Multi-stack electrolytic cell power distribution method and system based on wind power and photovoltaic prediction

Abstract

The invention provides a multi-stack electrolytic cell power distribution method and system based on wind power and photovoltaic prediction, and is applied to the technical field of energy and control. The method comprises the following steps: performing wind power prediction based on historical wind power generation data and a wind power prediction model to obtain predicted wind power; the wind power prediction model is a multivariable graph neural network based on an inverse residual structure; performing photovoltaic power prediction based on the historical photovoltaic power generation data and the photovoltaic power prediction model to obtain predicted photovoltaic power; based on the predicted wind power and the predicted photovoltaic power, the reference number of running electrolytic cells is determined; based on the predicted wind power and the predicted photovoltaic power, determining the total hydrogen production power by using a power optimization model considering the system cost; and power distribution is carried out based on the electrolytic cell operation state, the reference number and the total hydrogen production power. According to the invention, the problems of low service life of the electrolytic cell and low hydrogen production efficiency caused by frequent start and stop of the electrolytic cell due to fluctuation of wind power output of a wind-hydrogen system are solved.

Description

Technical Field

[0001] The present invention relates to the field of energy and control technology, and in particular to a multi-stack electrolyzer power distribution method and system based on wind power and photovoltaic prediction. Background Art

[0002] In recent years, renewable energy sources, such as wind and photovoltaic power generation, have experienced rapid growth, providing up to 80% of global electricity demand and contributing to global energy security and sustainable development. Renewable energy sources such as wind and solar have seen rapid global growth due to their unique advantages, and are recognized by the International Energy Agency as highly competitive, environmentally friendly, and energy-friendly.

[0003] Hydrogen is a clean energy source that produces no polluting gases during combustion. Currently, hydrogen is produced using two methods: fossil fuels and water electrolysis. The former generates large amounts of carbon emissions, which is detrimental to environmental protection and energy transition. The latter, on the other hand, produces no carbon emissions during production and is in line with the concept of green development. However, wind and solar power output is characterized by randomness, intermittency, and volatility. When connected to the grid for frequency regulation, wind and solar power may be abandoned, resulting in energy waste. When a wind-hydrogen system produces hydrogen, the volatility of wind power output can cause the electrolyzer to frequently start and stop, reducing its service life and hydrogen production efficiency. Summary of the Invention

[0004] In order to overcome the problems of low electrolyzer service life and hydrogen production efficiency caused by frequent start and stop of electrolyzers due to the volatility of wind power output in the above-mentioned wind hydrogen system, the present invention provides a multi-stack electrolyzer power allocation method and system based on wind power and photovoltaic prediction.

[0005] In one aspect, the present invention provides a method for allocating power to multiple electrolyzers based on wind power and photovoltaic power forecasts, the method comprising:

[0006] Perform wind power forecasting based on historical wind power generation data and a wind power forecasting model to obtain the forecasted wind power within the forecast period; the wind power forecasting model is a multivariate graph neural network based on an inverse residual structure;

[0007] Performing photovoltaic power forecasting based on historical photovoltaic power generation data and a photovoltaic power forecasting model to obtain predicted photovoltaic power within the forecast period;

[0008] determining a reference number of operating electrolyzers within the prediction period based on the predicted wind power and the predicted photovoltaic power;

[0009] Based on the predicted wind power and the predicted photovoltaic power, determining the total hydrogen production power using a power optimization model that takes system cost into consideration;

[0010] Power distribution of the multiple stacks of electrolyzers is performed based on the operating status of each electrolyzer in the multiple stacks of electrolyzers, the reference number and the total hydrogen production power.

[0011] On the other hand, the present invention also provides a multi-stack electrolyzer power distribution system based on wind power and photovoltaic prediction, comprising:

[0012] A wind power prediction unit is used to predict wind power based on historical wind power generation data and a wind power prediction model to obtain predicted wind power within a prediction period; the wind power prediction model is a multivariate graph neural network based on an inverse residual structure;

[0013] A photovoltaic prediction unit, configured to perform photovoltaic power prediction based on historical photovoltaic power generation data and a photovoltaic power prediction model, and obtain predicted photovoltaic power within the prediction period;

[0014] an electrolyzer quantity determination unit, configured to determine a reference quantity of operating electrolyzers within the prediction period based on the predicted wind power and the predicted photovoltaic power;

[0015] a total power determination unit, which determines the total hydrogen production power based on the predicted wind power and the predicted photovoltaic power using a power optimization model that takes system cost into consideration;

[0016] A power distribution unit is used to distribute the power of the multiple electrolyzer stacks based on the operating status of each electrolyzer in the multiple electrolyzer stacks, the reference number and the total hydrogen production power.

[0017] On the other hand, the present invention also provides an electronic device, comprising: at least one processor and a memory; the memory and the processor are connected via a bus;

[0018] The memory is used to store one or more programs;

[0019] When the one or more programs are executed by the at least one processor, any one of the above-mentioned methods for allocating power of multiple electrolyzers based on wind power and photovoltaic prediction is implemented.

[0020] On the other hand, the present invention also provides a readable storage medium having an execution program stored thereon. When the execution program is executed, any one of the above-mentioned methods for allocating power of multiple electrolyzers based on wind power and photovoltaic prediction is implemented.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] The present invention provides a multi-stack electrolyzer power allocation method based on wind power and photovoltaic prediction. Aiming at the operating characteristics of wind power and photovoltaic power generation, the wind power and photovoltaic power of the future time period are obtained through wind power prediction and photovoltaic power prediction based on the inverse residual results of a multivariate graph neural network, and the number of operating electrolyzers is planned based on the prediction results. The electrolyzer power allocation of the total hydrogen production power is performed considering the system cost and the health status of the electrolyzer. The method can realize the pre-matching calculation of the electrolyzer hydrogen production power based on the accurate prediction of wind and solar power output, and avoid the frequent start and stop of the electrolyzer due to the real-time wind power output fluctuation. By allocating power considering the cost and the operating status of the electrolyzer, the rational use of the electrolyzer is realized, the service life of the multi-stack electrolyzer is extended, and the hydrogen production efficiency is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a flow chart of a multi-stack electrolyzer power allocation method based on wind power and photovoltaic prediction according to the present invention;

[0024] Figure 2 Schematic diagram of the wind power prediction process of the multivariate graphical neural network based on the inverse residual results of the present invention;

[0025] Figure 3 Schematic diagram of the processing process of the time convolution submodule of the present invention;

[0026] Figure 4 Schematic diagram of the dilated convolution processing process of the temporal convolution submodule of the present invention;

[0027] Figure 5 Schematic diagram of the processing process of the graph convolution submodule of the present invention;

[0028] Figure 6 Schematic diagram of the processing process of the Mix-hop propagation layer of the present invention;

[0029] Figure 7 Schematic diagram of the processing process of the inverted residual submodule of the present invention;

[0030] Figure 8 Schematic diagram of the structure and processing process of the long-term and short-term time series network of the present invention;

[0031] Figure 9 Schematic diagram of the structure of the recurrent layer of the long-term and short-term time series network of the present invention;

[0032] Figure 10 This is a schematic diagram of power distribution results of multiple electrolytic cells according to an embodiment of the present invention;

[0033] Figure 11 This is a system structure diagram of the wind power-photovoltaic-multi-stack electrolyzer system of the present invention;

[0034] Figure 12 This is a second flow chart of the multi-stack electrolyzer power allocation method based on wind power and photovoltaic prediction according to the present invention;

[0035] Figure 13 This is a schematic structural diagram of a multi-stack electrolyzer power distribution system based on wind power and photovoltaic prediction according to the present invention;

[0036] Figure 14 Schematic diagram of the structure of the electronic device of the present invention. DETAILED DESCRIPTION

[0037] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0038] Example 1:

[0039] The present invention provides a method for allocating power of multiple electrolyzers based on wind power and photovoltaic prediction, the flow diagram of which is shown in FIG. Figure 1 As shown, including:

[0040] Step S110, performing wind power forecasting based on historical wind power generation data and a wind power forecasting model to obtain forecasted wind power within a forecast period; the wind power forecasting model is a multivariate graph neural network based on an inverse residual structure;

[0041] Step S120, performing photovoltaic power prediction based on historical photovoltaic power generation data and a photovoltaic power prediction model to obtain predicted photovoltaic power within a prediction period;

[0042] Step S130, determining a reference number of operating electrolyzers within a prediction period based on the predicted wind power and the predicted photovoltaic power;

[0043] Step S140, based on the predicted wind power and the predicted photovoltaic power, the total hydrogen production power is determined using a power optimization model that takes system cost into consideration;

[0044] Step S150 , performing power distribution of the multiple electrolytic cells based on the operating status of each electrolytic cell in the multiple electrolytic cells, a reference number, and the total hydrogen production power.

[0045] In this embodiment, wind power forecasting is performed using an inverted residuals multivariate time series forecasting with graph neural networks (Inverted-MTGNN) based on an inverted residual structure. Combined with photovoltaic power forecasting, the reference number of operating electrolyzers is planned based on the forecast results. The total hydrogen production power is determined by considering the system cost. Finally, the power of multiple electrolyzers is allocated based on the operating status of the electrolyzers. By accurately predicting the number of operating electrolyzers based on wind and solar power output, and allocating the electrolyzer power of the total hydrogen production power based on the system cost and the health status of the electrolyzers, a pre-matching calculation of the electrolyzer hydrogen production power based on the wind and solar power output forecast can be achieved, thereby avoiding frequent start-up and shutdown of the electrolyzers due to real-time wind power output fluctuations. By allocating power based on cost and electrolyzer operating status, the rational use of electrolyzers is achieved, the life of multiple electrolyzers is extended, and the hydrogen production efficiency is improved.

