Method and device for determining off-grid power of photovoltaic coupling electrolytic hydrogen production system

Abstract

The invention relates to the technical field of photovoltaic coupling electrolytic hydrogen production, and particularly discloses a grid-off power determination method and device for a photovoltaic coupling electrolytic hydrogen production system, and the method comprises the steps: obtaining the operation data of the photovoltaic coupling electrolytic hydrogen production system in a current time period; processing the operation data by using a target off-grid power determination model so as to determine the target off-grid power of the photovoltaic coupling electrolytic hydrogen production system; the target offline power determination model comprises a space-time convolutional neural network and a long short-term memory neural network; the space-time convolutional neural network is used for performing feature extraction on the operation data in the current time period to obtain a time sequence feature, a space feature and a coupling relation feature among multi-source data of the operation data; and the long-short-term memory neural network is used for performing time sequence processing on the time sequence characteristics, the spatial characteristics and the coupling relation characteristics among the multiple far data to determine the target off-network power. According to the method, the prediction accuracy of the off-network power can be improved.

Description

Technical Field

[0001] This specification relates to the technical field of photovoltaic-coupled electrolytic hydrogen production, and particularly to a method and device for determining the power fed into the grid for a photovoltaic-coupled electrolytic hydrogen production system. Background Art

[0002] New energy sources represented by photovoltaic power are rich, clean, and highly efficient, and can be used by humans permanently. However, their inherent volatility and intermittency pose more challenges to the power grid. When peak shaving is difficult, it is necessary to curtail photovoltaic power to maintain the safety and stability of the power grid. Electrolytic water hydrogen production consumes a large amount of electric energy, resulting in a high hydrogen production cost. Using the surplus electric energy generated by curtailed photovoltaic power for hydrogen production not only reduces the hydrogen production cost but also reduces carbon emissions during the production process, and at the same time solves the problem of energy waste.

[0003] A photovoltaic-coupled electrolytic hydrogen production system requires an efficient energy management system to ensure system efficiency. In the energy management system, the prediction of the power fed into the grid has strong guiding significance for system decision-making. There are many influencing factors and complex mechanisms for the power fed into the grid. The accuracy of the existing power fed into the grid prediction methods is not high enough to meet the actual application requirements.

[0004] In response to the above problems, no effective solution has been proposed yet. Summary of the Invention

[0005] Embodiments of this specification provide a method and device for determining the power fed into the grid for a photovoltaic-coupled electrolytic hydrogen production system to solve the problem of low accuracy in predicting the power fed into the grid in the prior art.

[0006] Embodiments of this specification provide a method for determining the power fed into the grid for a photovoltaic-coupled electrolytic hydrogen production system, including:

[0007] Obtain the operation data of the photovoltaic-coupled electrolytic hydrogen production system in the current time period, where the operation data includes: photovoltaic power generation data, electrolyzer operation parameters, hydrogen energy storage system status data, grid interaction data, meteorological factor data, and time data;

[0008] Use the target power fed into the grid determination model to process the operation data in the current time period to determine the target power fed into the grid at one or more moments in the future time period for the photovoltaic-coupled electrolytic hydrogen production system;

[0009] Among them, the target power grid connection power determination model includes a spatio-temporal convolutional neural network and a long short-term memory neural network; the spatio-temporal convolutional neural network is used to extract features from the operation data within the current time period to obtain the temporal features, spatial features, and coupling relationship features between multi-source data of the operation data; the long short-term memory neural network is used to perform temporal processing on the temporal features, spatial features, and coupling relationship features between multi-source data output by the spatio-temporal convolutional neural network to determine the target power grid connection power of the photovoltaic-coupled electrolytic hydrogen production system at one or more moments in the future time period.

[0010] In one embodiment, the spatio-temporal convolutional neural network includes: a time dimension convolutional layer, a spatial dimension convolutional layer, an environment and device coupling convolutional layer, and a pooling layer; the time dimension convolutional layer is used to extract features of input data at different time scales and short-term fluctuation patterns; the spatial dimension convolutional layer is used to process the spatial distribution data of multiple meteorological stations to capture the spatial difference features of the output of the photovoltaic array; the environment and device coupling convolutional layer is used to calculate the correlation weight between meteorological factor data and electrolyzer operation parameters through a cross-attention layer.

[0011] In one embodiment, the long short-term memory neural network includes an adaptive gated long short-term memory neural network; the adaptive gated long short-term memory neural network is used to dynamically adjust the weights of the input gate, forget gate, and / or output gate according to the input data.

[0012] In one embodiment, a hydrogen storage feedback link is introduced into the long short-term memory neural network, and the hydrogen storage feedback link is used to splice the hydrogen storage system state data with the hidden state and input features of the long short-term memory neural network to form an extended input vector to capture the influence features of the hydrogen storage system state data on the power grid connection power.

[0013] In one embodiment, the target power grid connection power determination model is constructed in the following manner:

[0014] Obtain the power grid connection power data and historical operation data of the photovoltaic-coupled electrolytic hydrogen production system in the historical time period; the historical operation data includes: historical photovoltaic power generation data, historical electrolyzer operation parameters, historical hydrogen storage system state data, historical grid interaction data, historical meteorological factor data, and historical time data;

[0015] Preprocess the power grid connection power data and the historical operation data to obtain a training sample set and a label set; the training sample set includes the preprocessed historical operation data, and the label set includes the power grid connection power data corresponding to each sample in multiple samples in the training sample set;

[0016] Training a preset model using the training sample set to obtain a target off-grid power determination model; the target off-grid power determination model is used to determine the off-grid power at one or more target moments in a future time period of the photovoltaic-coupled electrolytic hydrogen production system.

[0017] In one embodiment, preprocessing the off-grid power data and the historical operation data to obtain a training sample set and a label set, including:

[0018] Removing outliers from the off-grid power data and the historical operation data;

[0019] Performing interpolation filling and normalization processing on the historical operation data;

[0020] Aligning the processed off-grid power data and historical operation data according to timestamps to obtain a training sample set and a label set.

[0021] The embodiments of this specification also provide an off-grid power determination device for a photovoltaic-coupled electrolytic hydrogen production system, including:

[0022] An acquisition module, configured to acquire the operation data of the photovoltaic-coupled electrolytic hydrogen production system in the current time period, where the operation data includes: photovoltaic power generation data, electrolyzer operation parameters, hydrogen energy storage system state data, grid interaction data, meteorological factor data, and time data;

[0023] A determination module, configured to process the operation data in the current time period by using the target off-grid power determination model to determine the target off-grid power at one or more moments in a future time period of the photovoltaic-coupled electrolytic hydrogen production system;

[0024] Wherein, the target off-grid power determination model includes a spatio-temporal convolutional neural network and a long short-term memory neural network; the spatio-temporal convolutional neural network is used to extract features from the operation data in the current time period to obtain the temporal features, spatial features, and coupling relationship features between multi-source data of the operation data; the long short-term memory neural network is used to perform temporal processing on the temporal features, spatial features, and coupling relationship features between multi-source data output by the spatio-temporal convolutional neural network to determine the target off-grid power at one or more moments in a future time period of the photovoltaic-coupled electrolytic hydrogen production system.

[0025] The embodiments of this specification also provide a computer device, including a processor and a memory for storing processor-executable instructions, and when the processor executes the instructions, the steps of the off-grid power determination method for the photovoltaic-coupled electrolytic hydrogen production system described in any of the above embodiments are implemented.

[0026] The embodiments of this specification also provide a computer-readable storage medium, on which computer instructions are stored, and when the instructions are executed, the steps of the method for determining the power fed into the grid for the photovoltaic-coupled electrolytic hydrogen production system described in any of the above embodiments are implemented.

