curve scenario, and construct the operating conditions based on the clustering results. The output curve clustering analysis can be divided into similarity measurement based on output curve morphology clustering, which requires smoothing and filtering of the original output curve; and similarity measurement based on data feature clustering, which extracts features from the original data and retains the processed data information to the greatest extent.

[0004] For the electrolytic stack attenuation test in the PEM hydrogen production system, it is necessary to extract a typical operating condition curve that is representative of the long time scale as the test input to ensure the effectiveness and reliability of the test. However, for the wind power PEM electrolysis hydrogen production system, on the one hand, due to the fast fluctuation response speed of PEM electrolysis hydrogen production, the hydrogen production system usually has a fluctuation response speed of minutes (min). At this time, the randomness of wind power output fluctuation is stronger, and a more accurate output curve fluctuation characterization index is required. On the other hand, the PEM electrolytic stack will affect the operating life under continuous low load, continuous overload and partial variable load conditions, and the output curves of different forms may have the same operating characteristics, which will also lead to errors in the generated operating conditions. At the same time, the operating curve used for the electrolytic stack attenuation test usually needs to be representative to shorten and simplify the test process. Therefore, the typical operating curve of the PEM electrolytic stack formed needs to be a periodic and cyclic test, which can reflect the output fluctuation of the site on a long time scale in terms of data characteristics, and can also reflect the main operating quantity characteristics required in the attenuation test. Summary of the invention

[0005] The purpose of the present invention is to address the above-mentioned problems existing in the field of typical working condition generation of electrolytic stacks in wind power hydrogen production systems, and to provide a method and device for generating working conditions of a PEM electrolytic stack in a wind power hydrogen production system. The technical solution is as follows:

[0006] A method for generating typical operating conditions of a PEM electrolysis stack in a wind power hydrogen production system, characterized in that the method comprises:

[0007] S1. extracting the characteristic quantity of the wind farm output curve to form a high-dimensional output characteristic quantity matrix;

[0008] S2. Use principal component analysis to reduce the dimension of the obtained high-dimensional output feature matrix;

[0009] S3. Cluster analysis and evaluation of the extracted principal components were performed using partition-based K-Means clustering, fuzzy C-means clustering, hierarchical agglomerative clustering, probability-based Gaussian mixture model clustering and clustering evaluation system;

[0010] S4. Based on the proportion probability and curve characteristics of each typical output curve after clustering, a typical operating condition curve under the wind power hydrogen production scenario is obtained by combination;

[0011] S5. Simplify the typical operating curve of the electrolytic stack in the wind power hydrogen production system according to the power interval simplification rule, and generate the load change process between the power interval segments according to the weighted average load change rate of each typical operating curve, and combine them to obtain the typical operating curve suitable for the PEM electrolytic stack attenuation test in the wind power hydrogen production system.

[0012] Preferably, the characteristic quantities of the wind farm output curve extracted in step S1 include the operating condition characteristic quantities and morphological characteristic quantities of the second-order difference mean of the downtime, low-load time, low-load average power, low-load power variance, overload time, overload average power, overload power variance, average climbing rate, average load reduction rate and output change rate, which together constitute the wind power output curve characteristic quantity matrix.

[0013] Preferably, the low-load time in the characteristic quantity is the time when the wind power PEM hydrogen production system operates at 5% to 30% of the rated operating power, and the downtime is the time when the wind power PEM hydrogen production system operates at less than 5% of the rated operating power; the calculation formulas for the average climbing rate and the average load reduction rate are as follows:

[0014]

[0015] In the formula, ΔP avg,i Represents the average ramp rate of the wind power output curve, ΔP avg,d represents the average load shedding rate of the wind power output curve, where the subscripts i and d represent the power ramp-up and power reduction processes, respectively. Respectively represent ti Time and t d Wind power output level at time t i ,t d Represent the power ramp-up time and power drop-down time respectively, N i Indicates the total amount of fluctuation in the power ramp phase, N d Indicates the total amount of fluctuation in the power reduction phase;

[0016] The calculation formula of the second-order difference mean of the output change rate is as follows:

[0017] Δ 2 P avg，k =ΔP avg，k+1 -ΔP avg，k (3)

[0018] In the formula, Δ 2 P avg,k It represents the second-order difference component of the average load rate of the output curve at time k, ΔP avg,k It represents the average climbing rate or load reduction rate at time k.