[0046] In one implementation, Figure 2 As shown, the wind power prediction model includes a graph learning module, a standard convolution module, multiple graph convolution modules and an output module. Each graph convolution module includes a time convolution submodule, a graph convolution submodule and an inverse residual submodule connected in sequence. In the above step S110, wind power prediction is performed based on historical wind power generation data and the wind power prediction model to obtain the predicted wind power within the prediction period, including the following process:

[0047] The historical wind power generation data is input into the graph learning module and the standard convolution module respectively, and the adjacency matrix output by the graph learning module and the initial features output by the standard convolution module are obtained accordingly;

[0048] The initial features are sequentially passed through multiple graph convolution modules for graph convolution processing based on the inverted residual structure to obtain extracted features. In the graph convolution processing process of each graph convolution module, graph convolution processing based on the inverted residual structure is performed based on the adjacency matrix. The output of each graph convolution module is residually connected with the input and then used as the input of the next graph convolution module.

[0049] Perform feature concatenation on historical wind power generation data, the intermediate results of each graph convolution module, and the extracted features to obtain concatenated features;

[0050] The connection features are processed by convolution and nonlinear activation through the output module to obtain the predicted wind power within the prediction period.

[0051] In this implementation, the historical wind power generation data can be the wind power generation data of the latest month or several days, and the prediction period can be one or several days before or after the current day. For example, the wind power generation data of the latest month can be used to predict the wind power of the next day. Figure 2 As shown, historical wind power generation data can be used as the model's original input X, which is fed into the graph learning module and the standard convolution module. The output of the graph learning module is fed into the graph convolution submodule of each graph convolution module. The output of the standard convolution module passes through multiple graph convolution modules in sequence. The output of each graph convolution module is residually connected with the input of the previous graph convolution module and then fed into the next graph convolution module. The output of the temporal convolution submodule in each graph convolution module, the original input, and the output of the last graph convolution module are skip-connected and then fed into the output module. The output data Y of the output module is the predicted wind power. Specifically, the skip connection layer passes through the convolution layer, and the data dimension is changed to the same dimension as the output data. This layer serves to normalize the data information. The data of the temporal convolution submodules of different layers are merged through the skip connection layer and the output of the last graph convolution submodule before being fed into the output layer. The output layer passes through two standard convolution layers and a ReLU activation layer to adjust the output to the specified sequence length.

[0052] The graph convolution processing based on the inverted residual structure is performed within each graph convolution module, which can include the following processes:

[0053] The input of the graph convolution module is processed by the temporal convolution submodule to perform dilated convolution processing of different scales and splicing of channel dimensions to obtain the intermediate results of the graph convolution module;

[0054] The intermediate results and the adjacency matrix are used as the input of the graph convolution submodule, which performs multi-layer feature extraction and nonlinear transformation to obtain the selected features.

[0055] The selected features are passed through the inverted residual submodule for dimensionality-increasing convolution and depth-wise separable convolution to obtain the output of the graph convolution module.

[0056] In this implementation, the graph learning module can automatically extract unidirectional relationships between variables and integrate external knowledge such as variable attributes. The graph learning module can adaptively extract spatial relationships between variables from time series data and use these relationships to learn the adjacency matrix. The adjacency matrix is ​​represented as follows:

[0057] M1=tanh(αE1Θ1)

[0058] M2=tanh(αE2Θ2)

[0059]

[0060] Where E1 and E2 are the initial node embeddings, Θ1 and Θ2 are model parameters, α is the saturation rate of the activation function, M1 and M2 are the adjacency matrix model outputs, and A represents the output of the graph learning module. The superscript T represents the matrix transpose, ReLU represents the activation function corresponding to the rectified linear unit, and tanh represents the hyperbolic tangent activation function. After establishing the adjacency matrix, it is sparsified to reduce the computational cost of the neural network.

[0061] The standard convolution module is used to perform standard convolution processing on its input data.

[0062] Each graph convolution module includes a temporal convolution submodule, a graph convolution submodule, and an inverted residual submodule connected in sequence. For example, Figure 3 As shown in , the temporal convolution submodule includes two dilated convolution layers, the outputs of which are processed by different activation functions (such as tanh function and sigmoid function) and then fused and output. Figure 4 As shown in , each dilated convolution layer contains multiple standard one-dimensional dilated convolution kernels and connection layers. The standard one-dimensional dilated convolution kernels are used to extract high-dimensional temporal features. There are multiple different convolution kernels in the dilated convolution layer, which increases the receptive field without increasing the model parameters, thereby obtaining long-term dependencies. Figure 4 As shown in the figure, in the dilated convolution layer, the sequence is input into filters with different convolution kernels, such as 1×2, 1×3, 1×6, and 1×7 convolution kernels. The outputs of the remaining three filters are intercepted according to the minimum output length of the four filters, and the connection layer is used to splice the outputs of the four filters in the channel dimension.

[0063] The graph convolution submodule can be used to integrate the node information formed by the variables and the neighbor node information, for example, Figure 5 As shown in the figure, the graph convolution submodule contains two Mix-hop propagation layers, which respectively input the horizontal information and vertical information of the adjacency matrix. The outputs of the two Mix-hop propagation layers are H out1 and H out2 , H out1 and H out2 After fusion, the output H of the current layer is obtained out The structure of the Mix-hop communication layer is as follows: Figure 6 As shown, in the Mix-hop propagation layer, there are an information transmission module and an information selection module. In the information transmission module, the following formula is used for calculation:

[0064] H(k)=βH in +(1-β)H(k-1)

[0065] Where H(k) represents the input of the graph convolution depth k in the Mix-hop propagation layer; H(k-1) represents the input of the graph convolution depth k-1 in the Mix-hop propagation layer, β is a hyperparameter used to maintain the original node information ratio, k represents the depth of the graph convolution propagation layer, and H in Represents the hidden layer input of the current layer and the output of the previous layer. After data is processed by the information transfer module, node information is recursively propagated through the Multilayer Perceptron (MLP). As the number of network layers increases, the node representations of the same connected graph tend to be similar. To avoid data loss during selection, a parameter matrix can be introduced as a data feature selector.

[0066] The inverted residual submodule includes an inverted residual structure, which uses a dimensionality-increasing operation during feature extraction, resulting in a spindle-shaped network. The inverted residual structure uses depthwise separable convolution for feature extraction. The introduction of depthwise separable convolution greatly reduces network parameters, while enabling comprehensive feature extraction and reducing network computational complexity. In the inverted residual structure, the h-swish activation function and the h-sigmoid activation function reduce computational costs in quantization mode and accelerate model training efficiency. The activation function processing process is as follows:

[0067]

[0068] Where x d is the input of the inverted residual submodule.

[0069] For example, Figure 7 As shown in the figure, the inverted residual structure is the opposite of the residual structure. It first increases the dimension through 1×1 convolution, then performs convolution processing through 3×3 convolution kernel, and finally, reduces the dimension through ReLU6 activation function and 1×1 convolution. In the figure, Dwise represents the scale change of convolution processing, that is, from 3×3 to 1×1.

[0070] In this implementation, considering that traditional convolution extracts features after dimensionality compression, low-dimensional features have fewer parameters, thus reducing the overall computational complexity of the network. However, low-dimensional features are not conducive to fully extracting sample information, so an inverted residual structure is introduced. This inverted residual structure can further enhance the model's feature expression capabilities, improve the accuracy of the neural network, and thus enhance the accuracy of wind power prediction.

[0071] In some embodiments, before performing wind power prediction based on historical wind power generation data and a wind power prediction model, the correlation between power and other factors may be obtained using an improved grey correlation degree, which may include the following process:

[0072] Determine the relevant factors for wind power prediction, including wind speed, wind direction, temperature, humidity and air pressure at different altitudes;

[0073] Obtain the wind power historical factor sequence and the corresponding wind power sequence corresponding to each relevant factor;

[0074] Based on the cosine distance between each wind power historical factor sequence and the wind power sequence, the correlation between the wind power historical factor sequence and the wind power sequence corresponding to each relevant factor is determined; the target wind power historical factor sequence with a correlation greater than a preset first threshold is screened out, and the target wind power historical factor sequence and wind power sequence are used as historical wind power generation data.

[0075] In this embodiment, grey correlation analysis is a method for quantitatively describing and comparing the development and change trends of a system. This method determines whether the connection between series is close by comparing the geometric similarity between a reference sequence and several comparison data series, and reflects the degree of correlation between the series. Grey correlation analysis is generally divided into five steps: determining the variable sequence, dimensionless variables, calculating the correlation coefficient, calculating the correlation, and sorting. In wind power prediction, when considering wind speed, wind direction, temperature, humidity, and air pressure at different heights and performing multivariate model prediction, an improved grey correlation is used. When calculating the correlation coefficient, the cosine distance is used to represent the similarity between the two sequences, and the correlation between the input variable and the wind power is calculated. The correlation coefficient calculation formula is as follows:

[0076]

[0077] Where: i (k) represents the grey correlation degree of input variable i (e.g. wind speed, wind direction, temperature, humidity, air pressure and irradiation intensity) to wind power; || represents the absolute value; y j (k) represents the jth data in the reference sequence k, x ij (k) represents the jth data point in the comparison sequence k for variable i (wind speed, wind direction, temperature, humidity, air pressure, and irradiance), n represents the sequence length, d represents the cosine distance between the reference sequence and the comparison sequence, and ρ represents the resolution coefficient, typically set to 0.5. Grey correlation analysis is used to determine the correlation between variables. Variables with a grey correlation greater than 0.5 are selected as the target wind power historical factor sequence.