[0027] In the embodiments of this specification, a method for determining the power fed into the grid for a photovoltaic-coupled electrolytic hydrogen production system is provided. The operating data of the photovoltaic-coupled electrolytic hydrogen production system in the current time period can be obtained. The operating data can include photovoltaic power generation data, electrolyzer operating parameters, hydrogen energy storage system status data, grid interaction data, meteorological factor data, and time data. The target power fed into the grid determination model can be used to process the operating data in the current time period to determine the target power fed into the grid at one or more moments in the future time period for the photovoltaic-coupled electrolytic hydrogen production system. The target power fed into the grid determination model includes a spatio-temporal convolutional neural network and a long short-term memory neural network. The spatio-temporal convolutional neural network can extract features from the operating data in the current time period to obtain the temporal features, spatial features, and coupling relationship features between multi-source data of the operating data. The long short-term memory neural network can perform temporal processing on the temporal features, spatial features, and coupling relationship features between multi-source data output by the spatio-temporal convolutional neural network to determine the target power fed into the grid at one or more moments in the future time period for the photovoltaic-coupled electrolytic hydrogen production system. In the above solution, through the spatio-temporal convolutional neural network, the spatio-temporal features and the coupling relationship features between multi-source data of various operating data including photovoltaic power generation data and meteorological data can be captured simultaneously, improving the model's processing ability for complex non-linear relationships; the long short-term memory neural network performs well on temporal data; by integrating the spatio-temporal convolutional neural network and the long short-term memory neural network together, the features of different scales of the factors affecting the power fed into the grid can be captured simultaneously, improving the accuracy and robustness of the power fed into the grid of the photovoltaic-coupled electrolytic hydrogen production system, and also reducing the overfitting risk of a single model and improving the generalization performance of the model. Description of the Drawings

[0028] The drawings described herein are used to provide a further understanding of this specification, form a part of this specification, and do not limit this specification. In the drawings:

[0029] Figure 1 Shows the flowchart of the method for determining the power fed into the grid for the photovoltaic-coupled electrolytic hydrogen production system in an embodiment of this specification;

[0030] Figure 2 Shows the flowchart of the method for determining the power fed into the grid for the photovoltaic-coupled electrolytic hydrogen production system in an embodiment of this specification;

[0031] Figure 3 Shows the flowchart of constructing the target power fed into the grid determination model in an embodiment of this specification;

[0032] Figure 4 Shows a schematic diagram of a power grid connection power determination device for a photovoltaic-coupled electrolytic hydrogen production system in an embodiment of this specification;

[0033] Figure 5 Shows a schematic diagram of a computer device in an embodiment of this specification. Detailed implementation manners

[0034] Hereinafter, the principles and spirit of this specification will be described with reference to several exemplary implementation manners. It should be understood that these implementation manners are provided only to enable those skilled in the art to better understand and then implement this specification, rather than limiting the scope of this specification in any way. On the contrary, these implementation manners are provided to make the disclosure of this specification more thorough and complete, and to be able to fully convey the scope of this disclosure to those skilled in the art.

[0035] Those skilled in the art know that the implementation manners of this specification can be implemented as a system, device, equipment, method, or computer program product. Therefore, the disclosure of this specification can be specifically implemented in the following forms, namely: completely hardware, completely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0036] An embodiment of this specification provides a method for determining the power grid connection power of a photovoltaic-coupled electrolytic hydrogen production system. Figure 1 Shows a flowchart of a method for determining the power grid connection power of a photovoltaic-coupled electrolytic hydrogen production system in an embodiment of this specification. Although this specification provides method operation steps or device structures as shown in the following embodiments or drawings, based on routine or non-creative labor, more or fewer operation steps or module units may be included in the method or device. In steps or structures where there is no necessary causal relationship logically, the execution order of these steps or the module structure of the device is not limited to the execution order or module structure described in the embodiments of this specification and shown in the drawings. When the method or module structure is applied to an actual device or terminal product, it can be executed sequentially or in parallel according to the method or module structure shown in the embodiments or drawings (for example, in a parallel processor or multi-threaded processing environment, or even a distributed processing environment).

[0037] Specifically, as Figure 1 shown, the method for determining the power grid connection power of a photovoltaic-coupled electrolytic hydrogen production system provided in an embodiment of this specification may include the following steps.

[0038] Step S101, obtain the operation data of the photovoltaic-coupled electrolytic hydrogen production system in the current time period, where the operation data includes: photovoltaic power generation data, electrolyzer operation parameters, hydrogen energy storage system state data, grid interaction data, meteorological factor data, and time data.

[0039] The method in this embodiment can be applied to a server. The operating data of the photovoltaic-coupled electrolytic hydrogen production system in the current time period can be obtained. The current time period refers to a preset time period starting from the current moment. Among them, the preset time period can be 1 day, 10 days, one month, one year, etc. For example, if the preset time period is 10 days and the current moment is February 17, 2025, the current time period can be the operating data during the period from February 8, 2025 to February 17, 2025. Another example, if the preset time period is 1 day and the current moment is February 17, 2025, the current time period can be the operating data during the period from February 16, 2025 to February 17, 2025. The operating data in the current time period can include the operating data at one or more moments in the current time period.

[0040] The operating data can include photovoltaic power generation data, electrolyzer operating parameters, hydrogen energy storage system status data, grid interaction data, meteorological factor data, and time data. Among them, the photovoltaic power generation data can include data such as photovoltaic power generation power, photovoltaic array temperature, and light intensity sequence. The electrolyzer operating parameters can include parameters such as electrolyzer operating current, electrolysis efficiency, and hydrogen production rate. The hydrogen energy storage system status data can include data such as hydrogen storage tank pressure and hydrogen demand. The grid interaction data can include data such as time-of-use electricity price, peak shaving instruction, and curtailment threshold. The meteorological factor data can include data such as temperature, humidity, wind speed, and cloud cover of multiple meteorological stations. The time data can include year, month, day, hour, minute-level timestamps, and holiday flags. In this embodiment, not only the operating data directly related to the off-grid power such as photovoltaic power generation data, grid interaction data, meteorological factor data, and time data are considered, but also the electrolyzer operating parameters and hydrogen energy storage system status parameters are considered. This is because in the photovoltaic-coupled electrolytic hydrogen production scenario, the electrolyzer operating parameters will directly affect the power consumption of the system, and thus affect the off-grid power. Moreover, when the pressure of the hydrogen storage tank is too high, it means that there is more hydrogen stored in the tank. At this time, to avoid safety risks caused by excessive pressure and prevent waste of resources due to excessive hydrogen storage, the system can correspondingly reduce the input power of the electrolyzer, thereby reducing the off-grid power. When the pressure of the hydrogen storage tank is too low, it indicates that the hydrogen inventory in the tank is insufficient and cannot meet the possible subsequent hydrogen demand. To quickly replenish hydrogen, the system can increase the operating power of the electrolyzer, increase the electric energy obtained from the grid, and accelerate the production of hydrogen, thereby increasing the off-grid power. It can be seen that there is a certain correlation between the electrolyzer operating parameters and the hydrogen energy storage system status parameters and the off-grid power. By obtaining the photovoltaic power generation data, electrolyzer operating parameters, hydrogen energy storage system status data, grid interaction data, meteorological factor data, and time data, and predicting the off-grid power based on these data, the prediction efficiency of the off-grid power can be further improved.

[0041] Step S102, use the target power grid connection power determination model to process the operation data within the current time period to determine the target power grid connection power of the photovoltaic-coupled electrolytic hydrogen production system at one or more moments within a future time period.

[0042] Specifically, after obtaining the operation data of the photovoltaic-coupled electrolytic hydrogen production system within the current time period, the target power grid connection power determination model can be used to process the operation data within the current time period to determine the target power grid connection power of the photovoltaic-coupled electrolytic hydrogen production system at one or more moments within a future time period.

[0043] The target power grid connection power determination model includes a spatio-temporal convolutional neural network and a long short-term memory neural network. The spatio-temporal convolutional neural network is used to extract features from the operation data within the current time period to obtain the temporal features, spatial features, and coupling relationship features between multi-source data of the operation data. The long short-term memory neural network is used to perform temporal processing on the temporal features, spatial features, and coupling relationship features between multi-source data output by the spatio-temporal convolutional neural network to determine the target power grid connection power of the photovoltaic-coupled electrolytic hydrogen production system at one or more moments within a future time period.