[0019] Preferably, each typical output curve of S4 is obtained by the following method:

[0020] Calculate the normalized distance between the characteristic quantity of each output curve in the cluster after cluster reduction and the characteristic quantity of the cluster center, and select the corresponding original output curve of the wind farm using the principle of minimum normalized distance sum of characteristic quantity. The calculation formula of minimum normalized distance sum of characteristic quantity is as follows:

[0021]

[0022] In the above formula, δ ij Indicates the deviation between the characteristic value of each curve in the cluster and the characteristic value of the cluster center, X ij represents the element in the i-th row and j-th column of the high-dimensional output feature matrix X, where i represents the i-th output curve, j represents the j-th output feature of the output curve, and m represents the total number of output feature quantities. represents the mean value of the j-th output characteristic in the k-th cluster; μ kj , σ kj Respectively represent the deviation value δ ij The sample mean and sample variance of the jth feature in the kth cluster, n k Represents the total number of curves in the kth cluster.

[0023] Preferably, the S4 typical operating condition curve is generated in the following manner:

[0024] Analyze the output proportion and output power level of each typical curve after clustering, and extract the main fluctuation segments of each typical curve;

[0025] The influence of typical curves with a weakening probability of less than 15% on the output level of the entire typical operating condition curve is analyzed by extracting only the fluctuation segments of such typical curves;

[0026] The typical curve with the highest probability after clustering is selected to form the main trend of the typical operating condition curve. If other typical curves other than the above curves have fluctuations at the same output level as the main trend, they are weighted combined according to the probability of each typical curve. The combination method is as follows:

[0027]

[0028] In the above formula, P f Indicates the fluctuating power synthesis part of the typical operating curve, K f represents the number of typical curves of weighted combination, K represents the total number of clusters, and w k represents the weight of the kth cluster, the sum of all cluster weights is 1, P kf represents the fluctuation segment of the kth typical output curve;

[0029] The output level characteristics of other typical curves except the fluctuation characteristics are weighted and superimposed to the corresponding output section according to the probability of each typical curve proportion, and the weighted combination method is the same as the above formula (6);

[0030] The typical operating condition curve obtained by the preliminary combination is adjusted by time series processing to make the output levels at the beginning and end of the operating condition curve consistent, and finally the typical operating condition curve under the wind power hydrogen production scenario is obtained.

[0031] Preferably, the power range simplification rule of the typical operating condition curve of the electrolytic stack in the wind power hydrogen production system described in S5 is: the load change process of the typical operating condition curve is reconstructed by the constant load change rate obtained by weighted combination, in the following manner:

[0032] Calculate the average rising rate and average falling rate of the output power of each typical curve after clustering;

[0033] The average rising rate and falling rate of each typical curve are combined according to probability weighting to obtain the fixed rising rate and fixed falling rate of the output power of the entire typical working condition curve;

[0034] According to the step power segments defined by the power interval simplification rule, the load variation process between adjacent power segments is reconstructed;

[0035] The load variation processes with overlapping time sequences are merged and the upper envelope of the overlapping load variation processes is retained. The envelope is a typical operating condition curve segment, as shown in the following formula:

[0036]

[0037] Where sup represents the supremum, represents a set of real numbers, P1(t) and P2(t) represent the power curves of two different load-changing processes, and t represents time.

[0038] The present invention also discloses a typical operating condition generating device for a PEM electrolytic stack in a wind power hydrogen production system, which is characterized by:

[0039] High-dimensional output characteristic quantity matrix generation module: obtained by extracting the characteristic quantity of the wind farm output curve;

[0040] High-dimensional output feature matrix dimensionality reduction and simplification module: Use principal component analysis to reduce the dimension of the high-dimensional output feature matrix;

[0041] Cluster analysis and evaluation module: Use partition-based K-Means clustering, fuzzy C-means clustering, hierarchical agglomerative clustering, probability-based Gaussian mixture model clustering and cluster evaluation system to perform cluster analysis and evaluation on the extracted principal components;

[0042] The typical operating curve module under the wind power hydrogen production scenario is obtained by combination: based on the proportion probability and curve characteristics of each typical output curve after clustering, the typical operating curve under the wind power electrolysis hydrogen production scenario is obtained by combination;

[0043] The typical operating curve module suitable for attenuation test of PEM electrolyzer in wind power hydrogen production system is obtained by combination: the typical operating curve is simplified according to the power interval simplification rule, and the load variation process between power interval segments is generated according to the weighted average load variation rate of each typical curve, and the typical operating curve suitable for attenuation test of PEM electrolyzer in wind power hydrogen production system is obtained by combination.

[0044] The present invention also discloses a non-volatile storage medium, characterized in that the non-volatile storage medium includes a stored program, wherein when the program is running, the device where the non-volatile storage medium is located is controlled to execute the above method.

[0045] The present invention also discloses an electronic device, characterized in that it comprises a processor and a memory; the memory stores computer-readable instructions, and the processor is used to execute the computer-readable instructions, wherein the computer-readable instructions execute the above method when they are executed.