[0078] In some embodiments, when forecasting wind power, the improved grey correlation method is first used to determine the correlation between power and other factors. The mean and standard deviation of the data are then calculated. Based on the calculated results, abnormal data are removed, and the data is normalized. The processed data is then input into the Inverted-MTGNN model, and the model parameters are set to obtain the forecast results.

[0079] In another implementation, the photovoltaic power prediction model is a long-term and short-term time series network, and the hyperparameters of the long-term and short-term time series network are determined based on Bayesian and hyperband optimization algorithms.

[0080] In this implementation, a Bayesian Optimization and Hyperband (BOHB) algorithm is introduced based on the traditional Long and Short-term Time-series Network (LSTNet) to establish a BOHB-LSTNet photovoltaic power prediction model. The BOHB algorithm quickly determines the optimal values ​​of the hyperparameters of the LSTNet neural network. This algorithm can effectively utilize parallel resources and has high reliability. Its function is similar to that of intelligent optimization algorithms. Both select a target value and search for parameter values ​​that meet the target value within the parameter setting domain. This method is often used in neural networks to optimize the batch size, number of neurons, learning rate, etc.

[0081] BOHB relies on Hyperband to determine how many configurations to evaluate with which budget. It uses Bayesian optimization to replace the random selection of configurations at the beginning of each Hyperband iteration. When the required number of configurations is reached, the number of configurations is continuously halved until the optimal configuration is found. This algorithm uses the Hyperband algorithm to rationally allocate resources, accelerating the evaluation of hyperparameter combinations and ultimately finding the optimal hyperparameter combination. During the continuous halving, Bayesian optimization replaces random sampling. The specific implementation process is as follows: first, the total control budget parameters are initialized; then, the number of randomly sampled configurations and the iteration resources for each configuration are calculated; second, the number of configurations to be evaluated is initialized and the corresponding iteration resources are allocated; finally, each evaluation point is obtained by maximizing the acquisition function, and the objective function is solved to obtain the optimal hyperparameters.

[0082] The long-term and short-term time series network consists of convolutional layers, recurrent layers and autoregressive models, such as Figure 8 As shown in the figure, the convolutional layer is the first layer of the LSTNet model. It is a convolutional network without pooling, designed to extract short-term patterns in the time dimension and local dependencies between variables. The convolutional layer consists of multiple filters (the height of the filter is set to the same as the number of variables), and the output of the k-th filter is as follows:

[0083] h k =RELU(W k *X+b k )

[0084] Where h kRepresents the output of the kth filter, * represents the convolution operation, and the RELU function is RELU(x)=max(0,x); W k 、b k are the weight vector and bias vector of the kth filter respectively; X is the input matrix. By filling zeros on the left side of the input matrix X, each vector h k The length is T l , the output matrix size of the convolutional layer is d c ×T l , where d c Indicates the number of filters.

[0085] The loop layer is an LSTM neural network with the following structure: Figure 9 As shown. Due to the problems of gradient vanishing and gradient exploding, the learning ability of recurrent neural networks is still limited, and the actual effect is often not ideal. LSTM neural networks have a relatively long memory time for valuable information, and its advantages make them widely used in time series prediction. Figure 9 As can be seen from the figure, the improvement of LSTM is to add three gates. The input gate controls whether the new state of the calculation can be updated to the storage unit, the forget gate controls the forgotten information in the previous storage unit, and the output of the output gate depends on the current storage unit.

[0086] The calculation formula of the LSTM model at time t is as follows:

[0087] i t =σ(w i x t +U i h t-1 +b i )

[0088] f t =σ(w f x t +U f h t-1 +b f )

[0089] o t =σ(w o x t +U o h t-1 +b o )

[0090]

[0091] h t =o t ⊙tanh(c t )

[0092] Where: it represents the output of the input gate at time t, f t represents the output of the forget gate at time t, o t represents the output of the output gate at time t, c t is the output of the memory neuron at time t, c t-1 represents the output of the memory neuron at time t-1; w i 、w f 、w o 、w c 、U i 、U f 、U o 、U c Represents the weight matrix of the hidden layer matrix of the input matrix of the input gate, forget gate, and output gate mentioned above; b i 、b f 、b o Represents the bias vector of the input gate, forget gate, and output gate; is the operational variable of one of the links, σ represents the Sigmoid activation function; tanh represents the hyperbolic tangent activation function, x t represents the input of the LSTM neural network at time t, h t represents the external state of the hidden layer at time t, h t-1 represents the external state of the hidden layer at time t-1.

[0093] The formula for the autoregressive model is as follows:

[0094]

[0095] Where m t Represents the prediction result of the autoregressive model component at time t, a s1 represents the coefficient at time s1; x t-s1 Represents the input of the autoregressive model at time t-s1; s1=1,2,…p, p represents the order of the autoregressive model, ω t represents the white noise at time t.

[0096] In some implementations, before performing photovoltaic power prediction based on historical photovoltaic power generation data and a photovoltaic power prediction model, the historical photovoltaic power generation data may be screened through the following process:

[0097] Determine candidate factors related to photovoltaic power prediction, including wind speed, wind direction, temperature, humidity, air pressure and irradiation intensity;

[0098] Obtain the photovoltaic historical factor sequence and the corresponding photovoltaic power generation power sequence corresponding to each candidate factor;

[0099] Based on Spearman correlation analysis, the correlation coefficient between the photovoltaic historical factor series and the photovoltaic power series corresponding to each candidate factor is determined;

[0100] Target photovoltaic historical factor sequences with correlation coefficients greater than a preset second threshold are screened out, and the target photovoltaic historical factor sequences and photovoltaic power generation power sequences are used as historical photovoltaic power generation data.

[0101] In this implementation, the Spearman correlation coefficient method is applicable to nonlinear, non-normally distributed data and can be used for correlation analysis of historical operating data of photovoltaic power plants. In photovoltaic power forecasting, when considering relevant candidate factors such as wind speed, wind direction, temperature, humidity, air pressure, and irradiation intensity, the Spearman correlation coefficient can be used to screen data. The Spearman correlation coefficient calculation formula is as follows:

[0102]

[0103] Where r s represents the Spearman correlation coefficient, is the rank of the characteristic parameter x (wind speed, wind direction, temperature, humidity, air pressure and radiation intensity) at the i-th point, is the order of the characteristic parameter y (photovoltaic power) at the i-th point, d i Indicates the position distance between the feature parameter x and the feature parameter y at the i-th point, n s The Spearman correlation coefficient is used to calculate the correlation between the input variables and power, and the variables with an absolute value of the Spearman correlation coefficient greater than 0.7 are selected as the target photovoltaic historical factor series.

[0104] In some embodiments, the prediction effects of the wind power prediction model and the photovoltaic power prediction model can be evaluated to determine the prediction accuracy of the model. For example, general evaluation indicators of the time series prediction model can be used to evaluate the prediction performance and data characteristics of wind power and photovoltaic power, including mean absolute error (MAE), normalized mean absolute error (NMAE), mean square error (MSE), root mean square error (RMSE), normalized root mean square error (NRMSE), and goodness of fit R. 2 (Goodness of Fit,R 2 ). It is defined as follows:

[0105]

[0106] NMAE=(MAE / (y max -y min ))×100%

[0107]

[0108] NRMSE = (RMSE / (y max -y min ))×100%

[0109]

[0110] Where y i represents the actual value of the i-th sample, is the predicted value of the i-th sample, m is the number of samples, y max and y min Indicates the maximum and minimum values ​​of the actual sample values.

[0111] In another implementation, considering that in the hydrogen production system, the power will fluctuate around a certain value over a period of time, resulting in frequent start and stop of multiple electrolyzers. Therefore, it is necessary to add a standby unit (i.e., a standby electrolyzer). At the same time, the electrolyzer array is rarely able to operate at full load. When the output is low, it is inevitable that some units will be on standby and some units will be shut down. This requires considering the power changes in the future period, and reasonably arranging the number of operating units (the number of operating electrolyzers) and the number of standby units to avoid long periods of standby for multiple units, increase the operating time, and avoid short-term start and stop of units to reduce unit losses. In order to address the above problems, the wind power and photovoltaic power within the period T are predicted, and the number of standby units is allocated according to the prediction results. For example, the determination of the reference number of operating electrolyzers within the prediction period based on the predicted wind power and predicted photovoltaic power in step S130 may include the following process:

[0112] Based on the predicted wind power and the predicted photovoltaic power, the total power generation within the forecast period is determined;

[0113] Based on the change in total power generation at each moment within the forecast period, continuous time periods in which total power generation fluctuates around a fixed value are selected and used as the segments to be optimized.

[0114] The average value of the total power generation of the section to be optimized is taken as the average power of each electrolyzer at each moment in the section to be optimized;

[0115] Based on the ratio of the average power and the rated power of each electrolytic cell, the reference number of electrolytic cells operating in the section to be optimized is determined.

[0116] For example, in order to cope with the changes in wind power output and photovoltaic output, the total power generation within the forecast period can be determined by the following process:

[0117] Perform first-order difference processing on the predicted wind power and predicted photovoltaic power to obtain the difference results at each moment in the prediction period;

[0118] The differential results at each moment are accumulated and summed to obtain the total generated power within the prediction period.