[0044] In one embodiment, the temporal features may include: time scale change features, short-term fluctuation patterns, and periodic features. The time scale change features reflect the change laws of the operation data at different time scales. Taking photovoltaic power generation data as an example, the power generation power will show different change trends at different time scales such as seconds, minutes, hours, and days. Within a short time scale, the power generation power may fluctuate instantaneously due to the rapid movement of clouds, resulting in a change in light intensity; while on a long time scale, such as within a day, the power generation power will show periodic changes with the rise and fall of the sun. The spatio-temporal convolutional neural network can capture these change features at different time scales.

[0045] The short-term fluctuation pattern refers to the ups and downs of the operation data within a short time. For example, the current of the electrolyzer may fluctuate up and down within a few minutes due to some short-term disturbances inside the device or small fluctuations in the external power grid. The spatio-temporal convolutional neural network can extract this short-term fluctuation pattern to help the model understand the immediate changes in the device operation state.

[0046] The periodic feature means that some operation data has obvious periodicity, such as temperature and light intensity in meteorological data. The temperature will rise and fall within a day, showing a certain periodicity; the light intensity changes gradually from sunrise to sunset with a one-day cycle. The spatio-temporal convolutional neural network can capture these periodic features for predicting the relevant data changes within a future time period.

[0047] In one embodiment, the spatial features may include the spatial distribution difference features of multiple meteorological stations and the spatial difference features of the power output of photovoltaic arrays. The spatial distribution difference features of multiple meteorological stations refer to the differences in meteorological data monitored by meteorological stations at different geographical locations, such as temperature, humidity, wind speed, cloud cover, etc. These difference features will affect the power generation capacity of photovoltaic arrays. The spatio-temporal convolutional neural network can process the spatial distribution data of multiple meteorological stations, can capture the meteorological differences between different stations, and the impact of these difference features on the power output of photovoltaic arrays. For example, the wind speed and light conditions of a meteorological station on the leeward slope of a mountain and a station on an open plain may be different, which may lead to differences in the power generation power of nearby photovoltaic arrays. The spatio-temporal convolutional neural network can extract such spatial difference features. The spatial difference features of the power output of photovoltaic arrays refer to that even in the same area, the power generation capacities of photovoltaic arrays at different positions will be different due to factors such as installation angles and shading conditions. The spatio-temporal convolutional neural network can capture the spatial distribution features of the power output of photovoltaic arrays, understand the actual power generation efficiency differences of each photovoltaic array, and provide a basis for accurately evaluating the power generation capacity of the entire photovoltaic power station.

[0048] In one embodiment, the coupling relationship features between multi-source data may include the association features between meteorological data and device parameters, the interaction features between different device parameters, and the connection between grid interaction signals and other data.

[0049] The association features between meteorological data and device parameters are used to characterize the coupling relationship features between the environment and the operating parameters of the electrolyzer. For example, the change in light intensity will affect the power generation power of photovoltaic power generation, and then affect the electrical energy input of the electrolyzer, resulting in changes in parameters such as the current and electrolysis efficiency of the electrolyzer. The spatio-temporal convolutional neural network can capture the influence relationship of such meteorological data on device parameters, as well as the degree of association at different time steps and spatial nodes.

[0050] The interaction between different device parameters can be the relationship of mutual influence between the operating parameters of the electrolyzer (such as current, electrolysis efficiency, hydrogen production rate), or the relationship of mutual influence between the operating parameters of the electrolyzer and the state parameters of the hydrogen energy storage system (such as the pressure of the hydrogen storage tank, hydrogen demand). The spatio-temporal convolutional neural network can mine the coupling relationship between these different device parameters. For example, when the hydrogen production rate increases, it may lead to an increase in the pressure of the hydrogen storage tank. The model can capture this dynamic change relationship between parameters to more accurately predict the operating state and the power fed into the grid of the system.

[0051] The connection between grid interaction signals and other data can be the correlation between grid interaction signals (such as time-of-use electricity price, peak shaving instruction, curtailment threshold) and photovoltaic power generation data, electrolyzer operation parameters. For example, the level of time-of-use electricity price will affect the operation strategy of the electrolyzer. If the electricity price is low, the operation power of the electrolyzer may be increased to produce more hydrogen; the peak shaving instruction will directly affect the power output of the system to the grid. The spatio-temporal convolutional neural network can extract the coupling relationship characteristics between these grid interaction signals and other multi-source data, enabling the model to better adapt to the operation requirements of the grid.

[0052] The long short-term memory neural network can perform temporal processing on the temporal features, spatial features, and coupling relationship features between multi-source data output by the spatio-temporal convolutional neural network to determine the target power output to the grid at one or more moments within a future time period for the photovoltaic-coupled electrolytic hydrogen production system.

[0053] In the above solution, through the spatio-temporal convolutional neural network, it is possible to simultaneously capture the spatio-temporal features and coupling relationship features between multi-source data of various operation data including photovoltaic power generation data and meteorological data, improving the model's processing ability for complex non-linear relationships; the long short-term memory neural network performs well on temporal data; by integrating the spatio-temporal convolutional neural network and the long short-term memory neural network together, it is possible to simultaneously capture the features of different scales of factors affecting the power output to the grid, improving the accuracy and robustness of the power output to the grid of the photovoltaic-coupled electrolytic hydrogen production system, and also being able to reduce the overfitting risk of a single model and improve the generalization performance of the model.

[0054] In some embodiments of this specification, the spatio-temporal convolutional neural network includes: a time dimension convolutional layer, a space dimension convolutional layer, an environment and device coupling convolutional layer, and a pooling layer; the time dimension convolutional layer is used to extract the features of input data at different time scales and short-term fluctuation patterns; the space dimension convolutional layer is used to process the spatial distribution data of multiple meteorological stations to capture the spatial difference features of the power output of the photovoltaic array; the environment and device coupling convolutional layer is used to calculate the correlation weights between meteorological factor data and electrolyzer operation parameters through a cross-attention layer.

[0055] Specifically, the spatio-temporal convolutional neural network mainly includes: a time dimension convolutional layer, a space dimension convolutional layer, an environment and device coupling convolutional layer, and a pooling layer. The input data of the target power output to the grid determination model can include photovoltaic power generation, photovoltaic array temperature, light intensity, electrolyzer current, electrolysis efficiency, hydrogen production rate, hydrogen storage tank pressure, time-of-use electricity price, peak shaving instruction, curtailment threshold, temperature, humidity, cloud cover of each meteorological station. The output of the target power output to the grid determination model is the power output of the system to the grid. The spatio-temporal convolutional neural network extracts temporal features, spatial features, and coupling relationships between multi-source data through time dimension convolution, space dimension convolution, and environment and device coupling convolution respectively. The specific construction steps are as follows.

[0056] The time - dimension convolutional layer can use a 1 - D convolutional kernel with a size of 3 and 32 convolutional kernels. It slides along the time axis with a stride of 1 and uses "same" padding to keep the length of the output sequence the same as the input. This convolutional layer is used to extract features of multi - dimensional quantities in the input at different time scales and short - term fluctuation patterns. "Same" is a mode of padding in the convolution operation. The purpose of the "same" padding mode is to make the length (in the time - dimension convolution) or size (in the space - dimension convolution) of the output feature map (or sequence) after the convolution operation the same as that of the input. The convolution formula is as follows:

[0057]

[0058] Among them, C t is the feature value at time step t obtained after operations such as convolution, w i is the convolutional kernel weight, w t+i-1 is the input data at time step t + i - 1, and b is the bias term. k is a positive integer. Here, since the 1 - D convolutional kernel used in the time - dimension convolutional layer has a size of 3, k = 3, indicating the upper limit of the number of terms in the summation. i = 1, 2, 3.

[0059] The space - dimension convolutional layer uses a 2 - D convolutional kernel with a size of 3×3 and 16 convolutional kernels. It slides along the time axis with a stride of 1 and also uses "same" padding. This convolutional layer is used to process the spatial distribution data of multiple meteorological stations and capture the spatial difference features of the output of the photovoltaic array. The convolution formula is as follows:

[0060]

[0061] Among them, C i,j represents the value of the output feature map at position (i, j), where i and j are the row index and column index respectively. w m,n is the convolutional kernel weight, x i+m-1,j+n-1 is the value of the input data at position (i + m - 1, j + n - 1). b is the bias term. k = 3 (because the convolutional kernel size is 3×3). m = 1, 2, 3; n = 1, 2, 3.