[0046] Beneficial Effects

[0047] Based on the operating characteristics of the wind power electrolysis hydrogen production system, the present invention obtains a 365×10 high-dimensional feature matrix by extracting the morphological characteristics and operating condition characteristics of the wind power output curve for analysis. Each feature in the matrix can reflect the fluctuation of the wind power output curve and the operating condition characteristics of the PEM electrolytic stack, that is, the attenuation problem of the PEM electrolytic stack caused by low overload operation, which can better meet the attenuation test requirements of the PEM electrolytic stack.

[0048] In addition, a typical operating curve generation method based on the proportion probability and curve characteristics of each typical output curve after clustering is proposed. The typical output curves of each cluster after clustering are selected by the minimum normalized distance of the characteristic quantity. This curve can maximize the characterization of the morphology and operating characteristic quantity characteristics of the original cluster cluster, which is more conducive to synthesizing the required typical operating curve of the PEM electrolytic stack; the typical curve with the highest proportion probability after clustering is selected to form the main trend of the typical operating curve, and the other typical curves are weightedly combined with the fluctuation fragments and output level characteristics according to the proportion probability of each typical curve. The output characteristics and fluctuation characteristics of each cluster curve are reflected through a synthetic operating curve, which simplifies the complexity of the operating curve used for the PEM electrolytic stack attenuation test. Further simplification is made on this basis. The operating curve is divided into power steps through the defined power interval simplification rule, and then the electrolytic stack attenuation test is carried out according to each power step. In addition, the load variation process of the typical operating condition curve was reconstructed, and the typical operating condition curve of the PEM electrolyzer at a fixed rising rate and a fixed falling rate was obtained, which simplified the typical operating condition load variation test process and was more conducive to eliminating the possible error that different load variation rates lead to huge differences in attenuation test results. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A schematic diagram of a process flow diagram of a typical operating condition of a PEM electrolysis stack in a wind power hydrogen production system provided by an embodiment of the present invention;

[0050] Figure 2 A schematic diagram of a typical wind farm output curve after clustering provided by an embodiment of the present invention;

[0051] Figure 3 A schematic diagram of a simplified typical operating curve of a PEM electrolysis stack in a wind power hydrogen production system provided in an embodiment of the present invention;

[0052] Figure 4 A schematic diagram of a typical operating curve of a PEM electrolysis stack in a wind power hydrogen production system provided in an embodiment of the present invention after a load variation process is added. DETAILED DESCRIPTION

[0053] In order to facilitate those skilled in the art to understand and implement the present invention, the present invention is further described in detail below in conjunction with embodiments. It should be understood that the implementation examples described herein are only used to illustrate and explain the present invention and are not used to limit the present invention.

[0054] A method for generating typical operating conditions of a PEM electrolysis stack in a wind power hydrogen production system comprises the following steps:

[0055] Step 1: Extract the working condition characteristics that meet the test requirements of the electrolyzer, taking into account the morphological characteristics of the wind power output curve, and select the downtime, low-load time, low-load average power, low-load power variance, overload time, overload average power, overload power variance, average climbing rate, average load reduction rate and output change rate second-order difference mean to form a high-dimensional output characteristic matrix. The working condition related quantities are defined as follows:

[0056]

[0057] The morphological features are defined as follows:

[0058]

[0059] Δ 2 P avg，k =ΔP avg，k+1 -ΔP avg，k (3)

[0060] Where ΔP avg,i Represents the average ramp rate of the wind power output curve, ΔP avg,d It represents the average load shedding rate of the wind power output curve, where the subscripts i and d represent the power ramp-up and power reduction processes, respectively. Respectively represent t i Time and t d Wind power output level at time t i ,t d Represent the power ramp-up time and power drop-down time respectively, N i Indicates the total amount of fluctuation in the power ramp phase, N d Indicates the total amount of fluctuation in the power reduction stage; Δ 2 P avg,k It represents the second-order difference component of the average load rate at the output curve at time k, which is used to reflect the stability of the load rate change between different fluctuation processes, ΔP avg,k It represents the average climbing rate or load reduction rate at time k.