[0119] In this implementation, wind power and photovoltaic power often fluctuate within a small range due to their volatility, randomness, and intermittency. To mitigate the increase or decrease in the number of electrolyzers in operation caused by these small fluctuations, a first-order difference process is performed on the wind power forecast and photovoltaic forecast to determine the total generated power. The first-order difference process is performed according to the following formula:

[0120] Δy t =y wind (t+1)+y PV (t+1)-y wind (t)-y PV (t)

[0121] Where Δy t represents the predicted power generation at time t, y wind (t) represents the predicted wind power at time t, y PV (t) represents the predicted photovoltaic power generation at time t, y wind (t+1) represents the predicted wind power at time t+1, y PV (t+1) represents the predicted photovoltaic power generation at time t+1.

[0122] Then, the first-order difference results at each moment are accumulated and calculated to obtain the total generated power within the forecast period, which is as follows:

[0123]

[0124] Where T is the duration of the prediction period, t represents the time point within the prediction period, and A c Indicates the total generated power during the forecast period.

[0125] After obtaining the total power generation, the area where the change in the total power generation fluctuates around a fixed value is screened out and the area is determined as the area to be optimized. For example, if the total power generation exceeds a given value C, the period T1 is recorded and the total power generation is recalculated from this moment. There are m moments in the period T1, which means that the total power generation at these m moments fluctuates around a certain value Pm. P m(s) is the average power of electrolytic cell s in period T1, which is replaced by the average power of period T1. The calculation formula is as follows:

[0126]

[0127] Where m represents the number of sampling points in period T1, and y(t) represents the total generated power at time t in period T1.

[0128] In the period T1, the number of operable units N corresponding to Pm m is the reference number. When the number of running units is less than N m When N m The calculation of is as follows:

[0129]

[0130] Where: P N (s) is the rated power of electrolytic cell s, Indicates a round-up operation.

[0131] When N m When >N, the number of operating units is set to the total number N of electrolytic cells in the multi-stack electrolytic cell.

[0132] In another implementation, during the electrolyzer optimization control cycle, the total hydrogen production power is calculated based on the economic function value to maximize the benefits of hydrogen electrolysis. For example, in step S140, the total hydrogen production power is determined based on the predicted wind power and predicted photovoltaic power using a power optimization model that takes system costs into account, including:

[0133] Considering the operation and maintenance costs of the electrolyzer, hydrogen storage tank, wind and solar curtailment penalty costs, and grid electricity purchase costs, an objective optimization function is constructed with the goal of minimizing system costs.

[0134] Determine the constraints of the objective optimization function based on power balance constraints, operational safety constraints of multiple electrolyzers, start and stop times constraints of the hydrogen storage tank, and state of charge range constraints of the hydrogen storage tank.

[0135] Based on the target optimization function and the constraint conditions, a power optimization model is obtained;

[0136] Based on the predicted wind power, predicted photovoltaic power, electrolyzer operation and maintenance costs, and hydrogen storage tank operation and maintenance costs, the power optimization model is solved to obtain the total hydrogen production power.

[0137] In this implementation, the objective optimization function is a multi-objective optimization function that takes into account the electrolyzer operation and maintenance costs, hydrogen storage tank operation and maintenance costs, wind power curtailment penalty costs, solar power curtailment penalty costs, and the cost of purchasing electricity from the grid. Specifically, the function is as follows:

[0138]

[0139] Where: C total represents the total economic cost of the system, C CO represents the electrolyzer operation and maintenance cost, C HO represents the operation and maintenance cost of the hydrogen storage tank, C Wind represents the penalty cost for wind curtailment, C PV represents the penalty cost of abandoning light, C EL represents the cost of purchasing electricity from the grid, Represents hydrogen sales revenue. Each item in the multi-objective optimization function (objective optimization function) is analyzed as follows:

[0140] (1) Electrolyzer operation and maintenance costs:

[0141]

[0142] Where: N is the total number of electrolytic cells in the system, β EO (s) represents the operation and maintenance cost coefficient of electrolytic cell s, represents the rated capacity of electrolytic cell s, c EO (s) represents the construction cost per unit capacity of electrolytic cell s.

[0143] (2) Hydrogen storage tank operation and maintenance costs:

[0144]

[0145] Where: β HO represents the operation and maintenance cost coefficient of the hydrogen storage tank l, c HO represents the unit capacity construction cost of hydrogen storage tank l, Indicates the capacity of the hydrogen storage tank l,

[0146] (3) Penalty cost for wind curtailment:

[0147]

[0148] Where: c Wind represents the penalty cost per unit of wind curtailment, represents the amount of wind power abandoned at time t, and T represents the duration of the prediction period.

[0149] (4) Penalty cost for abandoned light:

[0150]

[0151] Where: c PV represents the unit penalty cost of abandoning light, Indicates the amount of abandoned solar power at time t.

[0152] (5) Cost of purchasing electricity from the power grid:

[0153]

[0154] Where: c grid (t) represents the time-of-use electricity price at time t, P grid (t) represents the purchased power at time t, and τ represents the control step size.

[0155] (6) Hydrogen production income:

[0156]

[0157] Where: represents the hydrogen price per unit mass at time t, represents the hydrogen production of electrolyzer s at time t.

[0158] The constraints of the objective optimization function include power balance constraints, multi-stack electrolyzer operation constraints, and hydrogen storage tank constraints. The specific analysis is as follows:

[0159] (1) Power balance constraints

[0160] The power purchase from the grid, wind power output, photovoltaic output, wind curtailment, solar curtailment, electrical load, and power balance constraints of the multi-stack electrolyzer array are:

[0161]

[0162] Where: P grid (t) represents the amount of electricity purchased from the power grid at time t, P Wind (t) represents the predicted wind power at time t, P PV (t) represents the predicted photovoltaic power at time t, P L (t) represents the electrical load at time t, P EO (t) represents the total power of multiple electrolytic cells at time t, P EO,s (t) represents the power of electrolytic cell s at time t, and N is the total number of electrolytic cells in the system.

[0163] (2) Multi-stack electrolyzer operation constraints

[0164] Based on the operating characteristics of alkaline electrolytic cells, the total input power of multiple stacks of electrolytic cells must be less than the operating power of each electrolytic cell in an overloaded state. The minimum input power is the minimum operating power of a single electrolytic cell. Therefore, if the electrolytic cell is an alkaline electrolytic cell, its electrolytic cell power constraint is:

[0165]

[0166] Based on the operating characteristics of PEM (Proton Exchange Membrane Electrolysis Cell, fuel cell) electrolyzers, the total input power of multiple stacks of electrolyzers must be less than the operating power of each electrolyzer in an overloaded state. The minimum input power is the minimum operating power of a single electrolyzer. Therefore, if the electrolyzer is a PEM electrolyzer, its electrolyzer power constraint is:

[0167]

[0168] In order to avoid the frequent start and stop of the electrolytic cell and the loss of its service life, the number of start and stop times of the electrode cell is restricted:

[0169] 0≤Y≤Y

[0170] smax

[0171] Where: Y s Indicates the number of shutdowns of the sth electrolytic cell; Y max Indicates the maximum number of shutdowns allowed for the electrolyzer.

[0172] (3) Hydrogen storage tank constraints

[0173] During the charging and discharging process of the hydrogen storage tank, in order to ensure its reliability, the charge state of the hydrogen storage tank should be limited to a certain range. At the same time, in order to meet the operating conditions of the next day, the initial hydrogen storage volume in the hydrogen storage tank must be the same. The constraints are as follows:

[0174]

[0175] Q SOC (0) = Q SOC (T)

[0176] Where: Q SOC (t) represents the state of charge of the hydrogen storage tank at time t, Q SOC (0) represents the state of charge of the hydrogen storage tank at the initial moment of the electrolyzer optimization control cycle, Q SOC (T) represents the state of charge of the hydrogen storage tank at the end of the electrolyzer optimization control cycle, Indicates the lower limit of the hydrogen storage tank's state of charge. Indicates the upper limit of the hydrogen tank's state of charge.

[0177] Based on the above multi-objective optimization function and constraints, the total hydrogen production power can be obtained by optimizing the model solver. For example, the cplex solver in the Python environment of PyCharm 2021.3.1 x64 software can be used to solve the objective optimization function to obtain the total hydrogen production power. In addition, data such as hydrogen production, electrolysis efficiency, operating time, and number of starts and stops for each electrolyzer can also be solved.

[0178] In another implementation, considering that when the electrolyzer array is operating under overload conditions or low power conditions, the electrolyzers with earlier numbers will be in an overload state for a long time or the electrolyzers with later numbers will be in a shutdown state for a long time, resulting in damage to the internal materials of the electrolyzers, a decrease in hydrogen production efficiency, and a reduction in the overall operating life of the electrolyzer array. In order to avoid a single electrolyzer operating in an unstable state for a long time, the multiple electrolyzers are rotated and numbered to balance the operating state of each electrolyzer under unstable conditions. For example, the power allocation of the multiple electrolyzers based on the operating state, reference number, and total hydrogen production power of each electrolyzer in the multiple electrolyzers in step S150 includes the following processes:

[0179] Sequentially number each electrolytic cell based on its health status;

[0180] Determine the number of standby electrolytic cells based on the current actual number of operating electrolytic cells and the reference number;

[0181] The power of each electrolyzer is allocated based on the sequential number of the electrolyzer and the total hydrogen production power.