[0062] The environment - device coupling convolutional layer uses a cross - attention layer to calculate the correlation weights between meteorological data and electrolyzer parameters. This layer is a fully - connected network and uses the softmax function to normalize the attention weight α i,j . The cross - attention weight formula is:

[0063]

[0064] Among them, α i,jDenotes the attention weight between the i-th time step or spatial node of meteorological data and the electrolyzer parameters at the j-th time step. N is the total sequence length of the electrolyzer parameters. e i,j Is the correlation score between meteorological data and electrolyzer parameters. The correlation score is calculated by the dot product of the query and the key, reflecting the correlation between meteorological features and equipment parameters.

[0065]

[0066] Among them, Q i Is the query vector of meteorological data at the i-th time step or spatial node; K j T Is the key vector of the electrolyzer parameters at the j-th time step; d k Is a scaling factor used to prevent the dot product value from being too large and causing gradient instability.

[0067] The activation function selection of the model: ReLU(x) = max(0, x). The coupling relationship between the modeling environment and the equipment is shown through the attention weight, improving the model's ability to capture complex non-linear relationships.

[0068] The output of the spatio-temporal convolutional neural network model is a one-dimensional vector, serving as the input of the long short-term memory neural network.

[0069] In some embodiments of this specification, the long short-term memory neural network includes an adaptive gated long short-term memory neural network; the adaptive gated long short-term memory neural network is used to dynamically adjust the weights of the input gate, forget gate, and / or output gate according to the input data.

[0070] Specifically, the adaptive gated long short-term memory neural network is used to extract temporal features. Based on the traditional long short-term memory neural network, an adaptive gating mechanism is introduced, which can dynamically adjust the weights of the input gate, forget gate, and / or output gate according to the input data. In one embodiment, an adaptive forget gate can be introduced to enhance the model's ability to capture complex temporal data. The structure of the adaptive gated long short-term memory neural network is described as follows.

[0071] Input gate:

[0072]

[0073] The input gate determines how much feature information can enter the internal memory state of the cell.

[0074] Adaptive forget gate:

[0075]

[0076] The forget gate determines which information in the previous internal memory state can be forgotten. is a learnable coefficient used to adjust the influence weight of the electrolyzer state, S t is the hydrogen production rate. α×S t The addition of makes the forget gate adaptively and dynamically adjust the forget gate weight when the electrolyzer state changes (i.e., when S t changes), determining whether to retain or forget the memory state of the previous moment, thus affecting the selective forgetting mechanism of the LSTM. When the hydrogen production rate increases significantly, the model tends to retain more historical information to capture long-term trends.

[0077] Candidate values:

[0078]

[0079] Update the internal memory state:

[0080]

[0081] The formula for updating the internal memory state, which combines the input gate, forget gate, and new candidate values, is used to update the internal memory state.

[0082] Output gate:

[0083]

[0084] The activation function of this model selects the sigmoid function, and the formula is:

[0085]

[0086] In the above formula:

[0087] x t is the input at the current time step t.

[0088] h t-1 is the output state at the previous time step t-1.

[0089] i t , f t , o t are the output vectors of the input gate, forget gate, and output gate.

[0090] C t-1 is the internal memory state at the previous time step t-1.

[0091] is the candidate value.

[0092] C t is the internal memory state at the current time step t.

[0093] h t is the output state at the current time step t.

[0094] W xi is the weight related to the input of the input gate at the current time step t and, W t-1 is the weight related to the output h at the previous time step t - 1 of the input gate.

[0095] W xf is the weight related to the input x of the adaptive forgetting gate at the current time step t t and, W hf is the weight related to the output h at the previous time step t - 1 of the adaptive forgetting gate. t-1

[0096] W xc is the weight related to the input x of the candidate value at the current time step t t and, W hc is the weight related to the output h at the previous time step t - 1 of the candidate value. t-1

[0097] W xo is the weight related to the input x of the output gate at the current time step t t and, W ho is the weight related to the output h at the previous time step t - 1 of the output gate. t-1

[0098] b i is the bias term corresponding to the input gate. b f is the bias term corresponding to the adaptive forgetting gate. b c is the bias term corresponding to the candidate value. b o is the bias term corresponding to the output gate.

[0099] In some embodiments of the present specification, a hydrogen storage feedback link is introduced into the long short - term memory neural network, and the hydrogen storage feedback link is used to splice the hydrogen energy storage system state data with the hidden state and input features of the long short - term memory neural network to form an extended input vector, so as to capture the influence characteristics of the hydrogen energy storage system state data on the off - grid power.

[0100] Specifically, a hydrogen storage feedback link is added to the long short - term memory neural network, and the hydrogen storage feedback link is used to splice the hydrogen energy storage system state data with the hidden state and input features of the long short - term memory neural network to form an extended input vector, so as to capture the influence characteristics of the hydrogen energy storage system state data on the off - grid power.

[0101] In one embodiment, the hydrogen storage tank pressure data can be spliced with the hidden state and input features in the long short - term memory neural network to form an extended input vector ​​​, with a dimension of the hidden state dimension + input feature dimension + 1, replaces the input vector x in each calculation step. t x t is the input at the current time step t. h t-1 is the output state at the previous time step t - 1. P t is the hydrogen storage tank pressure data at the current time step t.

[0102] By introducing the hydrogen storage tank pressure data, the memory state of the long short-term memory neural network can be dynamically adjusted to ensure that the model can respond in a timely manner to the state changes of the hydrogen energy storage system. When the hydrogen storage tank pressure changes (such as being too high or too low), the extended input vector will change accordingly, thereby affecting the calculation of the forget gate, input gate, and candidate memory state. In this way, when the hydrogen storage tank pressure is too high, the model tends to reduce the input of new information and retain more historical information to avoid excessive adjustment of the off-grid power; when the hydrogen storage tank pressure is too low, the model tends to increase the input of new information to quickly adjust the off-grid power to meet the requirements of the hydrogen energy storage system.

[0103] The hydrogen energy storage feedback link enables the model to better capture the impact of the hydrogen energy storage system on the off-grid power. In the power grid peak shaving scenario, the model can dynamically adjust the off-grid power according to the hydrogen storage tank pressure, avoiding prediction errors caused by sudden changes in the state of the hydrogen energy storage system and improving the prediction accuracy and robustness.

[0104] In some embodiments of this specification, the target off-grid power determination model is constructed in the following manner: obtaining the off-grid power data and historical operation data of the photovoltaic-coupled electrolytic hydrogen production system during a historical time period; the historical operation data includes: historical photovoltaic power generation data, historical electrolyzer operation parameters, historical hydrogen energy storage system state data, historical grid interaction data, historical meteorological factor data, and historical time data; preprocessing the off-grid power data and the historical operation data to obtain a training sample set and a label set; the training sample set includes the preprocessed historical operation data, and the label set includes the off-grid power data corresponding to each sample in multiple samples in the training sample set; using the training sample set to train a preset model to obtain a target off-grid power determination model; the target off-grid power determination model is used to determine the off-grid power at one or more target times in a future time period of the photovoltaic-coupled electrolytic hydrogen production system. Through the above method, a target off-grid power determination model can be trained.

[0105] In some embodiments of this specification, preprocessing the down-grid power data and the historical operation data to obtain a training sample set and a label set includes: removing outliers from the down-grid power data and the historical operation data; performing interpolation filling and normalization processing on the historical operation data; aligning the processed down-grid power data and historical operation data according to timestamps to obtain a training sample set and a label set.

[0106] Specifically, organizing the historical data of the down-grid power and influencing factors of the photovoltaic-coupled electrolytic hydrogen production system includes the following multi-source heterogeneous data.

[0107] Photovoltaic power generation data: power generation power, photovoltaic array temperature, light intensity sequence;

[0108] Electrolyzer operation parameters: current, electrolysis efficiency, hydrogen production rate;

[0109] Hydrogen energy storage system status: hydrogen storage tank pressure, hydrogen demand;

[0110] Grid interaction signals: time-of-use electricity price, peak shaving instruction, curtailment threshold;

[0111] Meteorological factor data: temperature, humidity, wind speed, cloud cover of multiple meteorological stations;

[0112] Time data: year, month, day, hour, minute-level timestamps and holiday flags.