[0061] Step 2: Use principal component analysis (PCA) to reduce the dimension of the high-dimensional output feature matrix obtained in the above steps, highlight the characteristic factors that have a greater impact on the feature clustering effect, reduce the complexity of data calculation, and shorten the analysis time. The specific steps of PCA are as follows:

[0062] 1) Data standardization: The wind power high-dimensional output characteristic matrix is ​​standardized by mean variance to obtain the standardized matrix X;

[0063] 2) Calculate the covariance matrix R of the standardized sample matrix;

[0064]

[0065] In the above formula, r ij is the element in the i-th row and j-th column of the covariance matrix R, X ki is the element in the kth row and ith column of the standardized matrix X, Respectively represent the sample means of the i-th column and the j-th column in the matrix X;

[0066] 3) Calculate the eigenvalues ​​and eigenvectors of the covariance matrix R, where the eigenvalues ​​are λ1 to λ p , the corresponding eigenvectors are k1 to k p ;

[0067] 4) Calculate the variance contribution rate and cumulative variance contribution rate: The cumulative variance contribution rate of the i-th principal component α i The calculation formula is as follows:

[0068]

[0069] The variance contribution rate indicates the ability of the principal component to describe the original data. The principal components with a cumulative variance contribution rate exceeding 90% are selected for clustering and reduction of the output curves. The reduced principal components can not only reflect the main characteristics of the original high-dimensional data, but also reduce the amount of data calculation.

[0070] Step 3: Use partition-based K-Means clustering, fuzzy C-means clustering, hierarchical agglomerative clustering, and probability-based Gaussian mixture model (GMM) clustering to perform cluster reduction analysis on the extracted principal components. The implementation process of each clustering method is as follows:

[0071] 1. Partition-based K-Means clustering and fuzzy C-means clustering

[0072] K-Means clustering uses Euclidean distance to calculate the distance between sample points and the centroids of various clusters, and the classification of each sample point is determined by the minimum distance; fuzzy C clustering divides the samples into categories by calculating their membership to different classes. The objective functions of the two algorithms are shown in the following formulas:

[0073]

[0074] In the above formula, μ ik Represents the i-th sample point x i For cluster center c k The membership of each sample belongs to all classes, and the sum of the membership of each sample to all classes is 1.

[0075] 2. Hierarchical Agglomerative Clustering Based on Level

[0076] In agglomerative clustering, each sample point is grouped as a cluster, and the two samples with the highest similarity are merged into the same cluster by calculating the inter-cluster distance. The inter-cluster distance calculation method can be divided into the minimum distance Dmin, the maximum distance Dmax, the average distance Davg and the centroid distance Dcen.

[0077] 3. Probability-based Gaussian Mixture Model Clustering

[0078] By linearly combining sub-Gaussian distributions with different morphological characteristics, it can be used to fit the probability distribution of any shape. The iterative process is as follows:

[0079] 1) Initialization parameters: set the number of sub-Gaussian distributions K and the initial values ​​of each sub-Gaussian distribution, including weight, expectation, and variance ω k , μ k , σ k , and calculate the log-likelihood function value of the Gaussian mixture model as follows:

[0080]

[0081] In the above formula, N k (x|μ k ,σ k ) indicates that the mean is μ k , the variance is σ k Gaussian distribution function of

[0082] 2) E-step: According to the weight, expectation, and variance ω of each sub-Gaussian distribution k , μ k , σ k The initial value of γ is used to calculate the posterior probability k ;

[0083] 3) M-step: Iteratively solve ω based on the posterior probability calculated in the E-stepk , μ k , σ k new value;

[0084] 4) Convergence condition: Calculate the likelihood function value after parameter iteration according to the formula shown in 1) until the difference before and after iteration is not higher than the set convergence difference, and terminate the iteration; otherwise, return to step 2.

[0085] The clustering effect of the above clustering algorithms is evaluated by the inter-cluster separation and intra-cluster cohesion, and the optimal number of clusters is selected. The clustering evaluation indicators are Calinski-Harabaz index (CHI), Davies-Bouldin Index (DBI) and Silhouette Coefficient (SC), where the larger the CHI value, the better the clustering effect, the smaller the DBI value, the better the clustering effect, and the closer the silhouette coefficient is to 1, the better the clustering effect. Calculate the normalized distance between the characteristic quantity of each output curve in the cluster after cluster reduction and the characteristic quantity of the cluster center, and select the corresponding original output curve of the wind farm based on the principle of the minimum normalized distance sum of the characteristic quantity to characterize the typical input curve characteristics of the cluster. The calculation formula of the minimum normalized distance sum of the characteristic quantity is as follows:

[0086]

[0087] In the above formula, δ ij Indicates the deviation between the characteristic value of each curve in the cluster and the characteristic value of the cluster center, X ij represents the element in the i-th row and j-th column of the high-dimensional output feature matrix X, where i represents the i-th output curve, j represents the j-th output feature of the output curve, and m represents the total number of output feature quantities. represents the mean value of the j-th output characteristic in the k-th cluster; μ kj , σ kj Respectively represent the deviation value δ ij The sample mean and sample variance of the jth feature in the kth cluster, n k Represents the total number of curves in the kth cluster;