[0182] In this implementation, the state of health (SOH) of the electrolytic cell can be used to characterize the operating state of the electrolytic cell. The SOH of the electrolytic cell is represented by the voltage at the rated current and is defined as follows:

[0183]

[0184] Where: SOH(s,t) is the health status of electrolytic cell s at time t; α e is the state parameter, which is usually set to 90%; N is the total number of electrolytic cells; V N (s,t) is the rated power of electrolytic cell s at time t, V N (s,0) is the rated power of electrolyzer s at the initial moment.

[0185] In actual control, when the number of operating electrolyzers is less than the reference number, standby electrolyzers are added to meet operational demand. The stacks are arranged in descending order based on their state of health (SOH), with the electrolyzer with the highest SOH ranked first, and so on. Furthermore, according to the power allocation strategy, lower-numbered electrolyzers carry higher power, thus achieving a balancing effect.

[0186] Exemplarily, the power allocation of each electrolyzer is performed based on the sequence number of the electrolyzer and the total hydrogen production power, including the following process:

[0187] Based on the upper power limit, rated power and lower power limit of each electrolyzer, three power ranges are divided;

[0188] Based on the total hydrogen production power, determine the target power range corresponding to the section to be optimized;

[0189] The power allocation of the electrolytic cells is performed based on the power allocation strategy corresponding to the target power interval and the sequence number of each electrolytic cell.

[0190] In this embodiment, each electrolyzer has an upper power limit, a rated power, and a lower power limit, and the upper power limit, rated power, and lower power limit of different electrolyzers in the multi-stack electrolyzer are the same. For example, the three power intervals and the number of operating electrolyzers are as follows:

[0191] For different power ranges, the number of operating electrolytic cells corresponding to each power range is determined according to the following formula:

[0192]

[0193] Among them, N f Indicates the number of electrolytic cells in operation, P el represents the total power of hydrogen production, N represents the total number of electrolyzers in the hydrogen production system, P m (s) represents the operating power of electrolytic cell s in the section m to be optimized, P min (s) represents the lower limit power of electrolytic cell s, P N (s) represents the rated power of electrolytic cell s, P max (s) represents the upper limit power of electrolytic cell s, and s represents the electrolytic cell number.

[0194] If the total hydrogen production power is in the interval 0≤P el <N·P min (s), the power distribution of each electrolytic cell is carried out according to the following formula:

[0195]

[0196] If the total hydrogen production power is in the range N·P min (s)≤P el <N·P N (s), the power distribution of each electrolytic cell is carried out according to the following formula:

[0197]

[0198] If the total hydrogen production power is in the range N·P N (s)≤P el ≤N·P max (s), the power distribution of each electrolytic cell is carried out according to the following formula:

[0199]

[0200] For example, when the electrolyzer is controlled in real time, power allocation starts from the first electrolyzer. When the total hydrogen production power is equal to or greater than the minimum operating power, the unit is started. When the total hydrogen production power is equal to or greater than the lower limit power, the power of the second electrolyzer is allocated, and this continues until the Nth unit. After that, power allocation starts from the first electrolyzer again. When the total hydrogen production power is equal to or greater than the rated power, the power of the second electrolyzer is allocated, and this continues until the Nth unit. Finally, power allocation starts from the first electrolyzer again. When the total hydrogen production power is equal to or greater than the upper limit power, the power of the second electrolyzer is allocated, and this continues until the Nth unit. Figure 10 As shown in the figure, taking 4 electrolyzers as an example, the power allocation results of multiple electrolyzer stacks with total hydrogen production power in different power partitions are shown.

[0201] The following is an example of a specific embodiment to illustrate the application scenario of the method of the embodiment of the present invention. Figure 11 As shown, the wind power-photovoltaic-multi-stack electrolyzer system includes: a power grid, wind turbines, a photovoltaic power station, multiple electrolyzer stacks (electrolyzer 1, electrolyzer 2, ..., electrolyzer N), a hydrogen storage tank, wind power loads, photovoltaic loads, and hydrogen loads. Alternating Current / Direct Current (AD / DC) converters are used to convert AC / DC between the power grid, wind turbines, and photovoltaic power station and the multiple electrolyzer stacks. The system operates by utilizing wind and photovoltaic power generation to produce hydrogen, using hydrogen energy to meet the demands of the electricity load (wind power load and photovoltaic load) and hydrogen load. When wind power and photovoltaic power generation fall below the load demand, electricity is purchased from the power grid. Standby units are planned based on wind power and photovoltaic output forecasts, and the operating status of the electrolyzers is balanced based on multiple objectives.

[0202] The following is a specific example to illustrate the method flow of the embodiment of the present invention. Figure 12 As shown, the following process may be included:

[0203] 1) Input relevant data of wind farms, photovoltaic stations and multi-stack electrolyzers (such as historical wind power generation data and historical photovoltaic power generation data, electrolyzer rated power, etc.);

[0204] 2) Predict wind power based on the Inverted-MTGNN model and photovoltaic power based on the BOHB-MLSTNet model;

[0205] 3) Perform first-order difference on the prediction results and accumulate them to obtain the total power in the future time period and calculate the reference number of operating electrolytic cells based on this;

[0206] 4) Determine whether the optimization period has been reached. If so, proceed to step 5), otherwise proceed to step 8);

[0207] 5) Considering the electrolyzer operation and maintenance costs, hydrogen storage tank operation and maintenance costs, wind power abandonment penalty, solar power abandonment penalty, electricity purchase costs, and hydrogen sales revenue, calculate the economic objective function and constraints (i.e., obtain the target optimization control model);

[0208] 6) Determine whether the maximum iteration cycle of the optimizer has been reached. If so, proceed to step 7), otherwise, go to step 1.

[0209] 7) Obtain the total hydrogen production power of multiple electrolyzers and calculate the hydrogen production, electrolysis efficiency, operating time, and number of starts and stops.

[0210] 8) The electrolytic cells are numbered according to the SOH sorting, the electrolytic cell operating conditions are calculated based on the power of multiple electrolytic cells, and the electrolytic cell power is allocated in real time.

[0211] Based on the operating characteristics of wind power and photovoltaic power generation, this invention proposes a multi-stack electrolyzer economic operation and multi-stack power allocation method based on wind and solar power output prediction based on deep learning algorithm. Through the deep learning algorithm, wind power output and photovoltaic output are accurately predicted, the service life of the multi-stack electrolyzer is extended, the number of electrolyzer start-ups and shutdowns is reduced, and its adaptability to fluctuations in renewable energy is improved.

[0212] Example 2:

[0213] Based on the same inventive concept, the present invention also provides a multi-stack electrolyzer power distribution system based on wind power and photovoltaic prediction, the structural diagram of which is shown in FIG. Figure 13 Shown, including:

[0214] The wind power prediction unit 1110 is used to predict wind power based on historical wind power generation data and a wind power prediction model to obtain predicted wind power within a prediction period; the wind power prediction model is a multivariate graph neural network based on an inverse residual structure;

[0215] The photovoltaic prediction unit 1120 is used to perform photovoltaic power prediction based on historical photovoltaic power generation data and a photovoltaic power prediction model to obtain predicted photovoltaic power within a prediction period;

[0216] an electrolytic cell quantity determination unit 1130 for determining a reference quantity of operating electrolytic cells within a prediction period based on the predicted wind power and the predicted photovoltaic power;

[0217] The total power determination unit 1140 determines the total hydrogen production power based on the predicted wind power and the predicted photovoltaic power using a power optimization model that takes system cost into consideration;

[0218] The power distribution unit 1150 is used to distribute the power of the multiple electrolyzer stacks based on the operating status of each electrolyzer in the multiple electrolyzer stacks, the reference number and the total hydrogen production power.

[0219] In a possible implementation, the wind power prediction unit 1110 includes:

[0220] A graph learning module is used to obtain an output adjacency matrix based on the input historical wind power generation data;

[0221] A standard convolution module is used to obtain the initial features of the output based on the input historical wind power generation data;

[0222] Multiple graph convolution modules are used to perform graph convolution processing based on the inverted residual structure on the initial features to obtain extracted features. In the graph convolution processing process of each graph convolution module, graph convolution processing based on the inverted residual structure is performed based on the adjacency matrix. The output of each graph convolution module is residually connected with the input and then used as the input of the next graph convolution module.

[0223] The skip connection module is used to perform feature connection on the historical wind power generation data, the intermediate results of each graph convolution module, and the extracted features to obtain the connection features;

[0224] The output module is used to perform convolution processing and nonlinear activation processing on the connection features through the output module to obtain the predicted wind power within the prediction period.

[0225] In one possible implementation, each graph convolution module includes:

[0226] The temporal convolution submodule is used to perform dilated convolution processing of different scales and channel dimension splicing on the input of the graph convolution module to obtain the intermediate results of the graph convolution module;

[0227] The graph convolution submodule is used to perform multi-layer feature extraction and nonlinear transformation on the intermediate results and adjacency matrix to obtain selected features;

[0228] The inverted residual submodule is used to perform up-dimensional convolution and depth-wise separable convolution on the selected features to obtain the output of the graph convolution module.

[0229] In a possible implementation, the photovoltaic power prediction model is a long-short term time series network, and the photovoltaic prediction unit 1120 is specifically configured to determine hyperparameters of the long-short term time series network based on a Bayesian and hyperband optimization algorithm.