[0113] Performing refined preprocessing on the sample set, and the specific steps are as follows:

[0114] Outlier processing:

[0115] Apply the 3σ criterion to stationary data (such as photovoltaic power generation), and remove data beyond the mean ± 3 times the standard deviation; for non-stationary data, combine the equipment operation threshold to remove data beyond the physically feasible range.

[0116] On the basis of removing outliers, different interpolation methods are used to fill the data respectively according to different data characteristics.

[0117] Time dimension interpolation: Cubic spline interpolation is used for meteorological data (such as light intensity) to retain the characteristics of time series continuity, smoothness and periodicity;

[0118] Spatial dimension interpolation: Kriging interpolation is used for multi-station meteorological data, etc., to fill missing values using spatial correlation;

[0119] In addition, linear interpolation is used for electrolyzer parameters to avoid introducing high-frequency noise and at the same time retain the smooth change characteristics of the equipment state.

[0120] Finally, the processed and complete historical data arranged in the time series of year, month, week, and day is obtained. The data is collected once every minute for a month, with a total of 43,200 data points at each moment.

[0121] Normalize the above processed historical data.

[0122] The formula for Min - Max normalization:

[0123]

[0124] Where: x' is the value after normalization. x is the value of the original data point. x min is the minimum value in the dataset. x max is the maximum value in the dataset.

[0125] Align the normalized data according to the timestamps, and further divide the processed data into datasets:

[0126] Training set: The continuous data in the first 80% of the time period, used for model training;

[0127] Validation set: The data in the middle 10% of the time period, used for hyperparameter tuning;

[0128] Test set: The data in the last 10% of the time period, used for final model evaluation.

[0129] The specific ratio is adjusted according to the collected data.

[0130] The mean squared error function is selected as the loss function, and the formula is:

[0131]

[0132] Where, y i is the true value of the off - grid power, is the predicted value of the off - grid power.

[0133] The Adam optimizer is adopted, with a batch size of 32 and 100 training epochs. The early stopping method is used. If the loss of the validation set does not decrease for 5 consecutive epochs, the training is terminated in advance. Finally, the target off - grid power determination model is obtained.

[0134] In some embodiments of this specification, the operating data includes time series data and environmental data; the time series data includes: photovoltaic power generation data, electrolyzer operating parameters, hydrogen energy storage system status data, grid interaction data, and time data; the environmental data includes meteorological data; using the target off-grid power determination model to process the operating data within the current time period to determine the target off-grid power of the photovoltaic-coupled electrolytic hydrogen production system at one or more moments within a future time period, including: performing wavelet transform on the time series data to decompose the time series data into multiple components corresponding to multiple frequencies, so as to extract the local features and trend features of the time series data; using the target off-grid power determination model to process the operating data within the current time period, as well as the local features and the trend features, to determine the target off-grid power of the photovoltaic-coupled electrolytic hydrogen production system at one or more moments within a future time period.

[0135] In this embodiment, before inputting the data into the target off-grid power determination model, wavelet transform is first performed on the time series data (such as the power generation power of a photovoltaic power station, the working current of an electrolyzer, etc.). Wavelet transform can decompose the time series into components of different frequencies and extract the local features and trend features of the data. For example, the power generation power time series is decomposed into high-frequency components and low-frequency components through discrete wavelet transform (DWT). The high-frequency components reflect the short-term fluctuations (local features) of the power generation power, and the low-frequency components reflect its long-term trend (trend features). Inputting these features after wavelet transform and the operating data into the target off-grid power determination model, the spatio-temporal convolutional neural network can further extract more advanced features on this basis, thereby improving the ability to extract the features of time series data and more accurately capturing the change law of the off-grid power.

[0136] In some embodiments of this specification, the target off-grid power determination model further includes a graph-structured neural network. The method further includes: constructing graph-structured data based on the association relationship between multiple devices in the photovoltaic-coupled electrolytic hydrogen production system and the historical operating data. The graph neural network learns the graph-structured data to obtain node features and graph-structured features. The long short-term memory model is used to perform temporal processing on the feature data output by the spatio-temporal convolutional neural network and the feature data output by the graph neural network to determine the target off-grid power at one or more moments within a future time period.

[0137] In this embodiment, considering the mutual relationships among various components (photovoltaic power station, electrolyzer, power grid, etc.) in the photovoltaic-coupled electrolytic hydrogen production system, a graph structure can be constructed to represent these relationships. Graph neural networks can process graph-structured data, learn the correlation features among components, and capture the global information of the system and the interactive effects among components. Graph neural networks can extract graph-structured features, such as the energy transfer paths between different devices and the collaborative working modes between devices. These graph-structured features play an important role in understanding the overall operation mechanism of the system and predicting the power fed into the grid, because the power fed into the grid depends not only on the operating states of individual devices but also on the mutual relationships between devices.

[0138] In some embodiments of this specification, the multiple devices include: a photovoltaic power station, a photovoltaic-coupled electrolytic hydrogen production device, and a power grid; based on the correlation relationships among the multiple devices in the photovoltaic-coupled electrolytic hydrogen production system and the historical operation data, graph-structured data is constructed, including: using the photovoltaic power generation data of the photovoltaic power station, the power grid dispatching data of the power grid, and the operation parameter data of the photovoltaic-coupled electrolytic hydrogen production device as the nodes of the graph-structured data; based on the physical connection relationships, energy flow relationships, and information interaction relationships among the photovoltaic power station, the photovoltaic-coupled electrolytic hydrogen production device, and the power grid, edges are constructed among the multiple nodes to obtain graph-structured data. In this way, graph-structured data can be constructed.

[0139] Each embodiment in this specification is described in a progressive manner. The same or similar parts among the embodiments can be referred to each other, and the key points of each embodiment are the differences from other embodiments. Specifically, reference can be made to the descriptions of the relevant processing-related embodiments above, and details will not be repeated here.

[0140] The specific embodiments of this specification are described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be executed in a different order from that in the embodiments and still achieve the desired results. Additionally, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0141] The above method will be described below in conjunction with a specific embodiment. However, it should be noted that this specific embodiment is only for better explaining this specification and does not constitute an improper limitation to this specification.

[0142] In this specific embodiment, a method for determining the power fed into the grid for a photovoltaic-coupled electrolytic hydrogen production system is proposed. Please refer to Figure 2, which shows a flow chart of a method for determining off-grid power for a photovoltaic coupled electrolysis hydrogen production system in this specific embodiment. Please refer to Figure 3 , shows a flow chart of constructing a target off-grid power determination model in this specific embodiment. Figure 2 and Figure 3 As shown, the method in this specific embodiment includes the following contents.

[0143] Step 1: Collect the historical operation data of the off-grid power of the photovoltaic coupled electrolysis hydrogen production system and its influencing factors, and establish a feature database.

[0144] The historical data of off-grid power and influencing factors of photovoltaic coupled electrolysis hydrogen production system are collated, including the following multi-source heterogeneous data:

[0145] Photovoltaic power generation data: power generation, photovoltaic array temperature, and light intensity sequence;

[0146] Electrolyzer operating parameters: current, electrolysis efficiency, hydrogen production rate;

[0147] Hydrogen energy storage system status: hydrogen storage tank pressure, hydrogen demand;

[0148] Grid interaction signals: time-of-use electricity prices, peak load instructions, and solar curtailment thresholds;

[0149] Meteorological factor data: temperature, humidity, wind speed, and cloud cover at multiple meteorological stations;

[0150] Time data: year, month, day, hour, minute-level timestamps and holiday marks.

[0151] The sample set is preprocessed in a refined manner. The specific steps are as follows:

[0152] Outlier handling:

[0153] The 3σ criterion is applied to stable data (such as photovoltaic power generation) to eliminate data that exceeds ±3 times the standard deviation of the mean; for non-stationary data, the data that exceeds the physically feasible range is eliminated in combination with the equipment operation threshold.

[0154] (1.2) On the basis of eliminating outliers, different interpolation methods are used to fill in the data according to different data characteristics.