[0088] Step 4: According to the proportion probability of each typical output curve and the output power level, the fluctuation characteristic segments of each typical curve after clustering are extracted through peak / trough recognition, and the typical curve with the highest proportion probability after clustering is selected to form the main trend of the typical working condition curve. If other typical curves other than the above curves have fluctuations at the same output level as the main trend, they are weighted combined according to the proportion probability of each typical curve. The combination method is as follows:

[0089]

[0090] In the above formula, P f Indicates the fluctuating power synthesis part of the typical operating curve, K f represents the number of typical curves of weighted combination, K represents the total number of clusters, and w k represents the weight of the kth cluster, the sum of all cluster weights is 1, P kf represents the fluctuation segment of the kth typical output curve;

[0091] The output level characteristics of other typical curves except the fluctuation characteristics are weighted and superimposed to the corresponding output section according to the probability of the output level of each typical curve, and the weighted combination method is the same as the above formula. At the same time, the typical operating condition curve usually needs to be tested cyclically. For the typical operating condition curve under the minimum cycle period (1440min), the output level of the head and the end needs to be consistent. Therefore, it is necessary to perform time series processing on the typical operating condition curve of the above combination so that the output level of the head and the end is the same, and finally obtain the typical operating condition curve under the wind power hydrogen production scenario.

[0092] Step 5: To facilitate the attenuation test of the electrolytic stack, the typical operating curve of the electrolytic stack in the wind power hydrogen production system is simplified according to the power range simplification rule in combination with the original output level of the wind farm. The power range is divided into 8 levels in the simplified rule, as shown below.

[0093]

[0094] The simplified typical operating curve is in a step shape, and the load change rate between different step power segments needs to be set to achieve smooth load change. The entire typical operating curve uses a fixed rising rate and a fixed falling rate for load change. The fixed rate is obtained by weighting the average load change rate of each typical curve according to probability, reflecting the average load change rate level of the wind farm output curve. The calculation process of the average load change rate ΔP is as follows:

[0095]

[0096] In the above formula, ρk represents the probability of the kth typical curve, and the sum of the probabilities of all typical curves is 1. k Represents the average ramp-up / down rate of the kth typical curve.

[0097] The load variation process between each step power is formed according to the obtained fixed rising rate and fixed falling rate, and the minimum cycle period is maintained unchanged. The load variation processes with overlapping time sequences are merged, and the upper envelope of the overlapping load variation processes is retained. The envelope is a typical operating condition curve segment, as shown in the following formula:

[0098]

[0099] Where sup represents the supremum, represents a set of real numbers, P1(t) and P2(t) represent the power curves of two different load-changing processes, and t represents time.

[0100] Example:

[0101] A method for generating operating conditions of a PEM electrolysis stack in a wind power hydrogen production system, as shown in the attached Figure 1 The structure shown comprises the following steps:

[0102] Step 1: Extract the operating condition characteristics that meet the electrolyzer test requirements, while taking into account the morphological characteristics of the wind power output curve, and select the downtime, low-load time, low-load average power, low-load power variance, overload time, overload average power, overload power variance, average climbing rate, average load reduction rate and second-order difference mean of output change rate to form a high-dimensional output characteristic matrix.

[0103] In the embodiment of the present invention, the calculation example data uses the output data of a wind farm in a certain area with a sampling rate of 1 minute for 365 days throughout the year. The feature quantity matrix to be extracted is a high-dimensional matrix of 365×10. The above-mentioned operating condition characteristics and morphological characteristic indicators of the wind farm output curve within each day correspond to each row vector of the feature quantity matrix. Each column vector of the matrix is ​​the corresponding indicator value of a specific feature quantity throughout the year, so as to reflect the output level and fluctuation of the wind farm throughout the year. The purpose of extracting the above feature quantity indicators is to obtain more accurate output curve characteristics, and at the same time, it can characterize the impact of the output curve on the operating conditions of the electrolytic stack in the wind power hydrogen production system, which is more consistent with the data required for the electrolytic stack attenuation test.

[0104] Step 2: PCA is used to reduce the dimension of the high-dimensional output feature matrix obtained in the above steps.

[0105] By performing principal component analysis on the original high-dimensional output feature matrix, the variance contribution rate of each feature indicator to the whole can be obtained. The principal component refers to several unrelated variables that can reflect the main characteristics of the original data. The variance contribution rate refers to the proportion of the difference of the principal component to the total difference of the original data. The greater the contribution rate, the stronger the explanatory power of the principal component for the total difference of the original feature. Several principal components with the largest variance contribution rates are selected to represent the original output feature matrix, and the high-dimensional feature is simplified by dimensionality reduction. Through calculation, the cumulative variance contribution rate of the first five principal components in this embodiment exceeds 90% and reaches 90.48%, which can be used to represent the data characteristics of the original output feature matrix.