[0230] In a possible implementation, the wind power prediction unit 1110 includes:

[0231] A relevant factor determination module is used to determine relevant factors for wind power prediction, including wind speed, wind direction, temperature, humidity and air pressure at different heights;

[0232] A wind power sequence acquisition module is used to obtain the wind power historical factor sequence and the corresponding wind power sequence corresponding to each relevant factor;

[0233] A correlation determination module is used to determine the correlation between the wind power historical factor sequence and the wind power sequence corresponding to each relevant factor based on the cosine distance between each wind power historical factor sequence and the wind power sequence;

[0234] The wind power data screening module is used to screen out target wind power historical factor sequences with a correlation greater than a preset first threshold, and use the target wind power historical factor sequences and wind power sequences as historical wind power generation data.

[0235] In one possible implementation, the photovoltaic prediction unit 1120 includes:

[0236] A candidate factor determination module is used to determine candidate factors related to photovoltaic power prediction, including wind speed, wind direction, temperature, humidity, air pressure and irradiation intensity;

[0237] A photovoltaic sequence acquisition module is used to obtain the photovoltaic historical factor sequence and the corresponding photovoltaic power generation power sequence corresponding to each candidate factor;

[0238] A correlation coefficient determination module is used to determine the correlation coefficient between the photovoltaic historical factor sequence and the photovoltaic power sequence corresponding to each candidate factor based on Spearman correlation analysis;

[0239] The photovoltaic data screening module is used to screen out target photovoltaic historical factor sequences with correlation coefficients greater than a preset second threshold value, and use the target photovoltaic historical factor sequences and photovoltaic power generation power sequences as historical photovoltaic power generation data.

[0240] In one possible implementation, the electrolytic cell quantity determination unit 1130 includes:

[0241] A total power determination module is used to determine the total generated power within the prediction period based on the predicted wind power and the predicted photovoltaic power;

[0242] The optimization section screening module is used to screen out continuous time periods in which the total power generation power fluctuates around a fixed value based on the change in the total power generation power at each moment within the forecast period, and use the continuous time periods as sections to be optimized;

[0243] A power determination module, configured to use the average value of the total generated power of the section to be optimized as the average power of each electrolytic cell at each moment in the section to be optimized;

[0244] The electrolytic cell data determination module is used to determine the reference number of operating electrolytic cells in the section to be optimized based on the ratio of the average power to the rated power of each electrolytic cell.

[0245] In a possible implementation, the total power determination module includes:

[0246] The differential submodule is used to perform first-order differential processing on the predicted wind power and predicted photovoltaic power to obtain differential results at each moment in the prediction period;

[0247] The summation submodule is used to accumulate and sum the differential results at each moment to obtain the total generated power within the prediction period;

[0248] The first-order difference processing is performed as follows:

[0249] Δy t =y wind (t+1)+y PV (t+1)-y wind (t)-y PV (t)

[0250] Where Δy t represents the predicted power generation at time t, y wind (t) represents the predicted wind power at time t, y PV (t) represents the predicted photovoltaic power generation at time t.

[0251] In a possible implementation, the total power determining unit 1140 includes:

[0252] An optimization function construction module is used to consider the operation and maintenance costs of the electrolyzer, the operation and maintenance costs of the hydrogen storage tank, the penalty costs for wind and solar power curtailment, and the grid's electricity purchase costs, and to construct a target optimization function with the goal of minimizing system costs;

[0253] A constraint determination module is used to determine the constraint conditions of the objective optimization function based on power balance constraints, operational safety constraints of multiple electrolyzers, start and stop times constraints of the hydrogen storage tank, and state of charge range constraints of the hydrogen storage tank;

[0254] An optimization model determination module, used to obtain a power optimization model based on a target optimization function and constraint conditions;

[0255] The model solving module is used to solve the power optimization model based on the predicted wind power, predicted photovoltaic power, electrolyzer operation and maintenance costs, and hydrogen storage tank operation and maintenance costs to obtain the total hydrogen production power.

[0256] In one possible implementation, the operating status of the electrolytic cell includes the health status of the electrolytic cell, and the power distribution unit 1150 includes:

[0257] a sorting module for sequentially numbering the electrolytic cells based on the health status of each electrolytic cell;

[0258] A standby quantity determination module is used to determine the number of standby electrolytic cells based on the current actual number of operating electrolytic cells and a reference number;

[0259] The power allocation module is used to allocate power to each electrolyzer based on the sequence number of the electrolyzer and the total hydrogen production power.

[0260] In one possible implementation, the power distribution module includes:

[0261] The partition submodule is used to divide each electrolyzer into three power intervals based on the upper power limit, rated power and lower power limit;

[0262] The interval determination submodule is used to determine the target power interval corresponding to the section to be optimized based on the total hydrogen production power;

[0263] The power allocation submodule is used to allocate power to the electrolytic cells based on the power allocation strategy corresponding to the target power range and the sequence number of each electrolytic cell.

[0264] In a possible implementation manner, the power distribution submodule is specifically configured to:

[0265] For different power ranges, the number of operating electrolytic cells corresponding to each power range is determined according to the following formula:

[0266]

[0267] Among them, N f Indicates the number of electrolytic cells in operation, P el represents the total power of hydrogen production, N represents the total number of electrolyzers in the hydrogen production system, P m (s) represents the operating power of electrolytic cell s in the section m to be optimized, P min (s) represents the lower limit power of electrolytic cell s, P N (s) represents the rated power of electrolytic cell s, P max (s) represents the upper limit power of electrolytic cell s, and s represents the electrolytic cell number;

[0268] If the total hydrogen production power is in the interval 0≤P el <N·P min (s), the power distribution of each electrolytic cell is carried out according to the following formula:

[0269]

[0270] If the total hydrogen production power is in the range N·P min (s)≤P el <N·P N (s), the power distribution of each electrolytic cell is carried out according to the following formula:

[0271]

[0272] If the total hydrogen production power is in the range N·P N (s)≤P el≤N·P max (s), the power distribution of each electrolytic cell is carried out according to the following formula:

[0273]

[0274] Example 3:

[0275] like Figure 14 As shown, the present invention also provides an electronic device, which may be a computer, a single-chip microcomputer, a smart mobile device, or the like. The electronic device in this embodiment may include a processor, a memory, a transceiver component, and the like. The memory, processor, and transceiver component are connected via a bus; the memory may be used to store an execution program, which may include instructions; and the processor may be used to execute the instructions stored in the memory. The memory may also be used to store data, which may be accessed and / or modified during the execution of the instructions.

[0276] The processor may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the storage medium to implement the corresponding method flow or corresponding function, so as to implement the steps of a multi-stack electrolyzer power allocation method based on wind power and photovoltaic prediction in the above embodiment.

[0277] Example 4

[0278] Based on the same inventive concept, the present invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory), which is a memory device in an electronic device for storing programs and data. It can be understood that the storage medium here can include both built-in storage media in the electronic device and, of course, extended storage media supported by the electronic device. The storage medium provides a storage space that stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space. These instructions can be one or more execution programs (including program codes). It should be noted that the storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The processor loads and executes one or more instructions stored in the storage medium, which can implement the steps of a multi-stack electrolyzer power allocation method based on wind power and photovoltaic prediction in the above embodiment.

[0279] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0280] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0281] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1The function specified in one or more boxes.

[0282] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0283] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit its scope of protection. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that after reading the present invention, those skilled in the art may still make various changes, modifications or equivalent substitutions to the specific implementation methods of the application, but these changes, modifications or equivalent substitutions are all within the scope of protection of the pending claims.

Claims

1. A method for allocating power to multiple electrolyzers based on wind power and photovoltaic power forecasting, characterized in that: The method comprises: Perform wind power forecasting based on historical wind power generation data and a wind power forecasting model to obtain the forecasted wind power within the forecast period; the wind power forecasting model is a multivariate graph neural network based on an inverse residual structure; Performing photovoltaic power forecasting based on historical photovoltaic power generation data and a photovoltaic power forecasting model to obtain predicted photovoltaic power within the forecast period; determining a reference number of operating electrolyzers within the prediction period based on the predicted wind power and the predicted photovoltaic power; Based on the predicted wind power and the predicted photovoltaic power, determining the total hydrogen production power using a power optimization model that takes system cost into consideration; Power distribution of the multiple stacks of electrolyzers is performed based on the operating status of each electrolyzer in the multiple stacks of electrolyzers, the reference number and the total hydrogen production power.

2. The method according to claim 1, characterized in that The wind power prediction model includes a graph learning module, a standard convolution module, multiple graph convolution modules, and an output module. The wind power prediction is performed based on historical wind power generation data and the wind power prediction model to obtain the predicted wind power within the prediction period, including: Inputting the historical wind power generation data into the graph learning module and the standard convolution module respectively, and obtaining the adjacency matrix output by the graph learning module and the initial features output by the standard convolution module accordingly; The initial features are sequentially passed through a plurality of graph convolution modules to perform graph convolution processing based on an inverted residual structure to obtain extracted features; wherein in the graph convolution processing process of each graph convolution module, graph convolution processing based on an inverted residual structure is performed based on the adjacency matrix, and the output of each graph convolution module is residually connected with the input and then used as the input of the next graph convolution module; Performing feature concatenation on the historical wind power generation data, the intermediate results of each graph convolution module, and the extracted features to obtain a concatenated feature; The connection features are subjected to convolution processing and nonlinear activation processing through the output module to obtain the predicted wind power within the prediction period.