[0155] Temporal dimension interpolation: cubic spline interpolation is used for meteorological data (such as light intensity) to preserve the continuity, smoothness and periodicity characteristics of the time series;

[0156] Spatial dimension interpolation: Kriging interpolation is used for multi-station meteorological data, and spatial correlation is used to fill missing values;

[0157] In addition, linear interpolation is used for the electrolyzer parameters to avoid introducing high-frequency noise while retaining the smooth change characteristics of the equipment status.

[0158] Finally, the processed and complete historical data arranged in the time series of year, month, week, and day is obtained. The data is collected once every minute within a month, with a total of 43,200 data points at different moments.

[0159] Normalize the historical data processed above.

[0160] Formula for Min-Max normalization:

[0161]

[0162] Where: x is the value of the original data point, and x' is the value after normalization. x min is the minimum value in the dataset. x max is the maximum value in the dataset.

[0163] Align the normalized data according to the timestamps, and further divide the processed data into datasets:

[0164] Training set: Continuous data in the first 80% of the time period, used for model training;

[0165] Validation set: Data in the middle 10% of the time period, used for hyperparameter tuning;

[0166] Test set: Data in the last 10% of the time period, used for final model evaluation.

[0167] The specific ratio is adjusted according to the collected data.

[0168] Step 2: Construct a spatio-temporal convolutional neural network (ST-ConvNet).

[0169] The constructed spatio-temporal convolutional neural network mainly includes: a convolutional layer in the time dimension, a convolutional layer in the space dimension, a convolutional layer for the coupling of environment and equipment, and a pooling layer. The input signals include photovoltaic power generation, photovoltaic array temperature, light intensity, electrolyzer current, electrolysis efficiency, hydrogen production rate, hydrogen storage tank pressure, time-of-use electricity price, peak shaving instruction, curtailment threshold, temperature, humidity, and cloud cover of each meteorological station. The output is the power fed into the grid of the system. ST-ConvNet extracts temporal features, spatial features, and the coupling relationship between multi-source data through convolutional operations in the time dimension, space dimension, and coupling of environment and equipment respectively. The specific construction steps are as follows:

[0170] The time - dimension convolutional layer can use a 1 - D convolutional kernel with a size of 3 and 32 convolutional kernels. It slides along the time axis with a stride of 1 and uses "same" padding to keep the length of the output sequence the same as the input. This convolutional layer is used to extract features of multi - dimensional quantities in the input at different time scales and short - term fluctuation patterns. "Same" is a mode of padding in the convolution operation. The purpose of the "same" padding mode is to make the length (in time - dimension convolution) or size (in space - dimension convolution) of the output feature map (or sequence) after the convolution operation the same as that of the input. The convolution formula is as follows:

[0171]

[0172] Among them, C t is the feature value at time step t obtained after operations such as convolution, w i is the convolutional kernel weight, w t+i-1 is the input data at time step t + i - 1, and b is the bias term. k is a positive integer. Here, since the 1 - D convolutional kernel used in the time - dimension convolutional layer has a size of 3, k = 3, indicating the upper limit of the number of terms in the summation. i = 1, 2, 3.

[0173] The space - dimension convolutional layer uses a 2 - D convolutional kernel with a size of 3×3 and 16 convolutional kernels. It slides along the time axis with a stride of 1 and also uses "same" padding. This convolutional layer is used to process the spatial distribution data of multiple meteorological stations and capture the spatial difference features of the power output of the photovoltaic array. The convolution formula is as follows:

[0174]

[0175] Among them, C i,j represents the value of the output feature map at position (i, j), where i and j are the row index and column index respectively. w m,n is the convolutional kernel weight, x i+m-1,j+n-1 is the value of the input data at position (i + m - 1, j + n - 1). b is the bias term. k = 3 (because the convolutional kernel size is 3×3). m = 1, 2, 3; n = 1, 2, 3.

[0176] The environment - equipment coupling convolutional layer uses a cross - attention layer to calculate the correlation weights between meteorological data and electrolyzer parameters. This layer is a fully - connected network that normalizes the attention weight α i,j using the softmax function. The cross - attention weight formula is:

[0177]

[0178] Among them, α i,jDenotes the attention weight between the i-th time step or spatial node of meteorological data and the electrolyzer parameters at the j-th time step. N is the total sequence length of the electrolyzer parameters. e i,j Is the correlation score between meteorological data and electrolyzer parameters. The correlation score is calculated by the dot product of the query and the key, reflecting the correlation between meteorological features and equipment parameters.

[0179]

[0180] Among them, Q i Is the query vector of meteorological data at the i-th time step or spatial node; K j T Is the key vector of the electrolyzer parameters at the j-th time step; d k Is the scaling factor, used to prevent the dot product value from being too large and causing gradient instability.

[0181] The activation function selection of the model: ReLU(x) = max(0, x). The coupling relationship between the modeling environment and the equipment is shown through the attention weight, enhancing the model's ability to capture complex non-linear relationships.

[0182] The output of the ST-ConvNet model is a one-dimensional vector, serving as the input of the AG-LSTM network.

[0183] Step 3, construct an Adaptive Gated Long Short-Term Memory Neural Network (AG-LSTM).

[0184] The Adaptive Gated Long Short-Term Memory Neural Network (AG-LSTM) is used to extract temporal features and make dynamic adjustments in combination with the feedback signal of the hydrogen storage system. The mentioned AG-LSTM, based on the traditional LSTM, introduces an adaptive forget gate and a hydrogen storage feedback link to enhance the model's ability to capture complex temporal data and its adaptability to sudden changes in equipment status. The specific settings of AG-LSTM are as follows:

[0185] The structure of AG-LSTM is described as follows.

[0186] Input gate:

[0187]

[0188] The input gate determines how much feature information can enter the internal memory state of the cell.

[0189] Adaptive forget gate:

[0190]

[0191] The forget gate determines which information in the previous internal memory state can be forgotten. The addition of makes the forgetting gate adaptively and dynamically adjust the forgetting gate weight when the electrolyzer state changes, determining whether to retain or forget the memory state at the previous moment. When the hydrogen production rate increases significantly, the model tends to retain more historical information to capture long-term trends.

[0192] Candidate values:

[0193]

[0194] Update the internal memory state:

[0195]

[0196] The formula for updating the internal memory state, combining the input gate, forgetting gate, and new candidate values, is used to update the internal memory state.

[0197] Output gate:

[0198]

[0199] The activation function of this model selects the sigmoid function, and the formula is:

[0200] .

[0201] In the above formula:

[0202] x t is the input at the current time step t.

[0203] h t-1 is the output state at the previous time step t-1.

[0204] i t , f t , o t are the output vectors of the input gate, forgetting gate, and output gate.

[0205] C t-1 is the internal memory state at the previous time step t-1.

[0206] is the candidate value.

[0207] C t is the internal memory state at the current time step t.

[0208] h t is the output state at the current time step t.

[0209] W xi is the weight related to the input of the input gate at the current time step t and is the weight related to the output h at the previous time step t-1 of the input gatet-1 The relevant weights.

[0210] W xf Is the weight related to the input x at the current time step t of the adaptive forget gate t W hf Is the weight related to the output h at the previous time step t - 1 of the adaptive forget gate t-1 The relevant weights.

[0211] W xc Is the weight related to the input x at the current time step t of the candidate value t W hc Is the weight related to the output h at the previous time step t - 1 of the candidate value t-1 The relevant weights.

[0212] W xo Is the weight related to the input x at the current time step t of the output gate t W ho Is the weight related to the output h at the previous time step t - 1 of the output gate t-1 The relevant weights.

[0213] b i Is the bias term corresponding to the input gate. b f Is the bias term corresponding to the adaptive forget gate. b c Is the bias term corresponding to the candidate value. b o Is the bias term corresponding to the output gate.

[0214] (3.2) Hydrogen storage feedback link in AG-LSTM

[0215] In addition, a hydrogen storage feedback link is added to the network model, and the hydrogen storage tank pressure data is concatenated with the hidden state and input features of the LSTM to form an extended input vector , whose dimension is the hidden state dimension + input feature dimension + 1, and replaces the input vector x in each calculation link t . x t Is the input at the current time step t. h t-1 Is the output state at the previous time step t - 1. P t Is the hydrogen storage tank pressure data at the current time step t.