[0106] Step 3: Cluster analysis and evaluation of the extracted principal components are performed using partition-based K-Means clustering, fuzzy C-means clustering, hierarchical agglomerative clustering, probability-based Gaussian mixture model clustering, and clustering evaluation system.

[0107] The three types of evaluation indicators, CHI, DBI and SC, are used to form a clustering evaluation system. The initial cluster center of each clustering method adopts the multiple random center method, and the minimum distance from the sample point to the cluster center is taken as the initial cluster center. As for the selection of the number of clusters, hierarchical agglomerative clustering does not need to give the number of clusters in advance; GMM clustering usually uses the Bayesian Information Criterion (BIC) to determine the optimal number of clusters, and selects the number of clusters with the minimum BIC value for cluster analysis. Other clustering methods use the clustering effectiveness evaluation system to select the optimal number of clusters. The above four types of clustering methods are used to perform cluster analysis on the simplified principal components, and the evaluation indicators of each clustering method are compared to obtain the optimal output curve clustering results. Among them, the CHI, DBI, and SC indicators of the fuzzy C-means clustering results are better than those of other clustering methods. Therefore, the clustering results of the fuzzy C-means can be used as the cluster results of the output curve. The clustering results divide the wind farm output curves into four categories, of which cluster 1 has 34 days, accounting for 9.3%; cluster 2 has 150 days, accounting for 41.1%; cluster 3 has 130 days, accounting for 35.6%; cluster 4 has 51 days, accounting for 14.0%. The original output curve feature values ​​are normalized, and the normalized distance sum between the output curve feature value in each cluster and the feature value of the cluster center is calculated. The output curve with the smallest normalized distance of the feature value is taken as the typical output curve of the cluster. The typical curve results of each cluster are shown in the attached figure. Figure 2 As shown in the figure, the four typical curves respectively represent the typical input curve characteristics of the above four types of clusters.

[0108] Step 4: Based on the proportion probability and curve characteristics of each typical output curve after clustering, a typical operating condition curve in the wind power electrolysis hydrogen production scenario is obtained by combination.

[0109] In order to facilitate the attenuation test of the electrolytic stack, it is necessary to combine the four types of typical output curves to synthesize a typical operating curve that meets the characteristics and proportion probability of each type of typical curve. Among the four typical curves obtained by the above steps, typical curves 2 and 3 account for 76.7% of the total number of output curves, which are the main curve characteristics of the combined operating curve. In addition, since the total proportion of typical curves 1 and 4 is relatively low, the influence of these two types of curves on the output power level of the entire typical operating condition is weakened when synthesizing the typical operating curve. Only the influence of typical curves 1 and 4 on the fluctuation characteristics of the operating curve is discussed. The output range of typical curve 2 covers operating conditions such as low load, overload, and shutdown. Typical curve 3 is mainly the fluctuation process under the low load state of the electrolytic stack. The main fluctuation feature fragments of each typical curve are extracted, and typical curves 2 and 3 are combined by probability weighting to obtain the main characteristics of the typical operating curve. Then, the fluctuation process of typical curves 1 and 4 at the corresponding output level is combined by probability weighting to obtain the typical operating curve in the wind power electrolysis hydrogen production scenario.

[0110] Further considering the consistency of the power at the beginning and end of the electrolytic stack attenuation test under a minimum cycle period (1440 min), the above typical operating condition curve is adjusted by translation so that the power at the beginning and end of the typical operating condition curve remains consistent, which is convenient for the electrolytic stack to perform cyclic power testing.

[0111] Step 5, simplify the typical operating condition curve of the electrolytic stack in the wind power hydrogen production system according to the power interval simplification rule, and generate the load change process between the power interval segments according to the weighted average load change rate of each typical operating condition curve, and combine them to obtain the typical operating condition curve suitable for the PEM electrolytic stack attenuation test in the wind power hydrogen production system.

[0112] The present invention simplifies the typical operating condition curve obtained above to obtain a typical operating condition curve simplified according to the power range simplification rule, as shown in the attached figure. Figure 3 As shown in the figure, the power levels where the dotted lines are located represent 100% rated power, 30% rated power, and 5% rated power, respectively, corresponding to the definition of the operating condition characteristic quantity extracted in step 1. At this time, the load change rate between each output step level of the operating condition curve is infinite, which does not meet the actual test requirements, and the operating condition curve still needs to be processed with load changes between each output power segment.