3. The method according to claim 2, characterized in that Each graph convolution module includes a temporal convolution submodule, a graph convolution submodule, and an inverted residual submodule connected in sequence. Each of the graph convolution modules performs graph convolution processing based on an inverted residual structure, including: Based on each of the graph convolution modules, the input of the graph convolution module is subjected to dilated convolution processing of different scales and splicing of channel dimensions by the temporal convolution submodule to obtain an intermediate result of the graph convolution module; Using the intermediate result and the adjacency matrix as inputs of the graph convolution submodule, performing multi-layer feature extraction and nonlinear transformation by the graph convolution submodule to obtain selected features; The selected features are passed through the inverted residual submodule to perform dimensionality-increasing convolution and depth-wise separable convolution to obtain the output of the graph convolution module.

4. The method according to claim 1, wherein The photovoltaic power prediction model is a long-term and short-term time series network, and the hyperparameters of the long-term and short-term time series network are determined based on Bayesian and hyperband optimization algorithms.

5. The method according to claim 1, wherein Before performing wind power prediction based on historical wind power generation data and the wind power prediction model, the method further includes: Determining relevant factors for wind power prediction, wherein the relevant factors include wind speed, wind direction, temperature, humidity and air pressure at different heights; Obtaining a wind power historical factor sequence and a corresponding wind power sequence corresponding to each of the relevant factors; Based on the cosine distance between each of the wind power historical factor sequences and the wind power sequence, a correlation degree between the wind power historical factor sequence corresponding to each of the relevant factors and the wind power sequence is determined; a target wind power historical factor sequence with a correlation degree greater than a preset first threshold is screened out, and the target wind power historical factor sequence and the wind power sequence are used as the historical wind power generation data.

6. The method according to claim 1, characterized in that Before performing photovoltaic power prediction based on historical photovoltaic power generation data and the photovoltaic power prediction model, the method further includes: determining candidate factors related to the photovoltaic power prediction, the candidate factors comprising wind speed, wind direction, temperature, humidity, air pressure, and irradiance; Obtaining a photovoltaic historical factor sequence and a corresponding photovoltaic power generation power sequence corresponding to each candidate factor; Determining the correlation coefficient between the photovoltaic historical factor sequence corresponding to each candidate factor and the photovoltaic power sequence based on Spearman correlation analysis; The target photovoltaic historical factor sequence having a correlation coefficient greater than a preset second threshold is screened out, and the target photovoltaic historical factor sequence and the photovoltaic power generation power sequence are used as the historical photovoltaic power generation data.

7. The method according to claim 1, characterized in that Determining a reference number of operating electrolyzers within the prediction period based on the predicted wind power and the predicted photovoltaic power includes: Determining the total generated power within the prediction period based on the predicted wind power and the predicted photovoltaic power; Based on the change in the total generated power at each moment within the forecast period, screening out a continuous time period in which the total generated power fluctuates around a fixed value, and using the continuous time period as a section to be optimized; Taking the average value of the total generated power of the section to be optimized as the average power of each electrolytic cell at each moment in the section to be optimized; Based on the ratio of the average power to the rated power of each electrolytic cell, a reference number of electrolytic cells operating in the section to be optimized is determined.

8. The method according to claim 7, characterized in that The determining of the total generated power within the prediction period based on the predicted wind power and the predicted photovoltaic power includes: Performing first-order difference processing on the predicted wind power and the predicted photovoltaic power to obtain difference results at each moment in a prediction period; Accumulating and summing the difference results at each moment to obtain the total generated power within the prediction period; The first-order difference processing is performed according to the following formula: Δy t =y wind (t+1)+y PV (t+1)-y wind (t)-y PV (t) Where Δy t represents the predicted power generation at time t, y wind (t) represents the predicted wind power at time t, y PV (t) represents the predicted photovoltaic power generation at time t, y wind (t+1) represents the predicted wind power at time t+1, y PV (t+1) represents the predicted photovoltaic power generation at time t+1.

9. The method according to claim 1, characterized in that The determining of the total hydrogen production power based on the predicted wind power and the predicted photovoltaic power using a power optimization model that considers system cost includes: Considering the operation and maintenance costs of the electrolyzer, hydrogen storage tank, wind and solar curtailment penalty costs, and grid electricity purchase costs, an objective optimization function is constructed with the goal of minimizing system costs. Determining the constraints of the objective optimization function based on power balance constraints, operational safety constraints of the multi-stack electrolyzers, start and stop times constraints of the hydrogen storage tanks, and state of charge range constraints of the hydrogen storage tanks; Based on the objective optimization function and the constraint conditions, obtaining the power optimization model; Based on the predicted wind power, the predicted photovoltaic power, the electrolyzer operation and maintenance cost, and the hydrogen storage tank operation and maintenance cost, the power optimization model is solved to obtain the total hydrogen production power.

10. The method according to claim 9, characterized in that The formula of the objective optimization function is as follows: Among them, C total represents the total economic cost of the system, C CO represents the electrolyzer operation and maintenance cost, C HO represents the operation and maintenance cost of the hydrogen storage tank, C Wind represents the penalty cost for wind curtailment, C PV represents the penalty cost of abandoning light, C EL represents the cost of purchasing electricity from the grid, represents the hydrogen sales revenue, N is the total number of electrolyzers in the system, β EO (s) represents the operation and maintenance cost coefficient of electrolytic cell s, represents the rated capacity of electrolytic cell s, c EO (s) represents the construction cost per unit capacity of electrolytic cell s, β HO represents the operation and maintenance cost coefficient of the hydrogen storage tank l, c HO represents the unit capacity construction cost of hydrogen storage tank l, Indicates the capacity of the hydrogen storage tank l, c Wind represents the penalty cost per unit of wind curtailment, represents the amount of wind power abandoned at time t, T represents the duration of the forecast period, c PV represents the unit abandonment penalty cost, represents the amount of abandoned solar power at time t, c grid (t) represents the time-of-use electricity price at time t, P grid (t) represents the purchased power at time t, τ represents the control step size, represents the hydrogen price per unit mass at time t, represents the hydrogen production of electrolyzer s at time t; The formula for the power balance constraint is as follows: Among them, P grid (t) represents the amount of electricity purchased from the power grid at time t, P Wind (t) represents the predicted wind power at time t, P PV (t) represents the predicted photovoltaic power at time t, P L (t) represents the electrical load at time t, P EO (t) represents the total power of multiple electrolytic cells at time t, P EO,s (t) represents the power of electrolytic cell s at time t.

11. The method according to claim 7, characterized in that The operating status of the electrolyzer includes a health status of the electrolyzer, and the power allocation of the multiple electrolyzer stacks based on the operating status of each electrolyzer in the multiple electrolyzer stacks, the reference number, and the total hydrogen production power includes: sequentially numbering each of the electrolytic cells based on their health status; Determining the number of standby electrolytic cells based on the current number of actually operating electrolytic cells and the reference number; Power allocation of each electrolytic cell is performed based on the sequence number of the electrolytic cells and the total hydrogen production power.

12. The method according to claim 11, characterized in that The power allocation of each electrolyzer based on the sequence number of the electrolyzer and the total hydrogen production power includes: Dividing the power range into three ranges based on the upper power limit, rated power and lower power limit of each electrolytic cell; Based on the total hydrogen production power, determining a target power range corresponding to the section to be optimized; Power allocation of the electrolytic cells is performed based on a power allocation strategy corresponding to the target power interval and a sequence number of each of the electrolytic cells.

13. The method according to claim 12, characterized in that The power allocation of the electrolyzers based on the power allocation strategy corresponding to the target power range, the sequence number of each electrolyzer and the total hydrogen production power includes: For different power ranges, the number of operating electrolytic cells corresponding to each power range is determined according to the following formula: Among them, N f Indicates the number of electrolytic cells in operation, P el represents the total power of hydrogen production, N represents the total number of electrolyzers in the hydrogen production system, P m (s) represents the operating power of electrolytic cell s in the section m to be optimized, P min (s) represents the lower limit power of electrolytic cell s, P N (s) represents the rated power of electrolytic cell s, P max (s) represents the upper limit power of electrolytic cell s, and s represents the electrolytic cell number; If the total hydrogen production power is in the interval 0≤P el <N·P min (s), the power distribution of each electrolytic cell is carried out according to the following formula: If the total hydrogen production power is in the interval N·P min (s)≤P el <N·P N (s), the power distribution of each electrolytic cell is carried out according to the following formula: If the total hydrogen production power is in the interval N·P N (s)≤P el ≤N·P max (s), the power distribution of each electrolytic cell is carried out according to the following formula:

14. A multi-stack electrolyzer power distribution system based on wind power and photovoltaic prediction, characterized in that: include: A wind power prediction unit is used to predict wind power based on historical wind power generation data and a wind power prediction model to obtain predicted wind power within a prediction period; the wind power prediction model is a multivariate graph neural network based on an inverse residual structure; A photovoltaic prediction unit, configured to perform photovoltaic power prediction based on historical photovoltaic power generation data and a photovoltaic power prediction model, and obtain predicted photovoltaic power within the prediction period; an electrolyzer quantity determination unit, configured to determine a reference quantity of operating electrolyzers within the prediction period based on the predicted wind power and the predicted photovoltaic power; a total power determination unit, which determines the total hydrogen production power based on the predicted wind power and the predicted photovoltaic power using a power optimization model that takes system cost into consideration; A power distribution unit is used to distribute the power of the multiple electrolyzer stacks based on the operating status of each electrolyzer in the multiple electrolyzer stacks, the reference number and the total hydrogen production power.