[0216] By introducing the pressure data of the hydrogen storage tank, the memory state of AG-LSTM can be dynamically adjusted to ensure that the model can respond in a timely manner to the state changes of the hydrogen energy storage system. When the pressure of the hydrogen storage tank changes (such as being too high or too low), the extended input vector will change accordingly, which in turn affects the calculation of the forget gate, input gate, and candidate memory state. In this way, when the pressure of the hydrogen storage tank is too high, the model tends to reduce the input of new information and retain more historical information to avoid excessive adjustment of the power output to the grid; when the pressure of the hydrogen storage tank is too low, the model tends to increase the input of new information to quickly adjust the power output to the grid to meet the requirements of the hydrogen energy storage system.

[0217] The hydrogen energy storage feedback link enables the model to better capture the impact of the hydrogen energy storage system on the power output to the grid. In the grid peak shaving scenario, the model can dynamically adjust the power output to the grid according to the pressure of the hydrogen storage tank, avoiding prediction errors caused by sudden changes in the state of the hydrogen energy storage system and improving prediction accuracy and robustness.

[0218] Step 4, model training and verification.

[0219] The mean squared error function is selected as the loss function, and the formula is:

[0220]

[0221] where y i is the true value of the power output to the grid, is the predicted value of the power output to the grid.

[0222] The Adam optimizer is adopted, the batch size is 32, and the number of training epochs is 100. The early stopping method is used. If the loss on the validation set does not decrease for 5 consecutive epochs, the training is terminated in advance. Finally, the ST-ConvNet-AG-LSTM prediction model for the power output to the grid is obtained.

[0223] The method for determining the power output to the grid for the photovoltaic-coupled electrolytic hydrogen production system proposed in this specific embodiment has the following beneficial effects: (1) Spatiotemporal feature extraction: Through the spatiotemporal convolutional neural network, the spatiotemporal features of photovoltaic power generation data and meteorological data can be captured simultaneously, improving the model's ability to handle complex nonlinear relationships. (2) Adaptive gating mechanism: Through the adaptive forget gate and temporal attention mechanism, the model parameters can be dynamically adjusted to adapt to the changes in the electrolyzer state and grid scheduling. (3) Hydrogen energy storage feedback link: By introducing the feedback signal of the hydrogen energy storage system, the long-term dependence learning of the model can be effectively constrained, improving prediction accuracy. (4) Strong generalization performance: By integrating the spatiotemporal convolutional neural network and the adaptive gating LSTM, the overfitting risk of a single model can be reduced, improving the generalization performance of the model.

[0224] Based on the same inventive concept, embodiments of this specification also provide a device for determining the power fed into the grid for a photovoltaic-coupled electrolytic hydrogen production system, as described in the following embodiments. Since the principle of the device for determining the power fed into the grid for a photovoltaic-coupled electrolytic hydrogen production system to solve problems is similar to that of the method for determining the power fed into the grid for a photovoltaic-coupled electrolytic hydrogen production system, the implementation of the device for determining the power fed into the grid for a photovoltaic-coupled electrolytic hydrogen production system can refer to the implementation of the method for determining the power fed into the grid for a photovoltaic-coupled electrolytic hydrogen production system, and the repeated parts will not be elaborated. As used hereinafter, the term "unit" or "module" can be a combination of software and / or hardware that can achieve a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated. Figure 4 is a structural block diagram of the device for determining the power fed into the grid for a photovoltaic-coupled electrolytic hydrogen production system according to an embodiment of this specification, as Figure 4 shown, including: an acquisition module 401 and a determination module 402, which will be described below.

[0225] The acquisition module 401 is used to acquire the operation data of the photovoltaic-coupled electrolytic hydrogen production system during the current time period, and the operation data includes: photovoltaic power generation data, electrolyzer operation parameters, hydrogen energy storage system status data, grid interaction data, meteorological factor data, and time data.

[0226] The determination module 402 is used to process the operation data during the current time period by using a target power fed into the grid determination model to determine the target power fed into the grid at one or more moments in the future time period for the photovoltaic-coupled electrolytic hydrogen production system.

[0227] Among them, the target power fed into the grid determination model includes a spatio-temporal convolutional neural network and a long short-term memory neural network; the spatio-temporal convolutional neural network is used to extract features from the operation data during the current time period to obtain the temporal features, spatial features, and coupling relationship features between multi-source data of the operation data; the long short-term memory neural network is used to perform temporal processing on the temporal features, spatial features, and coupling relationship features between multi-source data output by the spatio-temporal convolutional neural network to determine the target power fed into the grid at one or more moments in the future time period for the photovoltaic-coupled electrolytic hydrogen production system.

[0228] In some embodiments of this specification, the spatio-temporal convolutional neural network includes: a temporal dimension convolutional layer, a spatial dimension convolutional layer, an environment and device coupling convolutional layer, and a pooling layer; the temporal dimension convolutional layer is used to extract features of input data at different time scales and short-term fluctuation patterns; the spatial dimension convolutional layer is used to process spatial distribution data of multiple meteorological stations to capture spatial difference features of the output of the photovoltaic array; the environment and device coupling convolutional layer is used to calculate the correlation weight between meteorological factor data and the operating parameters of the electrolyzer through a cross-attention layer.

[0229] In some embodiments of this specification, the long short-term memory neural network includes an adaptive gated long short-term memory neural network; the adaptive gated long short-term memory neural network is used to dynamically adjust the weights of the input gate, forget gate, and / or output gate according to the input data.

[0230] In some embodiments of this specification, a hydrogen storage feedback link is introduced into the long short-term memory neural network, and the hydrogen storage feedback link is used to splice the hydrogen storage system state data with the hidden state and input features of the long short-term memory neural network to form an extended input vector, so as to capture the influence features of the hydrogen storage system state data on the grid-connected power.

[0231] In some embodiments of this specification, the target grid-connected power determination model is constructed in the following manner: obtaining the grid-connected power data and historical operation data of the photovoltaic-coupled electrolytic hydrogen production system during a historical time period; the historical operation data includes: historical photovoltaic power generation data, historical electrolyzer operation parameters, historical hydrogen storage system state data, historical grid interaction data, historical meteorological factor data, and historical time data; preprocessing the grid-connected power data and the historical operation data to obtain a training sample set and a label set; the training sample set includes the preprocessed historical operation data, and the label set includes the grid-connected power data corresponding to each sample in multiple samples in the training sample set; using the training sample set to train a preset model to obtain a target grid-connected power determination model; the target grid-connected power determination model is used to determine the grid-connected power of the photovoltaic-coupled electrolytic hydrogen production system at one or more target moments in a future time period.

[0232] In some embodiments of this specification, preprocessing the grid-connected power data and the historical operation data to obtain a training sample set and a label set includes: removing outliers from the grid-connected power data and the historical operation data; performing interpolation filling and normalization processing on the historical operation data; aligning the processed grid-connected power data and historical operation data according to timestamps to obtain a training sample set and a label set.

[0233] The embodiments of this specification also provide a computer device, which can be specifically referred to Figure 5Schematic diagram of the composition structure of a computer device for the method of determining the power fed into the grid provided by the embodiments of this specification for a photovoltaic-coupled electrolytic hydrogen production system. Specifically, the computer device may include an input device 51, a processor 52, and a memory 53. Among them, the memory 53 is used to store instructions executable by the processor. When the processor 52 executes the instructions, it implements the steps of the method for determining the power fed into the grid for the photovoltaic-coupled electrolytic hydrogen production system described in any of the above embodiments.

[0234] In this embodiment, the input device may specifically be one of the main devices for information exchange between a user and a computer system. The input device may include a keyboard, a mouse, a camera, a scanner, a light pen, a handwriting input board, a voice input device, etc.; the input device is used to input raw data and the programs for processing these data into the computer. The input device may also acquire and receive data transmitted from other modules, units, and devices. The processor may be implemented in any suitable manner. For example, the processor may take the form of, for example, a microprocessor or a processor and a computer-readable medium storing computer-readable program code (such as software or firmware) executable by the (micro)processor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, etc. The memory may specifically be a memory device used to store information in modern information technology. The memory may include multiple levels. In a digital system, anything that can store binary data can be a memory; in an integrated circuit, a circuit with a storage function without a physical form is also called a memory, such as a RAM, a FIFO, etc.; in a system, a storage device with a physical form is also called a memory, such as a memory module, a TF card, etc.