[0113] The present invention unifies the output power rising rate and falling rate of the entire typical operating condition curve to achieve a unified load change in the electrolytic stack test process and simplify the load change operation. The load change rate is obtained by weighting the average load change rate of each typical curve obtained by cluster analysis, and the weight is the curve share probability. After obtaining the fixed rising rate and falling rate, the corresponding load change process between power segments can be generated. At the same time, in order to keep the minimum cycle period unchanged at 1440min, it is necessary to combine the time-series overlapping load change processes and retain the upper envelope of the overlapping load change processes. The upper envelope forms a triangular wave-shaped peak-valley load change process. After the fixed load change process is combined, the typical operating condition of the PEM electrolytic stack in the wind power hydrogen production system can be obtained, as shown in the attached figure. Figure 4 As shown, the power level corresponding to the dotted line in the figure is the same as that in the attached figure. Figure 3 The typical operating curve of the PEM electrolytic stack shown in the figure finally generated will facilitate the analysis and study of the attenuation characteristics of the electrolytic stack in the laboratory.

Claims

1. A method for generating typical operating conditions of a PEM electrolysis stack in a wind power hydrogen production system, characterized in that: The method comprises: S1. Extract the characteristic quantity of the wind farm output curve to form a high-dimensional output characteristic quantity matrix; S2. Use principal component analysis to reduce the dimension of the obtained high-dimensional output feature matrix; S3, cluster analysis and evaluation of the extracted principal components were performed using partition-based K-Means clustering, fuzzy C-means clustering, hierarchical agglomerative clustering, probability-based Gaussian mixture model clustering and clustering evaluation system; S4. Based on the proportion probability and curve characteristics of each typical output curve after clustering, the typical operating condition curve under the wind power hydrogen production scenario is combined: Analyze the output proportion and output power level of each typical curve after clustering, and extract the main fluctuation segments of each typical curve; The influence of typical curves with a weakening probability of less than 15% on the output level of the entire typical operating condition curve is analyzed by extracting only the fluctuation segments of such typical curves; The typical curve with the highest probability after clustering is selected to form the main trend of the typical operating condition curve. If other typical curves other than the above curves have fluctuations at the same output level as the main trend, they are weighted combined according to the probability of each typical curve; The output level characteristics of other typical curves except the fluctuation characteristics are weighted and superimposed to the corresponding output section according to the probability according to the proportion of each typical curve; The typical operating curve obtained by preliminary combination is adjusted by time series processing to make the output levels at the beginning and end of the operating curve consistent, and finally the typical operating curve under the wind power hydrogen production scenario is obtained; S5. Simplify the typical operating curve of the electrolytic stack in the wind power hydrogen production system according to the power interval simplification rule, and generate the load change process between the power interval segments according to the weighted average load change rate of the typical operating curve, and combine to obtain a typical operating curve suitable for the PEM electrolytic stack attenuation test in the wind power hydrogen production system; wherein the power interval simplification rule of the typical operating curve of the electrolytic stack in the wind power hydrogen production system is: The load variation process of the typical operating condition curve is reconstructed by using the constant load variation rate obtained by weighted combination, and the following method is adopted: Calculate the average rising rate and average falling rate of the output power of each typical curve after clustering; The average rising rate and falling rate of each typical curve are combined according to probability weighting to obtain the fixed rising rate and fixed falling rate of the output power of the entire typical working condition curve; According to the step power segments defined by the power interval simplification rule, the load variation process between adjacent power segments is reconstructed; The load variation processes with overlapping time sequences are merged and the upper envelope of the overlapping load variation processes is retained. The envelope is a typical operating condition curve segment.

2. The method according to claim 1, characterized in that The characteristic quantities of the wind farm output curve extracted in step S1 include the operating condition characteristic quantities and morphological characteristic quantities of the second-order difference mean of the downtime, low-load time, low-load average power, low-load power variance, overload time, overload average power, overload power variance, average climbing rate, average load reduction rate and output change rate, which together constitute a high-dimensional wind power output characteristic quantity matrix.

3. The method according to claim 2, characterized in that The low-load time in the characteristic quantity is the time when the wind power PEM hydrogen production system operates at 5% to 30% of the rated operating power, and the downtime is the time when the wind power PEM hydrogen production system operates at less than 5% of the rated operating power; the calculation formulas for the average climbing rate and the average load reduction rate are as follows: In the formula, ΔP avg,i Represents the average ramp rate of the wind power output curve, ΔP avg,d represents the average load reduction rate of the wind power output curve, where the subscripts i and d represent the power ramp-up and power reduction processes, respectively. t represents the wind power output level at time t, N i Indicates the total amount of fluctuation in the power ramp phase, N d Indicates the total amount of fluctuation in the power reduction phase; The calculation formula of the second-order difference mean of the output change rate is as follows: D 2 P avg，k =ΔP avg，k+1 -ΔP avg，k (3) In the formula, Δ 2 P avg,k It represents the second-order difference component of the average load rate of the output curve at time k, ΔP avg,k It represents the average climbing rate or load reduction rate at time k.