15. The system according to claim 14, wherein: The wind power prediction unit includes: A graph learning module, configured to obtain an output adjacency matrix based on the input historical wind power generation data; A standard convolution module, configured to obtain output initial features based on the input historical wind power generation data; Multiple graph convolution modules, used to perform graph convolution processing based on an inverted residual structure on the initial features to obtain extracted features; wherein in the graph convolution processing process of each graph convolution module, graph convolution processing based on an inverted residual structure is performed based on the adjacency matrix, and the output of each graph convolution module is residually connected with the input and then used as the input of the next graph convolution module; a skip connection module, configured to perform feature connection on the historical wind power generation data, the intermediate results of each graph convolution module, and the extracted features to obtain connection features; The output module is used to perform convolution processing and nonlinear activation processing on the connection features through the output module to obtain the predicted wind power within the prediction period.

16. The system according to claim 15, wherein: Each of the graph convolution modules includes: The temporal convolution submodule is used to perform dilated convolution processing of different scales and channel dimension splicing on the input of the graph convolution module to obtain the intermediate results of the graph convolution module; A graph convolution submodule is used to perform multi-layer feature extraction and nonlinear transformation on the intermediate results and the adjacency matrix to obtain selected features; The inverted residual submodule is used to perform dimensionality-increasing convolution and depth-wise separable convolution on the selected features to obtain the output of the graph convolution module.

17. The system according to claim 14, wherein: The photovoltaic power prediction model is a long-term and short-term time series network, and the photovoltaic prediction unit is specifically used to determine the hyperparameters of the long-term and short-term time series network based on Bayesian and hyperband optimization algorithms.

18. The system according to claim 14, wherein: The wind power prediction unit includes: A relevant factor determination module is used to determine relevant factors for wind power prediction, wherein the relevant factors include wind speed, wind direction, temperature, humidity and air pressure at different heights; A wind power sequence acquisition module is used to obtain a wind power historical factor sequence and a corresponding wind power power sequence corresponding to each of the relevant factors; a correlation determination module, configured to determine the correlation between the wind power historical factor sequence corresponding to each of the relevant factors and the wind power sequence based on the cosine distance between each of the wind power historical factor sequence and the wind power sequence; The wind power data screening module is configured to screen out the target wind power historical factor sequence having a correlation greater than a preset first threshold, and use the target wind power historical factor sequence and the wind power sequence as the historical wind power generation data.

19. The system according to claim 14, wherein: The photovoltaic prediction unit includes: a candidate factor determination module, configured to determine candidate factors related to the photovoltaic power prediction, wherein the candidate factors include wind speed, wind direction, temperature, humidity, air pressure, and irradiation intensity; A photovoltaic sequence acquisition module is used to obtain the photovoltaic historical factor sequence and the corresponding photovoltaic power generation power sequence corresponding to each candidate factor; a correlation coefficient determination module, configured to determine, based on Spearman correlation analysis, a correlation coefficient between a photovoltaic historical factor sequence corresponding to each candidate factor and the photovoltaic power sequence; The photovoltaic data screening module is used to screen out the target photovoltaic historical factor sequence whose correlation coefficient is greater than a preset second threshold, and use the target photovoltaic historical factor sequence and the photovoltaic power generation power sequence as the historical photovoltaic power generation data.

20. The system according to claim 14, wherein: The electrolytic cell quantity determination unit includes: a total power determination module, configured to determine the total generated power within the prediction period based on the predicted wind power and the predicted photovoltaic power; an optimization section screening module, configured to screen out a continuous time period in which the total generated power fluctuates around a fixed value based on the change in the total generated power at each moment within the forecast period, and use the continuous time period as a section to be optimized; a power determination module, configured to take the average value of the total generated power of the section to be optimized as the average power of each electrolytic cell at each moment in the section to be optimized; The electrolytic cell data determination module is used to determine the reference number of operating electrolytic cells in the section to be optimized based on the ratio of the average power to the rated power of each electrolytic cell.

21. The system according to claim 20, wherein: The total power determination module includes: a difference submodule, configured to perform first-order difference processing on the predicted wind power and the predicted photovoltaic power to obtain difference results at each moment within a prediction period; A summing submodule, configured to accumulate and sum the differential results at each moment to obtain the total generated power within the prediction period; The first-order difference processing is performed according to the following formula: Δy t =y wind (t+1)+y PV (t+1)-y wind (t)-y PV (t) Where Δy t represents the predicted power generation at time t, y wind (t) represents the predicted wind power at time t, y PV (t) represents the predicted photovoltaic power generation at time t, y wind (t+1) represents the predicted wind power at time t+1, y PV (t+1) represents the predicted photovoltaic power generation at time t+1.

22. The system according to claim 14, wherein: The total power determination unit includes: An optimization function construction module is used to consider the operation and maintenance costs of the electrolyzer, the operation and maintenance costs of the hydrogen storage tank, the penalty costs for wind and solar power curtailment, and the grid's electricity purchase costs, and to construct a target optimization function with the goal of minimizing system costs; A constraint determination module is used to determine the constraint conditions of the objective optimization function based on power balance constraints, operation safety constraints of multiple electrolyzers, start and stop times constraints of the hydrogen storage tank, and state of charge range constraints of the hydrogen storage tank; An optimization model determination module, configured to obtain the power optimization model based on the target optimization function and the constraint conditions; A model solving module is used to solve the power optimization model based on the predicted wind power, the predicted photovoltaic power, the electrolyzer operation and maintenance costs, and the operation and maintenance costs of the hydrogen storage tank to obtain the total hydrogen production power.

23. The system according to claim 22, wherein: The formula of the objective optimization function is as follows: Among them, C total represents the total economic cost of the system, C CO represents the electrolyzer operation and maintenance cost, C HO represents the operation and maintenance cost of the hydrogen storage tank, C Wind represents the penalty cost for wind curtailment, C PV represents the penalty cost of abandoning light, C EL represents the cost of purchasing electricity from the grid, represents the hydrogen sales revenue, N is the total number of electrolyzers in the system, β EO (s) represents the operation and maintenance cost coefficient of electrolytic cell s, represents the rated capacity of electrolytic cell s, c EO (s) represents the construction cost per unit capacity of electrolytic cell s, β HO represents the operation and maintenance cost coefficient of the hydrogen storage tank l, c HO represents the unit capacity construction cost of hydrogen storage tank l, Indicates the capacity of the hydrogen storage tank l, c Wind represents the penalty cost per unit of wind curtailment, represents the amount of wind power abandoned at time t, T represents the duration of the forecast period, c PV represents the unit abandonment penalty cost, represents the amount of abandoned solar power at time t, c grid (t) represents the time-of-use electricity price at time t, P grid (t) represents the purchased power at time t, τ represents the control step size, represents the hydrogen price per unit mass at time t, represents the hydrogen production of electrolyzer s at time t; The formula for the power balance constraint is as follows: Among them, P grid (t) represents the amount of electricity purchased from the power grid at time t, P Wind (t) represents the predicted wind power at time t, P PV (t) represents the predicted photovoltaic power at time t, P L (t) represents the electrical load at time t, P EO (t) represents the total power of multiple electrolytic cells at time t, P EO,s (t) represents the power of electrolytic cell s at time t.

24. The system according to claim 20, wherein: The operating state of the electrolyzer includes the health state of the electrolyzer, and the power distribution unit includes: a sorting module, configured to sequentially number each of the electrolytic cells based on a health status of each of the electrolytic cells; A standby quantity determination module, configured to determine the number of standby electrolytic cells based on the current number of actually operating electrolytic cells and the reference number; A power allocation module is used to allocate power to each of the electrolyzers based on the sequence number of the electrolyzers and the total hydrogen production power.

25. The system according to claim 24, wherein: The power distribution module includes: A partitioning submodule, configured to divide the power into three power intervals based on the upper power limit, rated power and lower power limit of each electrolyzer; An interval determination submodule, configured to determine a target power interval corresponding to the section to be optimized based on the total hydrogen production power; The power allocation submodule is used to allocate power to the electrolytic cells based on the power allocation strategy corresponding to the target power interval and the sequential numbering of each electrolytic cell.

26. The system according to claim 25, wherein: The power distribution submodule is specifically used for: For different power ranges, the number of operating electrolytic cells corresponding to each power range is determined according to the following formula: Among them, N f Indicates the number of electrolytic cells in operation, P el represents the total power of hydrogen production, N represents the total number of electrolyzers in the hydrogen production system, P m (s) represents the operating power of electrolytic cell s in the section m to be optimized, P min (s) represents the lower limit power of electrolytic cell s, P N (s) represents the rated power of electrolytic cell s, P max (s) represents the upper limit power of electrolytic cell s, and s represents the electrolytic cell number; If the total hydrogen production power is in the interval 0≤P el <N·P min (s), the power distribution of each electrolytic cell is carried out according to the following formula: If the total hydrogen production power is in the interval N·P min (s)≤P el <N·P N (s), the power distribution of each electrolytic cell is carried out according to the following formula: If the total hydrogen production power is in the interval N·P N (s)≤P el ≤N·P max (s), the power distribution of each electrolytic cell is carried out according to the following formula:

27. An electronic device, characterized in that: include: at least one processor and memory; The memory and the processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, the multi-stack electrolyzer power allocation method based on wind power and photovoltaic prediction according to any one of claims 1 to 13 is implemented.

28. A readable storage medium, characterized in that An execution program is stored thereon, and when the execution program is executed, the multi-stack electrolyzer power allocation method based on wind power and photovoltaic prediction as described in any one of claims 1 to 13 is implemented.

Images