[0235] In this embodiment, the functions and effects specifically implemented by this computer device may be explained in comparison with other embodiments and will not be elaborated here.

[0236] This specification also provides a computer storage medium based on the method for determining the power fed into the grid for a photovoltaic-coupled electrolytic hydrogen production system. The computer storage medium stores computer program instructions, and when the computer program instructions are executed, they implement the steps of the method for determining the power fed into the grid for the photovoltaic-coupled electrolytic hydrogen production system described in any of the above embodiments.

[0237] In this embodiment, the above storage medium includes, but is not limited to, random access memory (RAM), read-only memory (ROM), cache, hard disk drive (HDD), or memory card. The memory can be used to store computer program instructions. The network communication unit can be set according to the standards specified by the communication protocol and is used as an interface for network connection communication.

[0238] In this embodiment, the functions and effects specifically implemented by the program instructions stored in the computer storage medium can be explained by comparison with other embodiments and will not be elaborated here.

[0239] Obviously, those skilled in the art should understand that the above-mentioned modules or steps of the embodiments of this specification can be implemented by a general-purpose computing device. They can be concentrated on a single computing device or distributed on a network composed of multiple computing devices. Optionally, they can be implemented by program codes executable by the computing device, so that they can be stored in the storage device and executed by the computing device. And in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules respectively, or multiple modules or steps among them can be made into a single integrated circuit module to implement. Thus, the embodiments of this specification are not limited to any specific combination of hardware and software.

[0240] It should be understood that the above description is for illustrative purposes rather than for limitation. By reading the above description, many embodiments and many applications other than the provided examples will be obvious to those skilled in the art. Therefore, the scope of this specification should not be determined with reference to the above description, but should be determined with reference to the full scope of the foregoing claims and the equivalents of these claims.

[0241] The above are only the preferred embodiments of this specification and are not used to limit this specification. For those skilled in the art, the embodiments of this specification can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this specification shall be included within the protection scope of this specification.

Claims

1. A method for determining the off-grid power of a photovoltaic coupled electrolysis hydrogen production system, characterized in that: include: Obtaining the operating data of the photovoltaic coupled electrolysis hydrogen production system in the current time period, wherein the operating data includes: photovoltaic power generation data, electrolyzer operating parameters, hydrogen energy storage system status data, grid interaction data, meteorological factor data and time data; The operation data in the current time period is processed using a target off-grid power determination model to determine a target off-grid power of the photovoltaic-coupled electrolysis hydrogen production system at one or more moments in a future time period; Among them, the target off-grid power determination model includes a spatiotemporal convolutional neural network and a long short-term memory neural network; the spatiotemporal convolutional neural network is used to extract features of the operating data in the current time period to obtain the timing features, spatial features and coupling relationship features between multi-source data of the operating data; the long short-term memory neural network is used to perform time series processing on the timing features, spatial features and coupling relationship features between multi-source data output by the spatiotemporal convolutional neural network to determine the target off-grid power of the photovoltaic-coupled electrolysis hydrogen production system at one or more moments in the future time period.

2. The method for determining the off-grid power of the photovoltaic coupled electrolysis hydrogen production system according to claim 1, characterized in that: The spatiotemporal convolutional neural network includes: a time dimension convolution layer, a space dimension convolution layer, an environment and equipment coupling convolution layer and a pooling layer; the time dimension convolution layer is used to extract the characteristics and short-term fluctuation patterns of the input data at different time scales; the space dimension convolution layer is used to process the spatial distribution data of multiple meteorological stations to capture the spatial difference characteristics of the output of the photovoltaic array; the environment and equipment coupling convolution layer is used to calculate the association weight between the meteorological factor data and the electrolyzer operating parameters through the cross attention layer.

3. The method for determining the off-grid power for a photovoltaic coupled electrolysis hydrogen production system according to claim 1, characterized in that: The LSTM neural network includes an adaptive gated LSTM neural network; the adaptive gated LSTM neural network is used to dynamically adjust the weights of an input gate, a forget gate and / or an output gate according to input data.

4. The method for determining off-grid power for a photovoltaic coupled electrolysis hydrogen production system according to claim 1, characterized in that: A hydrogen storage feedback link is introduced into the long short-term memory neural network, and the hydrogen storage feedback link is used to splice the hydrogen energy storage system state data with the hidden state and input features of the long short-term memory neural network to form an extended input vector to capture the impact characteristics of the hydrogen energy storage system state data on the off-grid power.

5. The method for determining the off-grid power of the photovoltaic coupled electrolysis hydrogen production system according to claim 1, characterized in that: The target off-grid power determination model is constructed in the following manner: Obtaining off-grid power data and historical operation data of the photovoltaic-coupled electrolysis hydrogen production system in a historical time period; the historical operation data includes: historical photovoltaic power generation data, historical electrolyzer operation parameters, historical hydrogen energy storage system status data, historical grid interaction data, historical meteorological factor data and historical time data; Preprocessing the off-grid power data and the historical operation data to obtain a training sample set and a label set; the training sample set includes the preprocessed historical operation data, and the label set includes the off-grid power data corresponding to each sample in a plurality of samples in the training sample set; The preset model is trained using the training sample set to obtain a target off-grid power determination model; the target off-grid power determination model is used to determine the off-grid power of the photovoltaic-coupled electrolysis hydrogen production system at one or more target moments in a future time period.

6. The method for determining the off-grid power of the photovoltaic coupled electrolysis hydrogen production system according to claim 5, characterized in that: Preprocessing the off-grid power data and the historical operation data to obtain a training sample set and a label set includes: Eliminating abnormal values ​​in the off-grid power data and the historical operation data; Performing interpolation filling and normalization processing on the historical operation data; The processed off-grid power data and historical operation data are aligned according to timestamps to obtain the training sample set and label set.

7. A device for determining off-grid power for a photovoltaic coupled electrolysis hydrogen production system, characterized in that: include: An acquisition module is used to acquire the operating data of the photovoltaic coupled electrolysis hydrogen production system in the current time period, wherein the operating data includes: photovoltaic power generation data, electrolyzer operating parameters, hydrogen energy storage system status data, grid interaction data, meteorological factor data and time data; A determination module, configured to process the operation data in the current time period using a target off-grid power determination model to determine a target off-grid power of the photovoltaic-coupled electrolysis hydrogen production system at one or more moments in a future time period; Among them, the target off-grid power determination model includes a spatiotemporal convolutional neural network and a long short-term memory neural network; the spatiotemporal convolutional neural network is used to extract features of the operating data in the current time period to obtain the timing features, spatial features and coupling relationship features between multi-source data of the operating data; the long short-term memory neural network is used to perform time series processing on the timing features, spatial features and coupling relationship features between multi-source data output by the spatiotemporal convolutional neural network to determine the target off-grid power of the photovoltaic-coupled electrolysis hydrogen production system at one or more moments in the future time period.

8. The device for determining off-grid power of a photovoltaic coupled electrolysis hydrogen production system according to claim 7, characterized in that: The spatiotemporal convolutional neural network includes: a time dimension convolution layer, a space dimension convolution layer, an environment and equipment coupling convolution layer and a pooling layer; the time dimension convolution layer is used to extract the characteristics and short-term fluctuation patterns of the input data at different time scales; the space dimension convolution layer is used to process the spatial distribution data of multiple meteorological stations to capture the spatial difference characteristics of the output of the photovoltaic array; the environment and equipment coupling convolution layer is used to calculate the association weight between the meteorological factor data and the electrolyzer operating parameters through the cross attention layer.

9. A computer device, characterized in that: The method comprises a processor and a memory for storing processor-executable instructions, wherein the processor implements the steps of the method according to any one of claims 1 to 6 when executing the instructions.

10. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the instructions are executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images