4. The method according to claim 1, characterized in that: The typical output curves of S4 are obtained in the following way: Calculate the normalized distance between the characteristic quantity of each output curve in the cluster after cluster reduction and the characteristic quantity of the cluster center, and select the corresponding original output curve of the wind farm using the principle of minimum normalized distance sum of characteristic quantity. The calculation formula of minimum normalized distance sum of characteristic quantity is as follows: In the above formula, δ ij Indicates the deviation between the characteristic value of each curve in the cluster and the characteristic value of the cluster center, X ij represents the element in the i-th row and j-th column of the high-dimensional output feature matrix X, where i represents the i-th output curve, j represents the j-th output feature of the output curve, represents the mean value of the j-th output characteristic in the k-th cluster; μ kj , σ kj Respectively represent the deviation value δ ij The sample mean and sample variance of the jth feature in the kth cluster, n k Represents the total number of curves in the kth cluster.

5. The method according to claim 1, characterized in that The weighted combination method in S4 according to the probability of each typical curve is as follows: In the above formula, P f Indicates the fluctuating power synthesis part of the typical operating curve, K f represents the number of typical curves of weighted combination, K represents the total number of clusters, and w k represents the weight of the kth cluster, the sum of all cluster weights is 1, P kf Represents the fluctuation segment of the kth typical output curve.

6. The method according to claim 1, characterized in that In step 5, the load variation processes with overlapping timing are merged, and the process of retaining the upper envelope of the overlapping load variation processes is shown in the following formula: Where sup represents the supremum, represents a set of real numbers, P1(t) and P2(t) represent the power curves of two different load-changing processes, and t represents time.

7. A typical operating condition generation device for a PEM electrolysis stack in a wind power hydrogen production system, characterized by: High-dimensional output characteristic quantity matrix generation module: obtained by extracting the characteristic quantity of the wind farm output curve; High-dimensional output feature matrix dimensionality reduction and simplification module: Use principal component analysis to reduce the dimension of the high-dimensional output feature matrix; Cluster analysis and evaluation module: Use partition-based K-Means clustering, fuzzy C-means clustering, hierarchical agglomerative clustering, probability-based Gaussian mixture model clustering and cluster evaluation system to perform cluster analysis and evaluation on the extracted principal components; The typical operating curve module under the wind power hydrogen production scenario is obtained by combination: According to the proportion probability and curve characteristics of each typical output curve after clustering, the typical operating curve under the wind power electrolysis hydrogen production scenario is obtained by combination: Analyze the output proportion and output power level of each typical curve after clustering, and extract the main fluctuation segments of each typical curve; The influence of typical curves with a weakening probability of less than 15% on the output level of the entire typical operating condition curve is analyzed by extracting only the fluctuation segments of such typical curves; The typical curve with the highest probability after clustering is selected to form the main trend of the typical operating condition curve. If other typical curves other than the above curves have fluctuations at the same output level as the main trend, they are weighted combined according to the probability of each typical curve; The output level characteristics of other typical curves except the fluctuation characteristics are weighted and superimposed to the corresponding output section according to the probability according to the proportion of each typical curve; The typical operating curve obtained by preliminary combination is adjusted by time series processing to make the output levels at the beginning and end of the operating curve consistent, and finally the typical operating curve under the wind power hydrogen production scenario is obtained; A typical operating curve module suitable for attenuation test of PEM electrolysis stack in wind power hydrogen production system is obtained by combination: the typical operating curve is simplified according to the power interval simplification rule, and the load change process between power interval segments is generated according to the weighted average load change rate of each typical curve, and the typical operating curve suitable for attenuation test of PEM electrolysis stack in wind power hydrogen production system is obtained by combination; wherein the power interval simplification rule of the typical operating curve of the electrolysis stack in the wind power hydrogen production system is: The load variation process of the typical operating condition curve is reconstructed by using the constant load variation rate obtained by weighted combination, and the following method is adopted: Calculate the average rising rate and average falling rate of the output power of each typical curve after clustering; The average rising rate and falling rate of each typical curve are combined according to probability weighting to obtain the fixed rising rate and fixed falling rate of the output power of the entire typical working condition curve; According to the step power segments defined by the power interval simplification rule, the load variation process between adjacent power segments is reconstructed; The load variation processes with overlapping time sequences are merged and the upper envelope of the overlapping load variation processes is retained. The envelope is a typical operating condition curve segment.

8. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored program, wherein when the program is executed, the device where the non-volatile storage medium is located is controlled to execute the method according to any one of claims 1 to 6.

9. An electronic device, characterized in that: It comprises a processor and a memory; the memory stores computer-readable instructions, and the processor is used to execute the computer-readable instructions, wherein the computer-readable instructions execute the method described in any one of claims 1 to 6 when executed.

Images